Intelligent Document Processing in End-to-End RPA Contexts: A Systematic Literature Review

Chapter 5 Intelligent Document Processing in End-to-End RPA Contexts: A Systematic Literature Review A. Martínez-Rojas , J. M. López-Carnicer , J. González-Enríquez , A. Jiménez-Ramírez , and J. M. Sánchez-Oliva Abstract Automating organizational processes typically involves document processing techniques for a large document set. For that purpose, the Intelligent Document Processing (IDP) paradigm has been studied for decades. With the fast emergence of Robotic Process Automation (RPA) in the process automation landscape, the industrial solution of IDP with RPA integration has risen significantly in the last few years. However, there is no up-to-date overview of the available knowledge in this area. Therefore, this chapter studies the current scientific knowledge about IDP and its integration into RPA through a systematic literature review that analyzed 77 primary studies. In addition, an industry review was performed, analyzing and characterizing 37 industrial tools. Although the results confirm the growth in the research interest in IDP in different dimensions, they also identify a lack of proposals that integrate IDP and RPA paradigms in confrontation with the industrial solutions that have increasingly led to its integration. This research has been supported by the NICO project (PID2019-105455GB-C31) of the Spanish Ministry of Science, Innovation and Universities and the CODICE project (EXP 00130458/IDI- 20210319 - P018-20/E09) of the Center for the Development of Industrial Technology (CDTI) and by the FPU scholarship program, granted by the Spanish Ministry of Education and Vocational Training (FPU20/05984). A. Martínez-Rojas (B )·J. M. López-Carnicer ·J. González-Enríquez ·A. Jiménez-Ramírez Departamento de Lenguajes y Sistemas Informáticos, Escuela Técnica Superior de Ingeniería Informática, Avenida Reina Mercedes, s/n., 41012 Sevilla, Spain e-mail: [email protected] J. M. López-Carnicer e-mail: [email protected] J. González-Enríquez e-mail: [email protected] A. Jiménez-Ramírez e-mail: [email protected] J. M. Sánchez-Oliva Servinform, S.A. Parque Industrial PISA, Calle Manufactura, 5, 41927 Mairena del Aljarafe, Sevilla, Spain e-mail: [email protected] © The Author(s) https://doi.org/10.1007/978-981-19-8296-5_5 95 96 A. Martínez-Rojas et al. Keywords Intelligent document processing (IDP) ·Robotic process automation (RPA) ·Systematic literature review ·Industrial review 5.1 Introduction The interest in the Robotic Process Automation (RPA) paradigm has increased in recent years [58,79]. One of the most recurrent topics is its integration with cognitive elements [18]. More precisely, contemporary literature acknowledges that the field of Intelligent Document Processing (IDP) is of utmost importance as a complementary technology in RPA [58], since IDP is a common element in multiple business domains related to RPA, e.g., back-office and accounting. Furthermore, in the industrial scope, most RPA companies include IDP components in their tools. This fact is mainly due to the dependency on documents within their business processes.1,2 The IDP paradigm refers to the set of processes that transform structured or unstructured documents into usable information [72]. Similar to RPA, this area has been significantly active since 2017 regarding topics like the improvement in Optical Character Recognition (OCR) systems [55], the inclusion of cognitive elements [49], and its integration with RPA [18]. Although the current situation of IDP is explored in some studies [8,19], they lack an in-depth study of the state of the art directly related to this topic. Meanwhile, other fields that are intrinsically related to IDP have been strongly studied, such as RPA [21,79], Artificial Intelligence (AI) integration [6,43], and specific areas inside IDP such as classification techniques [73]orOCR[52,55,87]. Nonetheless, they have been addressed as independent fields, thus lacking a broad vision in the area of IDP that shed light on their relationships and advances as a whole. Therefore, a comprehensive scientific study related to IDP in the context of RPA is necessary. Consequently, this chapter proposes a Systematic Literature Review (SLR) about IDP over the last years, which analyzes 77 documents and reports on the current state of this topic. In addition, the state of the industry related to this area is analyzed through the study of 37 tools. The analysis of these studies and tools reveals that the scientific studies are focused on the use of new techniques that improve the performance in specific phases of the IDP pipeline, leaving aside integration with RPA. In contrast, industrial tools emphasize massive document processing and integration with RPA. Finally, a series of gaps and challenges have been detected, which open up future lines of work in the field of IDP. The rest of the chapter is organized as follows. Section5.2 analyzes the related work that studies the importance of IDP in RPA contexts. Section5.3 describes the scientific scope methodology and its application, and Sect.5.4 describes the industrial research methodology and its application. Section 5.5 summarizes the 1https://www.automationanywhere.com/rpa/intelligent-document-processing. 