Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version

AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Project Title Achieving a New European Energy Awareness (AURORA) Grant Agreement 101036418 Coordinator Universidad Politécnica de Madrid, Dr. Ana B. Cristóbal Deliverable D.2.5. Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version Due Date 31 Aug. 2025 (M44) Responsible beneficiary UPM Contributing beneficiaries INSCICO, UEvora Dissemination Level CO Confidential Version 1.4 Ver Lead Author Ver. / revision description Submission Date 0.1 Lars Lorenz (inscico) Initial Version 24/11/2023 0.2 Lars Lorenz (inscico) & Alexander Gerber (inscico) Review Feedback by AG 28/11/2023 1.0 Lars Lorenz (inscico) & Alexander Gerber (inscico); Víctor Rodríguez Doncel (UPM) & Alejandra Remacha Delgado (UPM) Input by UPM for chapter 3 30/11/2023 1.1 Víctor Rodríguez Doncel (UPM), Alejandra Remacha (UPM), Javier Garrido (UPM) Sections on recommendations 17/07/2025 1.2 Afonso Cavaco (UEVORA) Gamification 23/07/2025 1.3 Lars Lorenz (INSCICO) Add input about AURORA app and technology 29/07/2025 1.4 Víctor Rodríguez Doncel (UPM) Document review 21/08/2025 Disclaimer The information in this document is provided as is and no guarantee or warranty is given that the information is fit for any particular purpose. The user thereof uses the information at its sole risk and liability. The content of this report reflects only the authors’ view. The European Commission is not responsible for any use that may be made of the information it contains. Statement of Originality This deliverable contains original unpublished work except where clearly indicated otherwise. Acknowledgement of previously published material and of the work of others has been made through appropriate citation, quotation, or both. Ref. Ares(2025)6953058 - 28/08/2025 AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Content 1 INTRODUCTION ...................................................................................................3 2 TECHNICAL FEATURES .........................................................................................4 2.1 Introduction .......................................................................................................................... 4 2.2 Co-creation Processes and User Testing .................................................................................. 4 2.3 Reach and Engagement ......................................................................................................... 4 2.4 Technical Description ............................................................................................................ 8 2.5 Feedback and Updates to the App .......................................................................................... 9 3 SOCIAL FEATURES OF THE APP ........................................................................... 14 3.1 Introduction ........................................................................................................................ 14 3.2 State of the art .................................................................................................................... 14 3.2.1 Introduction ................................................................................................................................... 14 3.2.2 Recommender systems for behavioural change ........................................................................... 15 3.2.3 Conclusions of the theoretical analysis ......................................................................................... 29 3.3 Recommendation system .................................................................................................... 33 3.3.1 System architecture and design..................................................................................................... 33 3.3.2 System interfaces ........................................................................................................................... 36 3.3.3 Implementation details .................................................................................................................. 39 3.4 Gamification ....................................................................................................................... 44 4 REFERENCES ..................................................................................................... 46 ANNEX I. RELATED PROJECTS .................................................................................... 49 AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. About this Document This deliverable is the final version of D2.1 and D2.3 and describes the technical and social features of the AURORA App, which is being used by citizens to track their energy behaviour. Section 2, on the technical features, starts with a description of the development process of the AURORA app. The following subsection presents the requirements and specifications for the app, developed under Task 2.3. These build upon and integrate the socio‑technical considerations identified in Tasks 1.2 and 2.2. The deliverable also analyses the app's reach and the feedback received and how it has been addressed. Section 3, on the social features, includes an extensive study on the state of the art, providing the psychological justification of the recommendations sent by the AURORA Recommender system –it is not just about sending arbitrary recommendations, but sending the recommendations that drive behavioural change. The AURORA Recommender is then described from a technical point of view (requirements and implemented functionalities), with a detailed list of the messages that are being sent. The document also introduces an interactive dashboard that offers a deeper exploration of the app's data, fostering transparency and user involvement. Finally, the implemented AURORA gamification strategies are described. 1 INTRODUCTION The AURORA Energy Tracker is a mobile and web-based application to inform citizens about the environmental impact of their energy choices and to encourage more sustainable behaviour (read more in D2.2). By providing clear and accessible insights into individual energy consumption and its associated ecological footprint, the app serves as a practical tool to raise awareness and empower users to make informed decisions in their daily lives – but beyond this intuition this deliverable justifies the psychological reasons behind it. Section 2 will present the technical features, Section 3 the social ones. This deliverable includes the requirements and specifications for the app produced in T2.3, integrating both the socio-technical aspects defined in T1.2 and T2.2, as well as the data models and algorithms for the provenance-based integral calculation of the citizen energy footprint. This document presents a comprehensive overview of both the technical components that underpin the Energy Tracker and the social features designed to foster engagement and behavioural change. Section 2 describes the technical features of the app, Section 3 focuses on the social dimension, covering strategies explored such as the gamification initiatives and the recommendation system. Together, these elements illustrate AURORA’s integrated approach, combining robust technological solutions with behavioural and social strategies to maximize the app’s impact and long‑term relevance within the participating communities. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. 2 TECHNICAL FEATURES 2.1 Introduction This chapter presents the technical features of the AURORA Energy Tracker, detailing how the app was designed, deployed, and iteratively improved to support citizen engagement with energy‑related environmental impacts. Section 2.2 describes the app design and testing methodology, Section 2.3 the reach and engagement. Section 2.4 describes the technical features of both the front end and back end and finally, Section 2.5 highlights major updates since the original launch and summarises the feedback received from users and stakeholders. 2.2 Co-creation Processes and User Testing The AURORA Energy Tracker has been developed in close collaboration with its future users through an extensive co‑creation process. This approach was pivotal to ensuring the success of the project as a whole, given its strong dependence on the app’s effectiveness and sustained user engagement. To this end, AURORA implemented a co‑creative design methodology for Environmental Impact Monitoring (WP2) and, in particular, for the development of the mobile and web applications. The methodology and its outcomes are detailed in Deliverables D2.1 to D2.4. Although the initial co‑creation phase was completed, the AURORA project has continued to integrate user feedback to further refine the tool and align it with citizen needs. This ongoing exchange was facilitated by the demo‑site partners through dedicated events promoting the app, as well as via the integrated in‑app feedback form. Chapter 2 of this deliverable (“Feedback and Updates to the App”) summarises the most significant feedback received and documents how it has been addressed in the latest release. The strong foundation built through early community involvement – including five Innovation Café events and structured survey interviews – has resulted in a process that enables continuous improvement of the app in response to evolving community needs and expectations. This iterative process was applied extensively during the beta testing of the AURORA app. Standardised online protocols were circulated to testers, guiding them through specific tasks and collecting structured feedback. While these tests primarily focused on validating core functionality, they were critical in ensuring a fully operational release version, which was successfully delivered in September 2023. A comprehensive report of all usability testing iterations is available in Deliverable 2.4. 