2https://www.abbyy.com/flexicapture/. 5 Intelligent Document Processing in End-to-End RPA Contexts ... 97 limitations of the current work, discussions, gaps, and challenges. Finally, Sect.5.6 concludes the chapter and describes future work. 5.2 Related Works Reviews and surveys related to RPA, cognitive elements integration, or related fields of study widely appear in the literature. For instance, [36] investigates the definition of RPA by the community presenting an SLR. Enríquez et al. [21] analyze scientific publications and industrial tools in RPA by performing a systematic mapping study. In addition, Syed et al. [79] present an SLR proposing 15 challenges for future research. However, these works are just focused on RPA, and the relation with the IDP context is minimal. A wide variety of studies are related to technologies associated with IDP and RPA, e.g., OCR. These studies compare industrial tools [80], study the integration of OCR with multiples technologies [77], and review technological OCR approaches [55]. Conversely to our work, these studies are just focused on one of the techniques that may be applied in IDP and RPA projects. Although IDP is strongly related to RPA, not many studies investigate them together. Some studies analyze a particular part of the field, as Baviskar et al. [9] that presents a SLR that focuses on the processing of unstructured documents enhanced by AI, indicating future lines of research. In addition, Baviskar et al. [8] examine the influence and status of IDP from different perspectives, e.g., publications trend by year, publications type (i.e., conference and journal publication, among others), publications scientific discipline (i.e., computer science, engineering, and mathematics, among others), organized by authors, countries, organizations, and languages, and the most cited ones. Although this paper delivers a comprehensive statistical review of the literature, it does not analyze paradigms and techniques, unlike our study. Furthermore, Cristani et al. [19] develops a study about industrial IDP until 2018 that investigates the scientific publications and the approaches IDP companies implement. This study only considers certain business documents (i.e., invoices and orders). The authors also suggest an IDP architecture based on the study of the industries and scientific community perspective. Unlike other approaches, the current chapter aims to study the IDP field and its relationship with RPA from a broad perspective for any scientific and industrial document domain. These analyses will answer a series of research questions, selecting only those studies and tools that satisfy specific criteria. Additionally, thanks to this methodological and in-depth scientific/industrial analysis, a series of untracked gaps and challenges have been identified in the field of IDP and the related RPA. 98 A. Martínez-Rojas et al. Fig. 5.1 Adapted methodology from Kitchenham and Charters [44] 5.3 Scientific Methodology The SLR follows a widely accepted engineering methodology for conducting systematic reviews proposed by Kitchenham and Charters [44]. Figure 5.1 illustrates the method followed, including minor changes (i.e., reduce some steps) to adapt it to perform the industrial review. 5.3.1 Planning This section describes the steps of the planning phase, i.e., the (RQs) to be solved, the search strategy, and the method’s criteria selection. 5.3.1.1 Research Questions In this study, a total of 7 RQs have been proposed to provide a comprehensive analysis of the state of the art in the context of IDP and RPA proposed (cf. Table5.1). 5.3.1.2 Search Strategy This step details how the studies were retrieved in the different digital libraries. Table5.2 shows both the mandatory and the optional keywords that helped to build the queries. These words were combined in different digital libraries, i.e., IEEE Xplore, Scopus, Science Direct, and ACM. The concrete query strings executed for each digital library are shown in Table5.3. 5.3.1.3 Inclusion and Exclusion Criteria It is necessary to filter the results obtained from the queries to obtain the final list of primary studies to be analyzed in depth. The studies retrieved from the different libraries are filtered according to the inclusion/exclusion criteria described in Table5.4. 5 Intelligent Document Processing in End-to-End RPA Contexts ... 