2.3 Reach and Engagement The AURORA Energy Tracker has been available for download since end of September 2023, with the web-based version following shortly after in February 2024 (dashboard) and June 2024 (web app). All versions had updates of varying scope along the way – highlights of those are detailed in section 2.5. To evaluate its reach and engagement, the most relevant statistics for a mobile application are downloads, views of the store listings, and the actual use of the app. The following sections take a deeper look at each of those metrics. Downloads and Engagement The AURORA Energy Tracker is available on both Android and iOS devices and distributed through their respective official stores, Google Play and Apple App Store. Each store therefore has separate download and engagement stats. The following stats were last updated on July 2025. Android - Google Play Store (July 2025) In total, the Android version of the app has been downloaded 1,642 times since its launch. 572 of those acquisitions happened through “Google Play Explore”, which means that users came across the app on the store naturally, 374 from searching for the app specifically, 436 through direct links to the store listing, such as those on the project website, and the rest through other AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. pathways. The store listing was viewed by 4,312 people, resulting in a conversion rate of about 38%. Figure 1. Downloads over time (Google Play) iOS - Apple App Store (July 2025) The iOS version of the app has been downloaded 1,011 times since launch. Apple breaks the number of downloads down slightly differently: 720 of those downloads were users discovering the app through the store, and 262 through referral links on the web. The store listing was viewed 2,130 times, resulting in a conversion rate of 47%. Figure 2. Downloads over time (iOS App Store) Combined Stats Combining both Google Play and App Store stats, the AURORA Energy tracker has been downloaded 2,653 times in total, with 6,442 store listing views since its launch at the end of September 2023. As both stores are completely separate, it is technically impossible to verify whether a single user has downloaded the app both from their iOS and Android devices. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Nevertheless, these estimates should be reasonably accurate, which is supported by the number of user accounts created (see below), as user accounts are compatible across devices. Figure 3. Downloads over time (Combined) User Accounts and Energy Tracking Besides the number of downloads AURORA is also closely tracking the number of accounts created and how they are using the app. The latest numbers are always available on the AURORA interactive dashboard and are automatically updated every day. Note that using AURORA through the web-based application does not require a download like for iOS or Android and hence the number of user accounts does not strictly match the number of downloads on those platforms. Overall, 1,646 accounts have been created. Of those users, 391 identify as female, 760 as male, 17 as non-binary, and 190 as other. In total, those users have created 155,900 consumptions. 154,445 are related to transportation, 998 to electricity, and 457 to heating. This amounts to a total of 470,481kg CO₂ emissions or 2,169,584 kWh of energy used. 383,787kg CO₂ are from transportation (1,604,255 kWh), 33,361kg CO₂ from electricity (232,769 kWh), and 53,333kg CO₂ from heating (332,561 kWh). Denmark For Denmark 70 accounts have been created. Of those users, 11 identify as female, 39 as male, 1 as non-binary, and 3 as other. In total, those users have created 2,070 consumptions. 1,964 are related to transportation, 66 to electricity, and 40 to heating. This amounts to a total of 40,036kg CO₂ emissions or 173,114 kWh of energy used. 24,387kg CO₂ are from transportation (95,149 kWh), 2,292kg CO₂ from electricity (41,441 kWh), and 13,358kg CO₂ from heating (36,523 kWh). Portugal For Portugal 470 accounts have been created. Of those users, 105 identify as female, 225 as male, 1 as non-binary, and 98 as other. In total, those users have created 17,029 consumptions. 16,748 are related to transportation, 239 to electricity, and 42 to heating. This amounts to a total of 110,618kg CO₂ emissions or 503,928 kWh of energy used. 100,014kg CO₂ are from transportation (414,291 kWh), 8,914kg CO₂ from electricity (53,425 kWh), and 1,689kg CO₂ from heating (36,213 kWh). AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Slovenia For Slovenia 217 accounts have been created. Of those users, 23 identify as female, 138 as male, 4 as non-binary, and 19 as other. In total, those users have created 27,997 consumptions. 27,613 are related to transportation, 213 to electricity, and 171 to heating. This amounts to a total of 108,178kg CO₂ emissions or 501,375 kWh of energy used. 92,661kg CO₂ are from transportation (377,121 kWh), 3,868kg CO₂ from electricity (40,211 kWh), and 11,650kg CO₂ from heating (84,043 kWh). Spain For Spain 446 accounts have been created. Of those users, 134 identify as female, 197 as male, 9 as non-binary, and 36 as other. In total, those users have created 101,798 consumptions. 101,353 are related to transportation, 332 to electricity, and 113 to heating. This amounts to a total of 140,611kg CO₂ emissions or 624,308 kWh of energy used. 127,861kg CO₂ are from transportation (554,855 kWh), 5,699kg CO₂ from electricity (34,709 kWh), and 7,051kg CO₂ from heating (34,744 kWh). United Kingdom For the United Kingdom 112 accounts have been created. Of those users, 32 identify as female, 48 as male, 1 as non-binary, and 5 as other. In total, those users have created 2,118 consumptions. 1,960 are related to transportation, 97 to electricity, and 61 to heating. This amounts to a total of 42,625kg CO₂ emissions or 242,661 kWh of energy used. 24,840kg CO₂ are from transportation (103,459 kWh), 4,199kg CO₂ from electricity (35,467 kWh), and 13,586kg CO₂ from heating (103,734 kWh). Data for all other European countries that are not AURORA demo sites is available on the dashboard and omitted here for brevity. Figure 4: Average emission data per country since January 2024 Figure 5: Label distribution for carbon emissions in 2025 across demo site countries AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Figure 6: Gender distribution across demo site countries Note that all charts were directly exported from the interactive dashboard, which can always be checked for the latest version of these, in addition to more data and various display options. For persistence, data is also preserved in Zenodo (https://doi.org/10.5281/zenodo.16920799). 2.4 Technical Description Technically speaking, the AURORA Energy Tracker comprises two parts, the frontend and backend. The frontend is defined as the app users can download to their mobile device or use in their browser with regards to the web app and is available for Android and iOS as a native application, as well as any web browser. The backend is decoupled from the front end, which allows it to run independently and serve all versions of the app with the same data. Frontend 1. User Interface and Experience: ○ Co-designed with prospective users in five countries through workshops and surveys (see D2.1 and 2.2). ○ Developed natively for both iOS and Android, adhering to the latest user interface standards of Apple and Google. ○ Later addition of web-based application, directly integrated to the interactive dashboard. 2. Data Entry and Calculation: ○ Users can input data for heating, electricity, transportation, and their local PV investments. ○ The app automatically calculates carbon emissions and energy usage (kWh) for each entry. ○ Users can add / edit / duplicate / delete multiple entries and view a monthly breakdown aggregated by year. 3. Energy Labels and Feedback: ○ A system of energy labels (e.g., "Ground Breaker (G)" to "Accomplisher (A+)") dynamically calculated and updated in real-time. ○ Immediate feedback to users as they add more data, instantly propagated across all their logged in devices. 4. User Support and Engagement Tools: ○ Scheduled notifications for data entry reminders. ○ Configurable recurring consumptions based on regular activities. ○ Integration with the PVGIS tool for estimating the impact of investing in photovoltaic installations. ○ Actual tracking of PV investments with dynamic offsets based on investment size (shares) and actual PV production data. ○ Personalised recommendations based on energy behaviour and user specific information. 5. Multilingual Support and Accessibility: ○ Available in six languages (English, Danish, German, Portuguese, Slovenian, and Spanish). AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. ○ User accounts are fully transferable between all platforms (iOS, Android, web) by simply logging in on a different device. 6. Privacy and Security: ○ Users have control over their data, with the ability to download or delete it at any time. ○ Support form for inquiries and assistance. 7. Open Source: ○ Source code for both iOS 1 and Android 2 versions is accessible on GitHub. Backend: The backend server is responsible for managing calculations, data aggregation, label assignment, and user authentication. 1. Server Architecture: ○ Common server architecture for all applications (iOS, Android, web). ○ All calculations and data management are performed directly on the server. ○ Uses cloud functions to reactively trigger computations based on user interaction or schedule (depending on the function and purpose). 2. Scalability and Performance: ○ Backend deployed on Google Firebase, providing high scalability and performance. ○ Allows adjustments and updates without downtime or app updates. 3. Calculation and Labelling: ○ Manages all calculations of carbon emissions and energy usage. ○ Dynamically calculates and assigns energy labels. ○ Regularly pulls latest PV production data from actual PV installations at the demo sites via the QPV API. ○ Regularly syncs all changes to consumptions with the Recommender System (UPM) and pulls in the latest recommendations. 4. User Authentication and Data Control: ○ Enables secure user authentication. ○ Handles requests by users to delete or download their personal data. 5. Open Source: ○ The source code for the backend running on Google Firebase is publicly available 3 . 