99 Table 5.1 Research questions RQ Question Aim RQ1 What are the types of publications of the proposals? Statistical analysis of studies based on the type of study (conference, article) and the number of publications RQ2 How are IDP approaches being validated? Verification of the types of validation in which the publications are included, whether scientific (running example, survey, or synthetic study), industrial (company case study, knowledge transfer, etc.), or without validation RQ3 What are the proposals researched? Analysis of the presented type of proposal (framework, algorithm, method, and theoretical analysis, among others) RQ4 How integrated is task automation? Analysis of integration with automated solutions or RPA. In case of not having it, evaluate its possible integration RQ5 What are the types of documents addressed? Analysis of the type of target document or specific application environment. Basic classification of the documents found, whether structured, unstructured, handwritten, or typed, and specify other relevant information about the documents RQ6 What are the main application domains? Context in which IDP proposals are developed. Domain refers to both the context of the problem being solved and the type of problem supported (banking, business, health, etc.) RQ7 What are the main phases in which IDP proposals are framed? Identification of the IDP pipeline phases more researched Table 5.2 Proposed keywords Mandatory words Optional words Document processing Intelligent, cognitive, automation, pipeline, structured, unstructured, handwritten, typewritten, preprocessing, segmentation, classification, extraction, detection, RPA, robotic process automation, OCR, character recognition 5.3.2 Conducting As depicted in Table5.5, the initial search obtained 1,405 studies in total. The pool of studies ended up with 77 primary studies after applying the five inclusion/exclusion criteria, removing the duplicated entries obtained from the different digital libraries, and including an expert recommendation. Additionally, to ensure the reliability of C5, one of the authors that did not execute this criterion randomly selected 30% of 100 A. Martínez-Rojas et al. Table 5.3 Executed queries Digital library Query IEEE XPLORE (“All Metadata”:“Document Processing”) AND (“All Metadata”:“intelligent” OR “All Metadata”:“cognitive” OR “All Metadata”:“automation” OR “All Metadata”:“pipeline” OR “All Metadata”:“character recognition” OR “All Metadata”:“automation” OR “All Metadata”:“structured” OR “All Metadata”:“unstructure” OR “All Metadata”:“handwritten” OR “All Metadata”:“typewritten” OR “All Metadata”:“preprocessing” OR “All Metadata”:“segmentation” OR “All Metadata”:“classification” OR “All Metadata”:“extraction” OR “All Metadata”:“detection” OR “All Metadata”:“RPA” OR “All Metadata”:“robot process automation”) Scopus TITLE-ABS-KEY ((“Document processing”) AND ((“intelligent” OR “cognitive” OR “automation” OR “pipeline” OR “RPA” OR “robot process automation” OR “structured” OR “unstructured” OR “handwritten” OR “typewritten” OR “preprocessing” OR “segmentation” OR “classification” OR “extraction” OR “detection” OR “OCR” OR “character recognition”))) Science Directa(“document processing”) AND (“intelligent” OR “cognitive” OR “automation” OR “pipeline”) (“document processing”) AND (“preprocessing” OR “segmentation” OR “classification” OR “extraction” OR “detection”) (“document processing”) AND (“OCR” OR “character recognition” OR “RPA” OR “robot process automation”) (“document processing”) AND (“structured” OR “unstructured” OR “handwritten” OR “typewritten”) ACM Title:(“document processing” AND (“intelligent” OR “cognitive” OR “automation” OR “OCR” OR “character recognition” OR “pipeline” OR “structured” OR “unstructured” OR “handwritten” OR “typewritten” OR “preprocessing” OR “segmentation” OR “classification” OR “extraction” OR “detection” OR “robot process automation” OR “RPA”)) OR Abstract: (“document processing” AND (“intelligent” OR “cognitive” OR “automation” OR “OCR” OR “character recognition” OR “pipeline” OR “structured” OR “unstructured” OR “handwritten” OR “typewritten” OR “preprocessing” OR “segmentation” OR “classification” OR “extraction” OR “detection” OR “robot process automation” OR “RPA”)) OR Keyword: (“document processing” AND (“intelligent” OR “cognitive” OR “automation” OR “OCR” OR “character recognition” OR “pipeline” OR “structured” OR “unstructured” OR “handwritten” OR “typewritten” OR “preprocessing” OR “segmentation” OR “classification” OR “extraction” OR “detection” OR “robot process automation” OR “RPA”)) aDue to search limitations, the 4 different searches were executed and pooled the results and applied it to check that studies were correctly included/excluded. The resulting primary studies are summarized, synthesizing its content in Tables 5.6 and 5.7.3 3A complete analysis of the analysis executed can be found on the sheet with the title “Scientific” within the Excel document available at https://doi.org/10.5281/zenodo.6400519. 5 Intelligent Document Processing in End-to-End RPA Contexts ... 