2.5 Feedback and Updates to the App Major app updates Since its initial release in September 2023, the AURORA app has undergone a major evolution. While there have been many larger and smaller updates, this section focuses on the most notable features that were added. Expansion of consumptions Additional options have been added to consumption tracking and existing ones were expanded as part of a major overhaul. This was a collaborative effort with WP1, which provided an updated 1 https://github.com/AURORA-H2020/AURORA-iOS 2 https://github.com/AURORA-H2020/AURORA-Android 3 https://github.com/AURORA-H2020/AURORA-Firebase AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. counteracts the main function of intelligence, the good management of one’s own behaviour, as well as reducing people’s satisfaction with their decisions, even if they have made good choices. This inability to choose occurs when consumers have to select from options that are difficult to compare. There is no doubt that having more options often enables people to achieve better objectives outcomes and enhances autonomy [7]. However, the availability of choices also might decrease users’ well-being, instead of producing a benefit, because the greater number of options, the more effort or help to choose is needed. Research shows that, faced with a wide range of choices, consumers are less likely to buy something, and in the case, they buy it, they are less satisfied with their choice [8]. Other researchers point out that, in consumption contexts, there exists a tendency to make consumption decisions that restore one’s sense of self-worth, which is called compensatory consumption [9][10], but does not imply that it is the right choice. For any of the above situations, choice architecture provides structures that will affect outcomes, and most importantly, avoids a low participation rate (Thaler and Sunstein, 2008) [3]. In addition to the progress in the technical aspects underpinning RSs, researchers have addressed their study in various areas related to the field of behavioural sciences, such as behaviourism, economic psychology and behavioural economics. Behaviourism, which may be defined as the theoretical stance in psychology that emphasises the role of learning and observing behaviours in understanding human actions, occupies an important position in the study of RSs from a psychological point of view to such an extent that it has become the currently dominant paradigm for constructing and evaluating RSs [11]. Economic psychology and behavioural economics are two examples of recent intersections of fields, aimed at developing interdisciplinary approaches to solving complex issues and phenomena. Thus, in order to reflect on the design of a suitable RS for the AURORA project, several works written by relevant authors such as Cass Sunstein, Richard Thaler, Amos Tversky and Daniel Kahneman have been thoroughly reviewed. One of the most important contributions within behavioural economics is the nudge theory, proposed by Thaler and Sunstein (2008) [3]. Nudge is defined as “any aspect of the choice architecture that alters people’s behaviour in a predictable way without forbidding any options or significantly changing their economic incentives”. A second definition shifts the balance towards policy making. Nudge is then: “a policy intervention that is intended to influence behaviour, but does not involve any incentive or sanction, mandate or regulation, and is more than just giving people information”. Nudges should be both choice-enhancing --or at least not choice-restricting- - and transparent (Halpner, 2015) [12]. And yes, nudges can be transparent and also effective, at least according to a recent report by the European Commission [13]. Nudging in AURORA seems to be ok, then. This is also confirmed by the thorough ethical study made recently by one of the authors if this deliverable [47]. This sort of orientation towards socially desirable behaviour must comply with certain principles. Sunstein and Thaler stress that genuine nudge rejects the idea of mere mandate or instruction, this is the reason why a nudge should also be easily avoidable; moreover, the proposed decision environment or choice architecture should be characterised by simplicity and clarity, which implies relatively low cost. To be clearer: nudges respect autonomy understood as freedom of choice because nudges do not modify people’s choice set and do not offer significant (financial) incentives. This implies that people should be nudged to choose the things that optimize their welfare according to their own individual conception of welfare. The authors advocate the design of policies that help the least educated sectors of society while imposing the lowest possible costs on the most educated, that is, a context of so-called libertarian paternalism in which designing a decision-making architecture to nudge citizens is the justified option for legislators. Numerous examples of initiatives and decisions made by corporations referred as nudges are provided throughout the book, among them, one of the most clarifying and simplest examples of nudging might be the following: “Putting fruit at eye level counts as a nudge. Banning junk food does not”. Nowadays the theory of nudge is a key tool in everyday life, showing its wide impact in both academia and business. However, criticisms have been made of the theory of nudge regarding its concept of autonomy and welfare, and have attempted to outline their long-term AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. consequences, especially on the subject of how democracy and deliberative processes would be affected. A recent systematic literature review focused on ethics issues with nudging (Kuyer and Gordijn, 2023) [14] identifies 1693 records in English language after filtering out duplicates. As a summary overview of the various arguments in favour of and against, nudging has been described [15] as an elegant idea whose efficacy, however, would rely on agreement on what the ’right’ answer is, therefore, the lack of consensus on a more controversial issue entails a nudge has the potential to be inadvertently harmful. Following on from the previous example of nudging, in the role of nothing more than an observer, it is simple to identify the problem emerged of this apparently non-harmful nudge. Fresh fruit and vegetables cut and served in take-away containers have become a common product in Western supermarkets due to their size, design and price, which made them attractive and ensure a healthy choice in the context of a dynamic and busy lifestyle. The nudge implicit would not be complete without locating these items in a placement to exert a positive influence on the customer and facilitate purchase. Although it is a successful measure, as many customers may have changed their behaviour and chosen the healthy take-away products instead of those in the junk food range, from the point of view of waste and green attitudes, it is also a failure. Since 2008, year in which Thaler and Sunstein proposed the Nudge Theory, the concept has spread among academics, corporations and governments, and is considered a successful soft intervention on behaviour. During this time, companies and governance agencies have emerged to design, implement and evaluate nudges. The latter usually receive the name of Behavioural Insights Teams (BIT), simplified as Nudge Units. Originally, BIT was conceived as a part of governments and are based in different countries. As time passes, some of them, for example the United Kingdom’s Behavioural Insight Team, created by Prime Minister David Cameron in 2009, have become partly private entities. Psychologist David Halpner, author of the book Inside the Nudge Unit How Small Changes Can Make a Big Difference (2015) [12], devotes the last chapter before conclusions to the risks and limitations of nudging. Here Halpner goes back to 2003 at the time when British media received a think piece entitled Personal Responsibility and Changing Behaviour: the state of knowledge and its implications for public policy [16], with the sentence “This is not a statement of government policy.” printed on the bottom of each page. The impact prompted Prime Minister, Tony Blair at the time, to make a major speech distancing himself from the use of behavioural approaches by government. This is an example of how delicate it is to address the issue of nudges. In the USA, Nudge Units have been seen as the channel for providing powerful forms of psychological strategies to governments and private interests, what has led its main proponents to focus on choice-enhancing nudges. In contrast, other critics have accused governments of opting for nudging rather than implementing appropriate policies. While Halpner clearly has seen that behavioural insights can benefit, and even enhance the character of democracy itself, he has analysed what he considers to be the three major deficiencies in nudge techniques: • Lack of transparency – that behavioural approaches are too close to the dark arts of propaganda and subconscious manipulation (a concern of the right). • Lack of efficacy – that behavioural approaches are an excuse for not acting more decisively and effectively. • Lack of accountability – that the behavioural scientists and decisionmakers behind these approaches need to be more answerable to those they affect. The usefulness of nudging has been also studied in the context of an EU project, like NUDGE (NUDging consumers towards enerGy Efficiency through behavioural science) 5 or ForestAgriGreenNudge (GREEN NUDGEs for sustainable FORESTry and AGRIcultural practices post 2027) –other projects are analysed in Annex 1. NUDGE’s approach starts with an exhaustive survey widely disseminated as it is been completed by citizens from 29 European countries. The final sample includes 3129 citizens between 18 and 100 years and Europeans. The researchers recognised a slight over or underrepresentation in relation to age, gender, and educational attainment for some groups. Annex 1 of this paper presents an overview of the NUDGE project, however, for the purpose of this state of the art, attention is focused on the survey itself, whose theoretical body is built on three models of human behaviour: the Theory of Planned Behaviour 5 https://cordis.europa.eu/project/id/957012 AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. (Ajzen, 1991) [17], the Value-Belief-Norm theory (Stern et al., 1999) [18] and the Prototype Willingness model (Gerrard et al., 2008) [19]. The novelty of this approach is precisely the cross-analysis of such models, in addition to the use of the five variables of gender, country of residence, income, age and level of education, and the development of a single instrument based on such models and variables. As a result of the survey, the researchers distinguish six profiles of energy consumers, which they describe and for which they recommend a different type of intervention to help improve their energy consumption. The psychology of preferences The title of this section reproduces the title of the brief article written by Daniel Kahneman and Amos Tversky in 1981 [20], where they studied risky decisions from a mathematical perspective. The section threads together the following contributions and theories on the analysis of decision making: • The importance of others, social contagion, collective consciousness and shared value. • The Prospect Theory (Kahneman and Tversky, 1979) [21] and the emotion of fear of loss. • The Regret Theory (Loomes and Sudgen, 1981 [22]; and the emotions of regret aversion and anticipated regret. • Integrity and regulation. The philosopher and economist Adam Smith and his classical work The Wealth of Nations (1776) usually occupy the top position in rankings of the most influential thinking of all time in economics. However, it is in a previous text, The Theory of Moral Sentiments (1759), where the role of psychology is defined in economics. In this work, Smith explains that morality is not driven by an innate moral sense, but by humanity’s natural sociability, that is, the need for the approval from one’s peers. In part III, chapter II, “Of the love of Praise, and of that of Praise-worthiness; and of the dread of Blame, and of that of Blame-worthiness”, Smith claims: Man naturally desires, not only to be loved, but to be lovely; or to be that thing which is the natural and proper object of love. He naturally dreads, not only to be hated, but to be hateful; or to be that thing which is the natural and proper object of hatred. He desires, not only praise, but praise-worthiness; or to be that thing which, though it should be praised by nobody, is, however, the natural and proper object of praise. He dreads, not only blame, but blameworthiness; or to be that thing which, though it should be blamed by nobody, is, however, the natural and proper object of blame. (TMS, III.ii.1) According to Smith, the most important factor in the decisions of human beings is the quality of deserving admiration. As long as humans are conscious social animals, the importance of this factor will not be seriously altered. On the other hand, people are much more likely to adopt or accept a change if they feel that people like them have recommended it to them. At present, the influence of neighbours on energy-related behaviour is a well-documented phenomenon in the fields of social psychology and behavioural economics –that is why AURORA conceived the idea of comparing data from different users. This concept often falls under the umbrella of social norms and peer influence, where individuals are influenced by the behaviour and practices of those around them. For example, Wolske et al. explored how peer influence shapes household energy behaviours, focusing on the mechanisms behind social influence and its effects on energy consumption [23] while Bonan et al. conducted a comprehensive study examining how social norms can be leveraged to promote energy conservation effectively [24]. In AURORA some of the messages will provide a comparison of energy labels or raw consumption within the same cluster. In the 1970s and 1980s, crucial steps were taken with regard to the concept of loss aversion. Kahneman considers that the concept of loss aversion is undoubtedly the most important contribution of the psychology of behavioural economics. Changing user behaviour requires training and awareness-raising exercises, incentive recommendations and feedback evaluations to induce permanent change [25] (Himeur et al., 2021). From their point of view, recommendations have to strike a balance between user comfort and energy efficiency, while the final decision should be in the hands of the end-users. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. The psychological analysis of RSs and their effect on human behaviour is really complex. Even in very repeatable scenarios, such as a website making A/B testing on a very homogeneous category of users, variability of observations is large, and results are very sensitive to demographics, time of day, previous user experience, referral source or the user intent. A good system for certain person, may not work in another context, even if it is the same person. In this line, Cremonesi et al. (2012) analysed RSs and found empirical evidence that their objective quality was not always a good indicator of the potential for persuasion of an RS: “how the users see an RS, and ultimately, the RS’s capability to affect the attitude or behavioural sphere, are induced by a complex plethora of factors, among which algorithmic performance is less crucial than we may expect” [26]. Exogenous factors such personal agenda, age range or nationality, among others, are decisive. In the area of improving energy efficiency behaviours, the EU PENNY project conducted scientific experiments (including A/B testing) with the intention of enhancing the design of policies to maximise energy efficiency behaviours –to read more on PENNY, see Annex 1 to or the book summarizing their findings [27]. Cremonesi et al. (2012) highlighted that in the opinion of users on an RS and how this tool can affect users’ attitude, factors distinct from algorithmic performance are more relevant. Based on this assumption, the RS design should address towards the search of what is important in the community where the energy community is located. There is a common agreement of economic incentives have the highest individual average effect, for this reason they appear in the two most effective intervention combinations described by Composto and Weber (2022). However, these authors underline some of the combinations of behavioural interventions, which are more effective than a single behavioural tool: defaults with reframing, commitments with observable behaviour, and feedback with energy saving information. For example, regarding the last combination, the energy saving information should be provided in the course of dissemination and engaging activities organised by the AURORA demo sites, of which many have taken place already, combined with messages through the app aimed at collecting opinions and feedback. On the other hand, should not neglect other communicating opportunities, in this regard, external content (news, scientific papers, local events or just the availability of a new solar charging station) may be pointed to. In the same way, the app may propose users –who will be free to refuse them– certain commitments based on information provided in the demo sites and their own interests, as users would act more or less rationally in forming and modifying attitudes, according to beliefs and values, rather than performing behaviours as a result of conditioning (Oinas-Kukkonen and Harjumaa, 2009). On this matter, it would be an approach of major significance, to define what participants in the energy communities understand by a sustainable lifestyle, followed by the inclusion of a list of questions or tips regarding its achievement. Figure 1 summarises Oinas’ primary support tasks, their requirements and example of applications given by the authors as well as several possible implementations in AURORA project. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Figure 1: Primary support Task, as in Oinas-Kukkonen, adapted for AURORA Figure 2 summarises Oinas’ dialogue support tasks. Figure 3 represents Oinas’ credibility support tasks. Finally, Figure 4 capture the support support tasks. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Figure 2: Dialogue support Task, as in Oinas-Kukkonen, adapted for AURORA AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Figure 3: Dialogue support Task, as in Oinas-Kukkonen, adapted for AURORA. Himeur et al. (2021) remind us of a golden rule: recommendations have to strike a balance between user comfort and energy efficiency, which can be combined with a personalised approach to send reminders based on the individual’s schedule (Phiri and Trevorrow, 2019). For this purpose, a user segmentation is needed, at least per location and age group. The former is already available, whereas the latter should be included. In the bibliography consulted, the Fogg’s Behaviour Model (FBM) designed to change behaviour stands out [28]. His contribution will be enriched with other works such as the Scurati et al. (2020) [29] framework for designing products that encourage sustainable behaviour. The FBM posits that behaviour results from three key factors: motivation, ability, and triggers (Figure 5). The model states that for an objective behaviour to occur, the person must simultaneously have sufficient motivation, sufficient ability and an effective trigger. Fogg’s work has significantly influenced the field of persuasive design, enabling practitioners to create effective interventions that encourage behaviour change. Such a model may suit AURORA problem, identified as follows: although AURORA partners claim high engagement and response from participants in all demos, the tools used by the project are not able to measure these phenomena. In other words, all the three Fogg’s key factors are already there, but they need to be rethought. In conclusion, this section has presented different theoretical frameworks to analyse the behavioural change. How important are they quantitatively? –Both the project NUDGE and the work of Tabi [30] posed this question before. Individuals were clustered in both cases, and it was found that different clusters responded differently to different stimuli. AURORA tries to pursue then multiple approaches, as described in Section 3.3.3. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Figure 4: Social support Task, as in Oinas-Kukkonen, adapted to AURORA Figure 5: Fogg’s Behaviour Model Methodologies to design a RS Designing products that encourage citizens behaviour change is becoming one of the most popular trends in design research today [29]. Indeed, the proposed actions do not only target citizens (as we do in AURORA) but also policy makers and other change agents [31] within integrated strategies. RSs are trained to match users’ preferences, prior decisions and characteristics from the data collected on their interactions with the qualities of the items for sale. From a technical perspective, RSs have gone through their own evolution from a first generation based on knowledge, content and collaborative filtering, or a hybrid technique, to a third generation related to the emergence of deep learning, passing by a matrix factorization-based approach, among others. Therefore, the usefulness of RSs lie in their ability to help users discover products and services that they might not find on their own. Cremonesi et al. (2012) explore the persuasiveness of RSs, presenting two vast empirical studies that address a number of research questions [26]. In view of the large number of variables that need to be controlled and the lack of sufficient factual work, the team of researchers defends its work as one of the most far-reaching studies in this context. Among other questions, their research seeks to elucidate whether it is possible to isolate those properties of an RS design that seem more effective in achieving persuasive goals. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Another important finding when designing tools to modify user behaviour is gamification, which is an interesting field of work that has been widely researched over the last three decades. However, for to the purpose of this state of the art, it will not be discussed here. Having defined the theoretical corpus to be reviewed for this paper, it is time to identify the key design factors. Oinas-Kukkonen and Harjumaa listed in 2009 twenty-eight design principles for persuasive system content and functionality, describing example software requirements and implementations [32]. In a previous work (2008), the authors had defined persuasive systems as “computerised software or information systems designed to reinforce, change or shape attitudes or behaviours or both without using coercion or deception” [33]. This definition implies that the potency of persuasive systems lies in the combination of the interpersonal and mass communication attributes of Internet-based systems, and, from a strictly psychological perspective, involves three key features, the possibility of reinforce a current attitude or behaviour in order to make them more resistant to variation, the capacity of change in an individual’s response due to (e. g., certain events or socioeconomic pressure), and the ability to shape behavioural outcomes both the existing and unknown situations. Fogg published a work that is worth reading in parallel to the above mentioned, which includes an eight-steps design process based on 15 years of experience in researching at Stanford University and designing persuasive technologies for industry and testing solutions [34]. In this paper, Fogg explains what he considers “best practices” for developing digital experiences that influence users, listed in eight steps as follows: AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. According to Fogg, the first four steps can be rearranged depending on design conditions and eventualities. The second step involves choosing what designers consider to be the right audience, although they frequently have no choice. In that case, Fogg points out that there are still multiple ways to benefit from flexibility in the design process. In step 6, the author recognises that imitating what already works and adapting successful approaches to behaviour and target audiences is a more reliable method than starting from scratch. The relevance of context cannot be understated, and this can also be seen in the conclusions of Adaji and Adisa’s research, which indicates that personalised persuasive technologies are most effective in inducing desired behaviour change. Researchers have not treated RSs evaluation in much detail. It is common to other papers reviewed, the fact that the long-term efficacy of persuasive strategies for sustainability is unknown, as researchers typically evaluate their tools and techniques for a very short period of time, rarely more than two months [35]. This also means that researchers have no strategy to keep the participants interested in the technologies over time, which is a recurrent problem both in the literature and in the European projects reviewed. Contrary to regular approach, Pu et al. (2011) [36] provide both a long and a short version of an evaluation framework called ResQue (Recommender systems’ Quality of user experience), designed for measuring the qualities of the recommended items, the system’s usability, usefulness, interface and interaction qualities, users’ satisfaction with the systems, and the Step 2: Choose a receptive audience Step 3: Find what prevents the target behaviour Step 4: Choose a familiar technology channel Step 5: Find relevant examples of persuasive technology Step 6: Imitate successful examples Step 7: Test and iterate quickly Step 1: Choose a simple behaviour to target Step 8: Expand on success AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. A more elaborated listed of messages if listed in the Annexes. Figure 8: Notification types and some notifications Figure 9: Initially identified messages types and some messages AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. 3.3 Recommendation system This system implements the main ideas and design principles identified in the previous sections. 3.3.1 System architecture and design The system architecture consists of five key components, each with a specific role in generating and delivering personalized recommendations aimed at modifying user behaviour concerning green aspects. They are described in Figure 10. Figure 10: Architecture diagram AURORA Backend The AURORA backend (already described in Section 2) is part of the schema as provider of user data and a future possible user interface. Recommendation Engine The recommendation engine, operating in Madrid at UPM, uses data from both the backend (userrelated information) and other external sources (non-user information) to generate recommendations. It is based on a Rule-Based system that evaluates conditions and makes decisions on the most relevant suggestions for each user. The system ensures that recommendations are personalized and contextually appropriate. The system is able to parse rules in both Drools format (executable by a Drools engine) plus rules in a simple format. Filter Module Once the recommendation engine produces the recommendations, the filter module selects the best message and determines the optimal time for delivering notifications. This process ensures that notifications are both effective and timely, increasing their likelihood of influencing user behaviour. The module is also in charge of avoiding message overload, limiting the delivered messages. Translation Module The translation module handles the adaptation of recommendations into different languages and forms. It ensures that text is varied to avoid repetition and makes the notifications culturally appropriate for the target audience. The system modifies the content’s language and tone as necessary, based on user preferences or regional needs. OpenAI has been used for this, but only tested in Spanish and English. User Interface The visualization (AURORA App, a Dashboard, or a separated system) is the final component in the system. It is responsible for delivering the personalized notifications to users. Once the filter and translation modules have processed the recommendation, the user interface sends the notification to the user’s device at the optimal time or to the dashboard, with the appropriate language and message variation. Expected changes as a result of implementing the Recommender: (1) The ability to receive a message from the Recommender System and issue a AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. push notification; (2) Adding an email address in the information for contact; (3) Ability to opt-in or opt-out from notifications. The backend and the user context provide information about every user. The set of pre-loaded rules fire one or more messages, that are listed in a recommendation stack. The filter module selects the best message for each user. An LLM-based module translates the message into different languages, possibly paraphrasing messages to make them more variate. The Recommendation Engine, the Filter Module and the Translation Module are running at UPM servers. Feedback is obtained in the app from the user, further characterising the user and helping improve the filter module. Messages can also drive the user to external web pages or Twitter messages. Technical requirements Requirements for the Recommendations Module Technical requirements follow. • Language: The system must operate in English. • User management: It must be possible that test users are created via an API call. It must be possible to obtain real users from the app via API calls. When users delete the App or revoke the consent to process their personal data, they should be removed. • Performance and Scalability The recommendation engine must handle up to 70,000 users (10 times the official estimate) and about 1,000 rules (10 times our estimate). • Data input The system should support data import/export via APIs, ensuring compatibility with other systems for data synchronization. The engine must integrate seamlessly with the backend (probably a security token or a key are necessary) to receive user context data and any other relevant external data sources (which have a variety of formats). • Data output The engine must be capable of producing messages are represented in JSON. { "id ": "00" , "userid ": "001" , "threadid ": "0", "title ": " title of the notification ", " body ": " this is the body recommendation, in escaped html", "datetime ": " generated date and time ", " priority ": "0 -5" } • On the rules The engine must be based on a rule-based system for decision-making, where business rules define the conditions for recommending specific actions or content. Rules are represented in the Drools syntax or an AURORA Syntax. • Logging Error logs and debugging information must be stored for quick issue resolution. Events to be logged: (1) creation modification or erasure a of user; (2) creation of a message as an ouptut; (3) rule base execution; (4) warnings and errors. • Security and privacy Message are personal data as per GDPR. Messages must be persisted encrypted. Access to sensitive user data should be restricted, and authorization mechanisms must be implemented. Non-Technical requirements follow. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. • User Feedback The system should support mechanisms for gathering and integrating user feedback to continuously improve the recommendations over time (perhaps an providing an email address in the app suffices). • Transparency (for developers) The recommendation engine should provide explanations to the developers about why specific recommendations are being made, logs should consider this. At some point, real-time monitoring tools should be in place to detect anomalies and optimize system performance. Requirements for the Filter and Translation Module The Filter Module takes the list of recommendations and ensures that only relevant messages are sent. It includes logic to enforce the one-message-per-week rule and respects user preferences (including opting out of notifications). The Translation Module adapts the message text into the user’s preferred language, ensuring no repetitive content is sent. Technical requirements follow. Message Filtering The Filter Module must receive a list of recommendations (messages) from the Rule-Based Engine. The system must ensure that only relevant messages are selected for delivery. This involves evaluating the context, user preferences, and recency of previous messages. The module must enforce the rule that no more than one message is sent to a user per week. If the users has specified in their preferences that opts-out of receiving messages, no message should be generated. Message Translation and Variation The Translation Module must take the filtered messages and adapt them for the user’s preferred language and context (EN,ES,DE,PT,DA,SL). The system should introduce variations in the messages to avoid repetition, ensuring that the content remains fresh and engaging. Language The system must accept English but produce Portuguese, German, Spanish, Slovenian and Danish (if reasonably possible). Logging Error logs and debugging information must be stored for quick issue resolution. Events to be logged: (1) Translation made (with input and output); (2) Message delivered to the AURORA app; (3) Warnings and errors. Security and privacy Message are personal data as per GDPR. Messages must be persisted encrypted. Access to sensitive user data should be restricted, and authorization mechanisms must be implemented. Authentication All endpoints require an API key or token for authentication. The same style as in the AURORA App should be used –in principle, a Bearer header in the request as follows: Authorization: Bearer <token> AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Notifications and messages With regard to choosing a receptive audience, the target audience should ideally be determined by the role of individuals and their connection to the different demo sites. In the university-based demo sites, the entire university community (students, professors, and staff) is likely to engage the AURORA project, whereas in the Forest of Dean demo site, settled in a council, the neighbours and institutions may follow AURORA project. The scheme behind AURORA RS lies on four general steps: 1. Elaborate inputs related to ecological awareness, specific environmental control beliefs and environmental justice appraisals 2. Choose a simple behaviour to target 3. Find what prevents the target behaviour 4. Test and iterate The aim of the first step is to make sure that user’s environmental awareness does indeed exist. This is a basic prerequisite before choosing a target behaviour. Users may be asked the following questions: • Do you think it is important to learn to adopt behaviours that benefit the environment and your wallet? • Do you feel part of the solution? • How often do you read newsletters, magazines or other publications written by an environmental group? • How often do you vote for a candidate in an election at least in part because he or she is in favour of strong environmental protection/conservation? • Tourism has an environmental impact - Do you want to know more about how to mitigate the negative effects of tourism? 3.3.2 System interfaces API The AURORA Recommender API provides programmatic access to the AURORA Recommender component (plus Filter and Translation). The API follows an HTTP REST architecture and uses Bearer authentication for secure access. It is hosted on the production server at: The AURORA Recommender runs as a Java service. The system exposes the API using Jetty. This is the endpoint: https://aurora.linkeddata.es/api/ Authorization is handled via Bearer authentication (HTTP Bearer). The API is fully documented using Swagger UI, which allows any authorized user with valid credentials to explore and test the available endpoints directly. Screen captures of the implemented methods follow. The system is compatible with the data structures used by the AURORA App backend. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Authentication and Session Management • POST /api/login – Authenticates a user and establishes a session. • GET /api/logout – Terminates the session for the authenticated user. User Operations These endpoints manage individual users and their data: • GET /api/user/{id} – Retrieves information about a specific user by ID. • POST /api/user/{id} – Creates or updates user information. AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. • DELETE /api/user/{id} – Deletes a user. • GET /api/user/{id}/auroraconsumptions – Retrieves all energy consumption data for a user in AURORA format. • POST /api/user/{id}/auroraconsumptions – Uploads new consumption data for a user. • DELETE /api/user/{id}/auroraconsumptions – Removes all stored consumption data for a user. • GET /api/user/{id}/consumptions – Retrieves all consumption data for a user (alternative representation). • GET /api/user/{id}/recs – Retrieves personalized recommendations generated for a user. • GET /api/user/{id}/run – Triggers the rule engine manually for a given user to generate all possible recommendations (primarily for testing). Recommendation Operations These endpoints handle individual recommendations: • GET /api/rec/{id} – Retrieves a specific recommendation by ID. • POST /api/rec/{id} – Creates a new recommendation entry. • DELETE /api/rec/{id} – Deletes a specific recommendation. Bulk User and Recommendation Management These endpoints perform actions across multiple users or recommendations: • GET /api/users/ids – Lists all user IDs. • GET /api/users/statistics – Provides aggregated statistics about users. • GET /api/users/random – Generates a temporary, non-persistent random user (for testing). • DELETE /api/users – Deletes all user data. • GET /api/recs – Retrieves all recommendations for all users (can be heavy). • DELETE /api/recs – Deletes all recommendations. • GET /api/recs/ids – Lists all recommendation IDs. Engine and Server Operations • GET /api/status – Returns the operational status of the server. • GET /api/engine/run – Executes the recommendation engine globally. Debugging interface and integration features A dedicated testing user interface has been implemented to facilitate controlled evaluation of the AURORA Recommender. Once authenticated, the tester can generate synthetic users, import existing AURORA user profiles, upload transportation trip data extracted from Google Timeline, and import energy consumption data from Spanish utility bills. Through this interface, it is also possible to invoke the most relevant API methods directly, enabling efficient validation of data processing, rule execution, and recommendation generation workflows. The user interface is running here: https://aurora.linkeddata.es/ The system asks first for the credentials: AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. The web application requires user authentication before access. Once logged in, the interface presents an overview of the available users and their associated data. From this view, data can be downloaded in JSON format or uploaded directly from exports generated by the AURORA Dashboard. In addition, the application supports the upload of Spanish electricity bills as well as the import of mobility data from Google Timeline exports. All stored records can be browsed interactively, enabling users or testers to review and manage the information in a structured way. No personal information from the Google Timeline or the Spanish utility bills are sent to the servers –processing is made in the client side. The system permits forcing an engine run, deleting recommendations, etc. 3.3.3 Implementation details System description The AURORA Recommender is a standalone Java 11 application designed to generate personalized sustainability recommendations based on user consumption data. The component is implemented as a lightweight Java web application using the Jetty embedded server, enabling it to run without requiring an external application server. The API follows the HTTP REST style and leverages standard Java Servlet APIs for request handling. Core Functionalities Rule-Based Recommendation Engine AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. The main logic of the recommender is built on Drools, a forward-chaining rule engine that allows declarative modeling of sustainability rules and recommendation logic. This is an example of rule expressed in Drools: rule "Shower Efficiency" when $u: UserProfile(heatingFuel == "electric", householdSize > 2) then insert(new Recommendation( "Reduce showers by 2min - could save " + (householdSize * 25) + "kg CO₂ monthly", 2 )); end With this meaning: "If a user heats water with electricity and lives in a household with more than 2 people, recommend reducing shower time by 2 minutes to save CO ₂ ." Rules are authored and maintained separately from the core Java code, allowing flexible adaptation of the recommendation strategies without recompiling the system. Also, a custom syntax has been created for rules that live in a Excel spreadsheet. This spreadsheet and the easy syntax make the addition of rules possible for every. The custom syntax parses rules whose antecedent is in this format: country="UK" AND mainHeatingSystem="homePhotovoltaics" AND currentYear=2025 The engine can be executed for individual users or across the entire dataset, producing either targeted recommendations or bulk outputs. Data Processing and Serialization The recommender handles JSON-based data structures to represent user profiles, consumption histories, and generated recommendations. Jackson Databind is used for serialization and deserialization between Java objects and JSON. For additional data processing, particularly for structured reports or bulk operations, Apache POI is included, enabling import/export of Excel-based datasets. Persistence and Dynamic Configuration The Reflections library (org.reflections) supports dynamic discovery of annotated classes and resources at runtime, simplifying the registration of rules or components. This is used to enable different rule engines to co-exist (Drools, Excel based, etc.). The API can dynamically load available rule sets and apply them according to the context of each user. Email and Notification Support Jakarta Mail provides capabilities for sending email notifications. An email is sent every night, upon the regular execution. Logging and Monitoring The component includes robust logging capabilities to support monitoring and debugging. Logs are maintained for every critical operation (serving an API method, etc.). API Hosting and Security The application runs as a self-contained Jetty server exposing the REST API documented with Swagger UI, allowing authenticated users to test endpoints interactively. Security is implemented using Bearer Token Authentication, ensuring only authorized clients can access sensitive operations like retrieving or updating user data and triggering recommendation generation. Execution Environment Java 11 is required as the runtime environment. The application is packaged as a standard Java web application and can be deployed as a standalone service without requiring additional dependencies beyond those listed in the build configuration. Command line operation of the main functionalities is also possible. AURORA AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. Recommender is lightweight, adaptable, and maintainable, while supporting advanced rule-based reasoning over user-provided energy consumption data. Configuration # Minimal production configuration AUTH_TOKEN = your_project_token USERNAME = aurora_service_account PASSWORD = encrypted_db_password EMAIL_PASSWD = smtp_service_key # Optional integrations OPENAI_APIKEY = sk-proj-xxxxxxxxx WEATHER_APIKEY = 3xxxxxxxxxxxx1 DEEPSEEK_APIKEY = sk-xxxxxxxx # Operational parameters REPORT_SUSCRIBERS = [email protected],[email protected] DELAY = 0 System that runs the recommender system The system is a Dell PowerEdge R530 server running Ubuntu 16.04 (Xenial) with Linux kernel 4.4.0-210. It is equipped with two Intel Xeon E5‑2640 v4 processors, providing a total of 20 physical cores and 40 threads. The server has 192 GB of RAM, of which around 83 GB is currently in use. Storage is handled by a 2.4 TB PERC H730P Mini RAID controller. There is human maintenance that provides backups and other professional maintenance services. Configuration The configuration file, placed alongside the application's JAR file, contains essential parameters for system operation, including security credentials like the AUTH_TOKEN for API authentication, USERNAME and PASSWORD for service access, and EMAIL_PASSWD for notification services, along with integration keys for third-party services such as OPENAI_APIKEY for AI features, WEATHER_APIKEY for climate data, and DEEPSEEK_APIKEY for alternative AI processing, while REPORT_SUSCRIBERS defines report recipients and DELAY adjusts the minimum number of days between messages. Dictionary of user and context information The following list shows the elements that can become part of the rules. Variable Data Type Source Definition energyLabel enum inferred Latest energy label that is available monthlyElectricEnergy integer inferred Inferred from different electricity consumptions introduced, as an average month. Summer and winter bills are weighted to make the estimation. Data is in kWh peopleAtHome integer inferred From the consumptions introduced mainHeatingSystem enum inferred Inferred from different heating consumptions (biomass, oil, firewood, butane, electric, district, geothermal, solarThermal, locallyProducedBiomass, liquifiedPetroGas, naturalGas) mainElectricitySystem enum inferred Inferred from different electricity consumptions. default, defaultGreenProvider, pvInvestment, homePhotovoltaics mainTransportationSystem enum inferred Commuting is inferred between car, walking, metro, or bus. These types may appear in the trips: plane, otherBus, hybridCar, alternativeFuelBus, electricPassengerTrain, planeIntraEu, dieselBus, electricCar, electricBike, bike, metroTramOrUrbanLightTrain, motorcycle, electricScooter, planeExtraEu, walking, dieselPassengerTrain, electricBus, electricMotorcycle, fuelCar, hybridElectricBus, highSpeedTrain hasPlane boolean inferred True if the user has taken a plane this month hasIntraEUPlane boolean inferred True if the user has taken an intra EU plane this month AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. [34] BJ Fogg. Creating persuasive technologies: an eight-step design process. In Proceedings of the 4th international conference on persuasive technology, pages 1–6. ACM, 2009. [35] Ifeoma Adaji and Mikhail Adisa. A review of the use of persuasive technologies to influence sustainable behaviour. In Adjunct Proceedings of the 30th ACM Conference on User Modeling, Adaptation and Personalization, UMAP ’22 Adjunct, page 317–325, New York, NY, USA, 2022. Association for Computing Machinery. [36] Pearl Pu, Li Chen, and Rong Hu. A user-centric evaluation framework for recommender systems. In Proceedings of the fifth ACM conference on Recommender systems, pages 157–164, 2011. [37] Pearl Pu, Li Chen, and Rong Hu. A user-centric evaluation framework for recommender systems. In Proceedings of the fifth ACM conference on Recommender systems, pages 157–164. ACM, 2011. [38] Alexander Felfernig, Matthias Wundara, Thi Ngoc Trang Tran, Seda PolatErdeniz, Sven Lubos, Mohamad El Mansi, and Van Minh Le. Recommender systems for sustainability: overview and research issues. Frontiers in big Data, 6, 2023. [39] E. Kals and J. Maes. Sustainable development and emotions. In P. Schmuck and W. P. Schultz, editors, Psychology of Sustainable Development, pages 97–122. Springer, Boston, MA, 2002. [40] Peter Schmuck and Wesley P. Schultz. Sustainable development as a challenge for psychology. In Peter Schmuck and Wesley P. Schultz, editors, Psychology of Sustainable Development, pages 3–17. Springer, New York, NY, 2002. [41] Víctor Rodríguez-Doncel. La Web de Datos Enlazados: desarrollo técnico y fenómeno cultural. Bubok Publishing, 2021. [42] J. L. Austin. How to Do Things with Words. William James Lectures. Oxford University Press, Oxford, 1962. Edited by J. O. Urmson and Marina Sbisà. [43] L. Tsolas P. Conradie M. Amadori I. Koutsopoulos K. Ponnet S. Van Hove, M. Karaliopoulos. Profiling of energy consumers: psychological and contextual factors of energy behaviour, 2021. [44] C. L. Forgy. Rete: A fast algorithm for the many pattern/many object pattern match problem. In Readings in artificial intelligence and databases, pages 547–559. Morgan Kaufmann, 1989. [45] Manuela Sanguinetti and Maurizio Atzori. Conversational agents for energy awareness and efficiency: A survey. Electronics, 13(2):401, 2024. [46] G Phiri and P Trevorrow. Sustainable household food management using smart technology. In 2019 10th International Conference on Dependable Systems, Services and Technologies (DESSERT), pages 112–119. IEEE, 2019. [47] Jordana W Composto and Elke U Weber. Effectiveness of behavioural interventions to reduce household energy demand: a scoping review. Environmental Research Letters, 17(6):063005, 2022. [48] P. Kuyer, S. Shukla, F. O'Brolcháin, V. Rodríguez-Doncel, B. Gordijn. The ethics of digital nudging for sustainable energy consumption -- a closer look at the EU green deal. Technology in Society, 2025 [49] Dispositivo, Sistema y Procedimiento Inteligente para la Optimización del Consumo de Energía Eléctrica, V. Rodríguez, R. Enrich, P. Skovron y M. Torrent. Oficina Española de Patentes y Marcas, Publicación No. 2414581 (2013) AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. ANNEX I. RELATED PROJECTS This section presents the results of a constant search performed on the CORDIS webpage (https://cordis.europa.eu/). NUDGE (NUDging consumers towards enerGy Efficiency through behavioural science) 7 The NUDGE project was about making individualised behavioural nudging for responsible energy consumption. it addressed the need for responsible energy consumption by designing personalized behavioral interventions to promote lasting energy efficiency –much as if AURORA’s recommender had an entire EU project. NUDGE’s assumption is that current policy interventions often fail due to a lack of customization and ineffective testing conditions. Therefore, NUDGE combines consumer psychology analysis with intervention design, utilizing behavioural science methods to tailor strategies to individual behaviours. The project (2023-2023) tested new interventions across five EU countries, considering various demographics and energy use scenarios, to evaluate their effectiveness in real-world settings. The project was put into practice in five specific demo sites, with different targets: (i) Interdisciplinary project-based education on home energy consumption for children