101 Table 5.4 Inclusion and exclusion criteria Code Description C1 Mandatory and optional words must appear in the title, abstract, or keywords C2 Studies must have a publication date of 2017 or later. Although IDP has been studied for decades, a new interest has arisen as new ways of employing emerging technologies in this sector have been found through the use of RPA and AI (cf. Baviskar et al. [8]andin Fig.5.2), analyzing the publications in Scopus, one of the most prolific digital library between those used in the study This increase in the trend in IDP coincides in date with the increase in the number of publications in the field of RPA, as can be observed in the study presented by Enríquez et al. [21] C3 Studies must be written in English and fit the area of engineering and/or computer science C4 Only studies published in recognized rankings will be considered. More precisely, journal publications must belong to the Journal Citation Reports (JCR)a,and conferences must belong to the GII-GRIN-SCIE (GGS)2ranking C5 Studies must fit in the area of IDP and/or RPA. For this, the abstract and conclusions are analyzed aJournal Citation Reports, https://jcr.clarivate.com bConference Rating GGS, http://scie.lcc.uma.es/gii-grin-scie-rating/ratingSearch.jsf 5.3.3 Reporting Over the Research Questions The review methodology ends with the reporting step to answer the RQs. 5.3.3.1 RQ1. What are the Types of Publications of the Proposals? This section is intended to reveal the statistic analysis of the primary studies analyzed in Tables5.7 and 5.6. The origin is being investigated by means of the type of publication (i.e., journal or conference). In addition, the evolution of the trend in the number of publications is analyzed. As shown in Fig.5.3, the main type of contributions are conferences with 45 contributions (i.e., 58.44%) while 32 of them (i.e., 41.56%) correspond to journals. Additionally, as shown in Fig. 5.4, except for 2018, when there is a slight decrease in the number of publications, there is a clear upward trend up to the current date. Note that the number of publications in 2022 only considers the first month of the year and, therefore, it is expected that the trend will be kept. 5.3.3.2 RQ2. How are IDP Approaches Being Validated? The type of validation proposed in the primary studies for IDP is analyzed regarding the context where they took place. In this sense, three contexts were observed: (i) scientific context, where validations are experiments or case studies with synthetic or real data; (ii) industrial context, where the proposed validation is related to real-world 102 A. Martínez-Rojas et al. Fig. 5.2 Tendency of the number of IDP publications in the Scopus until December 2021 5 Intelligent Document Processing in End-to-End RPA Contexts ... 103 Table 5.5 Number of studies per digital library after applying the inclusion and exclusion criteria Criteria IEEE XPLORE Science direct Scopus ACM Totals C1 339 176 843 47 1405 C2 54 27 203 10 1243 C3 50 19 194 10 273 C4 27 17 88 7139 C5 22 12 68 5107 Duplicates 76 Recomm. 77 Table 5.6 Journal primary studies References Titles Years [86] Script identification of multi-script documents: a survey 2017 [33]An efficient open system for offline handwritten signature identification based on curvelet transform and one-class principal component analysis 2017 [84]Word graphs size impact on the performance of handwriting document applications 2017 [89]DivaServices-a RESTful web service for document image analysis methods 2017 [54] Sentence level matrix representation for document spectral clustering 2017 [3] Blind document image quality prediction based on modification of quality aware clustering method integrating a patch selection strategy 2018 [10]Text and non-text separation in offline document images: a survey 2018 [16]Deep neural networks for document processing of music score images 2018 [90]Paragraph vector representation based on word to vector and CNN learning 2018 [19]Future paradigms of automated processing of business documents 2018 [65] Text baseline detection, a single page trained system 2019 [4] A new threshold selection method based on fuzzy expert systems for separating text from the background of document images 2019 [82]CONFIRM—Clustering of noisy form images using robust matching 2019 [85] Document representation and classification with Twitter-based document embedding adversarial domain-adaptation and query expansion 2019 (continued) 110 A. Martínez-Rojas et al. Fig. 5.10 Total number of contributions per document processing phases Meanwhile, the remaining domains (i.e., music and agriculture) share the lowest level of interest with one publication (i.e., 1.3%). 5.3.3.7 RQ7. What are the Main Phases in Which IDP Proposals are Framed? This question aims to show which phases receive the most significant attention by the community. It should be borne in mind that the same proposal can deal with multiple document processing phases. As shown in Fig. 5.10, the field that attracts the most research interest is the improvement of the extraction/detection phase (i.e., 62.34%), which encompasses Natural Language Processing, OCR, entity detection, and other functionalities. Document segmentation ranks second (i.e., 28.57%). In the tail, preprocessing (i.e., 23.38%), classification (i.e., 18.18%), and other types of proposals not specifically related to a specific phase (e.g., configurations, training, and unitized phases) receive less interest from the community (i.e., 11.69%). 