in Belgium; (ii) Efficient control of heating and DHW preparation for Natural Gas consuming boilers in Greece; (iii) Optimization of EV charging with self-produced PV power in Germany; (iv) Healthy homes for long-lasting energy efficiency behaviour in Portugal and (v) Promoting distributed self-production for local energy communities in Croatia –the most related one to AURORA would be Germany’s and Portugal’s, perhaps. The saving behaviours that can be the target of AURORA’s message were categorised and studied by NUDGE [43], listed here again for the readers’ convenience. Saving behaviours related to heating and cooling: • Turning heating off while air conditioning is on • Closing windows when heating is on • Keeping the doors closed to unheated areas in winter • Closing curtains and/or blinds to prevent heat loss in winter and heat gain in summer • Wearing more clothes instead of turning the heating up • Lowering daytime/nighttime thermostat setting • Turning off heating when absent • Turning down temperature in unused rooms Saving behaviours related to water use: • Reducing hot water temperature in thermostat settings • Reducing the number of baths/showers per week • Turning off tap when soaping up/cleaning teeth • Preferring a shower over bathing • Reducing showering time Saving behaviours related to kitchen activities: • Filling the kettle only with the needed amount of water before boiling • Cooking with pots covered • Only using the dishwasher when fully loaded • Turning off the tap when washing dishes • Using the energy-saving program (e.g., eco-mode) of the dishwasher • Defrosting the freezer to remove icing • Optimizing the temperature set point of cold appliances, such as refrigerator and freezer, to prevent freezing of the interior of the appliance Saving behaviours related to the general use of appliances: • Switching on electric devices when sun is shining as PV production is high • Charging my electric vehicle when sun is shining as PV production is high • Only using the washing machine when fully loaded • Frequently doing laundry at lower temperature, i.e., 40°C instead of 60°C 7 https://cordis.europa.eu/project/id/957012 AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. • Only using the tumble dryer when fully loaded • Using a clothes line rather than a tumble dryer • Turning off all unnecessary appliances completely when not in use (not in stand-by) • Switching off the TV when no one is watching • Turning off lights when leaving a room NUDGE did not pay attention however to transportation modes, nor emphasised solar energy as a tool to achieve the decarbonisation –still the previous list, and the derived psychological analysis that is found in the report, are of great interest. Upon joint discussion, a list of messages was discussed in AURORA, results follow in Figures 11 and 12. Figure 11. NUDGE Messages for AURORA ISAAC (Increasing Social Awareness and ACceptance of biogas and biomethane) 8 Its main objective was the construction of a communicative model aimed to spread balanced information, based on environmental and economic benefits, between all the actors potentially involved in the implementation of biogas/biomethane. Although the topic differs from AURORA’s, 8 https://cordis.europa.eu/project/id/691875 AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. the participatory process model that was developed as the main project’s approach to reduce social conflict is of much interest. uCARe (You can also reduce emissions) 9 The mission of this EU project (2020-2022) consisted in 1) proposing measures based on the collection of dispersed high-quality data and by studying motorists’ behaviour, 2) analysing the impact of user-related reduction measures and possibilities to improve existing pollutant emission legislation and policies, 3) creating easy to use instruments based on measures targeting the reduction of emissions, and 4) launching information campaigns through stakeholders. Triangulum (Triangulum: The Three Point Project / Demonstrate. Disseminate. Replicate) 10 The project’s objective (2015-2020) was the design and implementation of innovative business models and the activation of citizens as co-creators in order to base the technologies in real-world city environments and facilitate replication. The project produced a valuable monitoring and impact assessment framework. BECoop (Unlocking the community energy potential to support the market uptake of bioenergy heating technologies). 11 BECoop aimed at providing the necessary conditions and supporting tools to unlock the underlying market potential of EU bioenergy in order to promote a wide deployment of such energy technologies across Europe. The relation with AURORA lies in the existence of base communities on which to operate. WinWind (Winning social acceptance for wind energy in wind energy scarce regions). 12 The cornerstone of the project is to select, analyse, discuss, replicate, test and disseminate viable solutions to increase social acceptance and thereby the uptake of wind energy. Respon-SEA-ble (Sustainable oceans: our collective responsibility, our common interest. Building on real-life knowledge systems for developing interactive and mutual learning media). 13 The project objective consists in developing targeted and sound communication material that raises awareness responsibility and interest in ensuring the sustainability of the ocean and of its ecosystems. HARP (Heating Appliances Retrofit Planning) . The aim of the project (2019-2022) was to raise consumer awareness of the opportunities of a planned replacement of heating appliances. Some of the messages in this project might be delivered by the AURORA’s recommenders system. REPLACE (Making heating and cooling for European consumers efficient, economically resilient, clean and climate-friendly). 14 Very similar to the former, this project (2019-2023) aimed at raising awareness of the benefits of heating and cooling replacements by highlighting success stories in target regions. PENNY (Psychological, social and financial barriers to energy efficiency) 15 The project aimed at improving citizen understanding on behavioural mechanisms in energy efficiency, following an interdisciplinary and broad behavioural science approach. 9 https://cordis.europa.eu/project/id/815002 10 https://cordis.europa.eu/project/id/646578 11 https://cordis.europa.eu/project/id/952930 12 https://cordis.europa.eu/project/id/764717 13 https://cordis.europa.eu/project/id/652643 14 https://cordis.europa.eu/project/id/847087 15 https://cordis.europa.eu/project/id/723791 AURORA – D2.5 Technical and Social features of an app devoted to inform citizens on their environmental impact related to energy choices. Final version This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036418. GRETA (GReen Energy Transition Actions) 16 The project objective carried out a research to develop models and frameworks that reveal factors impacting both individual and collective energy citizenship actions. Other projects Other EU-funded projects also touched topics of interest for AURORA. Thus, EnerGAware 17 (2015-2018) explored serious gaming towards building energy engagement in social housing (but social housing is not a target in AURORA) and ENTROPY’s 18 focused on public buildings – resembling AURORA’s intent with public universities. for example InBetween 19 monitored users’ energy consumption to determine the ideal time to activate a home appliance in order to minimise costs and optimise any potential presence of renewable energy sources like solar panels or wind turbines. One of the lessons learnt from these projects is that heavily monitoring, although privacy intrusive, yields always better results. The above selected European projects seem to match not only the spirit of AURORA project, but also share several of its specific objectives. However, the available documentation available does not always comprehensively include the explanation of approaches, initiatives or internal discussions that every project meeting involves, which would be an invaluable resource of experience and expertise. On the other hand, the teams in charge of maintain the technical resources available online, quit this task when the project ends. That said, the following points should be highlighted: Occasionally, the projects documentation confirms common thesis or ideas supported by the literature. At uCARE project, the team warns of the risk of including economical rewards in gamification strategies, as these could discourage further involvement in any environmental initiative if not included or ever lead to cheating. At Triangulum project, the team recognises that the citizen engagement and socioeconomic impacts are the two impact domains most difficult to quantify. In relation to the former, the experience of AURORA project can ratify this point. The Triangulum project points to the complex methodological issues of ensuring a representative sample and an accurate survey procedure in collecting data on individual response and infrastructure performance. BeCoop project highlights the importance of an initial review focused on public awareness campaigns as well as personal opinions, perceptions, and people’s understanding of such initiatives. In whole, these tasks uncover essential resources for positive behavioural change. In particular, this project, as its names indicates (Unlocking the community energy potential to support the market uptake of bioenergy heating technologies) supports the participation of the public in decision-making processes and constant involvement throughout the lifecycle of such projects can lead to more socially acceptable and sustainable outcomes, such as higher engagement in taking action. Obviously, engaging the public on a large scale entails a major barrier, which consists in both the people’s high levels of knowledge and the acceptance of the related energy concept promoted. BeCoop project draws attention to its surveys and campaigns in order to disseminate its work and raise awareness of bioenergy. 16 https://cordis.europa.eu/project/id/101022317 17 https://cordis.europa.eu/project/id/649673 18 https://cordis.europa.eu/project/id/649849 19 https://cordis.europa.eu/project/id/768776