5.4 Industrial Methodology In this section, IDP-related software or commercial tools are analyzed following the same method illustrated in Fig.5.1. 5 Intelligent Document Processing in End-to-End RPA Contexts ... 111 Table 5.8 Research questions RQ Question Aim RQ1 What are the main software tools on the market for intelligent document processing? Find and identify the main (and the most promising) software tools on the market for IDP RQ2 What are the main functionalities that can characterize each phase of the IDP pipeline? Identify the degree of support offered by the main tools available in the market for the functionalities of the IDP pipeline phases RQ3 How well supported are the main industry tools concerning the general characteristics identified? Identify the degree of support provided by the main tools on the market for the general characteristics RQ4 How widespread are AI and automation in the IDP sector? Determine how and to what extent the analyzed tools use AI and the status of IDP automation Table 5.9 Keywords Mandatory words Related words Document processing Intelligent, cognitive, automation, pipeline, software, tool 5.4.1 Planning In this phase, the need for this review is identified within the industrial environment in terms of tools or software platforms that allow IDP management. For this purpose, a series of RQs are formulated, and the definition of the review protocol. 5.4.1.1 Research Questions This study tries to answer the following RQs to provide a comprehensive analysis of the state of IDP in RPA (cf. Table 5.8). 5.4.1.2 Search Strategy The first step in defining the search strategy is to identify the generalist search engines to be used for queries. The search engines selected were Google, Bing, and AOL Search (cf. Table5.9).4,5,6 4https://www.google.es/. 5https://www.bing.com. 6https://search.aol.com/aol/webhome. 112 A. Martínez-Rojas et al. Table 5.10 Inclusion and exclusion criteria Code Description C1 Software or service that includes document processing, whether this is a compliment or the primary objective of the tool C2 Current software with technical support and updates C3 Read the specifications provided about the tool, including whether it is software that includes IDP that meets the above criteria. If in doubt, read the associated documentation and tutorials C4 Read the specifications provided about the tool, including whether it is software that includes IDP that meets the above criteria. If in doubt, read the associated documentation and tutorials 5.4.1.3 Inclusion and Exclusion Criteria The tools which are obtained from the search engines are filtering according to the following inclusion/exclusion criteria (cf. Table 5.10). The procedure was to try to answer each of the RQs posed and save each of the characteristics that defined it to define a classification criterion for each selected tool. Therefore, a series of defining characteristics will be obtained for each question, cf. Table5.11. Those tools for which it has not been possible to obtain transparent information obtain a result of N/A. There could be information about it, but not easily accessible, or it could be that, since they do not have such functionality, no reference to it is explicitly stated in any documentation. 5.4.2 Conducting The search process described in Sect.5.4.1 is carried out using different search engines. These searches result in a set of tools shown in Table 5.12.7 5.4.3 Reporting over the Research Questions Once all the industrial tools have been presented together with the knowledge extracted from each one, each research question can be answered. 7The relationship between these tools and the characteristics described in Table 5.11 are shown in the sheet with the title “Industrial” in the Excel document available at https://doi.org/10.5281/ zenodo.6400519. 5 Intelligent Document Processing in End-to-End RPA Contexts ... 113 Table 5.11 General features Ref General features Description G1 Automation level The level of automation available in the IDP tool. If it has some type of built-in automation (Yes), if it does not have any type of automation (No), if it has automation through integration with other platforms (Partial) G2 Phases of the IDP pipeline supported Phases of the IDP pipeline are supported by each evaluated tool. Identify what these phases are (Preprocessing, segmenting, extraction/detection, classification, others), indicating whether it is performed (Yes) or not contemplated (No) G3 Tool provides a web distribution If the tool is a web version (Yes), desktop (No), both through shared features (Partial) G4 Free license Whether the license type is free (Yes), partially free with limitations (Partial), or commercial only (No) G5 Acceptable learning curve Level of difficulty of the learning curve, based on a subjective opinion of the author of this report based on the tutorials and documentation found, indicating whether it is relatively easy (Yes), medium difficulty (Partial), or complex (No) G6 Deployed on own servers with no external visibility In case it is a web tool if it can be self-deployable: Yes, No G7 Process structured and unstructured information Support for structured, unstructured, or both documents G8 Graphical process editor The tool provides graphical editors to facilitate the user to build the process execution flow with, for example, drag&drop options (Yes) or not (No) G9 Visual adaptation and configuration Allows the adaptability of visual elements (icons, logos, etc.): Yes, No, Partial G10 Allows process configuration It allows documenting in detail the characteristics of each process, including objectives, metrics, deliverables, assumptions, team members, scope, stakeholders, customers, input data, output data, customer requirements, other comments, etc. It is classified as Yes or No G11 Simultaneous management of several processes The tool allows you to manage multiple document processing processes simultaneously: Yes or No G12 Reporting and statistics The tool provides statistics on the documents handled: Yes or No G13 Robot output flow configuration The tool allows you to define and configure the exit flow of the process in case of failures in the usual flow: Yes, No, Partial (continued) 114 A. Martínez-Rojas et al. Table 5.11 (continued) Ref General features Description G14 Debug process mode The tool has a debugging and step-by-step execution mode and the inspection of variables: Yes, No, in case of having some kind of trace that allows seeing the errors, but it is not a debugging mode, Partial G15 Export process The tool allows the export of processes configured in the platform: Yes, No, in case of having some kind of export but it is not an export functionality, Partial G16 Multiple users The tool allows access to multiple users at the same time: Yes or No G17 Supported languages The tool allows you to process documents in multiple languages. If it allows 5 or more languages, it will be considered multiple languages; if it has only accepted languages, they will be made explicit. 5.4.3.1 RQ1. What are the Main Software Tools on the Market for Intelligent Document Processing? As previously found, the field of IDP has been studied for decades, so there have been industrial tools for IDP years plaguing the market. An external report (i.e., Everest Group [72]) has been chosen to check the results and compare the tools mentioned in those reports with the tools found in the present study. This report marked as leaders the tools offered by WorkFusion, Automation Anywhere, Kofax, ABBYY, IBM, and Antworks. Then, except for HCL Technologies, the companies and products in the external report match the search results and have not been excluded by the inclusion/exclusion criteria. The conclusion is that these tools are not only the main ones according to the external reports, but they also coincide with the results of this study. 5.4.3.2 RQ2. What are the Main Functionalities that Can Characterize Each Phase of the IDP Pipeline? The IDP pipeline identified during the scientific domain contains the following phases: preprocessing, segmentation, extraction/detection, classification, and others. During the analysis of the industrial tools, it has been observed that the same pipeline phases are also widely used. Nevertheless, some additional phases were identified, such as document import, data export, and human interaction for data validation or creation of the IDP processes by choosing the phases. These phases are not presented in the scientific study and have not been considered for our IDP pipeline since they 5 Intelligent Document Processing in End-to-End RPA Contexts ... 115 Table 5.12 Industrial primary results Company Tool Automation anywhere Intelligent document processing UIPath Document understanding Blue Prism Decipher IDP Google Document AI Amazon AWS Textract/Comprehend/A2I IBM Datacap ABBYY FlexiCapture/FineReader engine Azure Form recognizer ROSSUM DATA CAPTURE ANTWORKS Document processing Hyland Brainware Intelligent Capture Foxit PDFCompresor/Maestro server OCR/Document transformation service SoftWorksAI Trapeze OpenText Intelligent capture + Magellan Parascript FormXtra Alfresco Intelligent information capture (via AWS) forms recognition and processing EPHESOFT Intelligent document processing WorkFusion Document intelligent Infrrd Intelligent document processing Edgeverve AssistEdge DMS-solutions elDoc Kofax Kofax transformation BIS Grooper Bizdata SmartDetect Hypatos Hypatos document processing Kanverse Kanverse IDP Parashift Parashift process documents HyperScience Intelligent document processing Appian Appian IDP Datamatics TruCap+ Docdigitizer Docdigitizer document processing Goneutrinos Intelligent document processing Acodis Intelligent document processing AutomationHero AutomationHero document processing Indico Indico intelligent process automation OpenBots OpenBots documents Nanonets Nanonets 116 A. Martínez-Rojas et al. Fig. 5.11 Total number of tools according to their phase in the IDP pipeline correspond to tasks related but not directly to document processing. However, some activities related to the IDP process, which some companies carry out, were cataloged in the phase others. This is because they could not be in a specific phase when performing actions that could be classified in more than one phase or carrying out process preparation work. The most widespread point, and why most of the tools can be included in the others phase, is the possibility of training or configuring the intelligent model that would be used in the IDP process. Regarding the pipeline phases identified during the execution of the industrial report, they coincide with those identified in the analysis of the scientific scope. Figure5.11 shows how many tools contemplate the fulfillment of the identified pipeline phases. As can be seen, all the tools perform the phases of classification and content extraction/detection, thus being the main pipeline functionalities used in the industry. It should be noted that another important phase is segmentation, where the different sections that make up the documents are identified. Meanwhile, preprocessing and others take the last position for some of the tools. In short, 65% of the tools fulfill all the phases of the pipeline since part of the tools do not have information on preprocessing or explicitly report the existence of additional phases (cf. “N/A” label in Fig.5.11). 5.4.3.3 RQ3. How Well Supported are the Main Industry Tools Concerning the General Characteristics Identified? Concerning the general characteristics that appear in Sect. 5.4.1, a series of data have been extracted that can be analyzed. 5 Intelligent Document Processing in End-to-End RPA Contexts ... 117 Fig. 5.12 Total number of tools per license type Fig. 5.13 Total number of tools according to their learning curve To begin with, in Fig.5.12 can be observed that almost all the tools have a commercial license, but some offer a community version with reduced capabilities or a free version with limited time use or with fewer features than the commercial version. Only 2 tools are open source, and these are precisely the ones that allow visual adaptation or modification of the software due to the nature of their license type. The graph represents the freeware of the tools, being the community and open source tools considered free, those with some limited freeware as partial, and those purely commercial as non-free. As can be seen, the industry focuses on commercial tools. Those with limited free, limited community,oropen source versions have fewer features than the paid versions. Only OpenText offers complete open source for free. Regarding the difficulty of using the tool, it is subjective data, so the perception of the difficulty of use of the tools may vary among users. However, as can be seen in Fig.5.13, all the tools offer a user-friendly interface, increasing complexity concerning the use of advanced functions. Half of the tools (i.e., 48.65%) in Fig.5.13 are easy to use, with well-defined steps, and some have slight complexity (i.e., 29.73%). Also, concerning the deployment of the system, it is extracted from Fig. 5.14 that half of the tools (i.e., 49%) allow self-deploy. Except for one tool, all of them claim to have features for handling both unstructured and structured documents, either through the tool’s functions or through thirdparty functions that can be included in these tools. 118 A. Martínez-Rojas et al. Fig. 5.14 Total number of tools per type of deployment Fig. 5.15 Total number of tools that allows building IDP process Fig. 5.16 Total number of tools that are able to run multiple processes Those that allow building an IDP process are scarce, and several of them are included in the elaboration of the RPA creation process. This reduces the IDP process creation level but allows some control over its status. Only 35% of the tools in Fig.5.15 contemplate the construction of the IDP process, most of them being a process with a defined pipeline, although they allow parameterizing and configuring the process already defined in the tool. Most of the tools investigated in Fig.5.16 allows the management of multiple IDP processes. Statistical reports of the IDP process are present in most of the tools (i.e., 90%) in Fig.5.17. However, there are variations in the amount of data shown in the reports, from some exhaustive tools to others that only offer information regarding case counts or process usage. 5 Intelligent Document Processing in End-to-End RPA Contexts ... 119 Fig. 5.17 Percentage of tools that generates statistics Fig. 5.18 Total number of tools that allows to configure processes Fig. 5.19 Percentage of tools that allows to debug processes In some tools, the output configuration of the IDP process is automatic, while in others it requires a specific configuration. There is a disparity of options depending on the difficulty of use intended to give to the tool. Although it is important to note that a significant number of tools lack the capability to provide error reporting, requiring manual validation from the user, generating failure reports, or simply failing to process documents that cause errors. In Fig.5.18, it can be seen that the same number of tools (i.e., 32.43%) of the tools allow and do not allow configuring the process, while a partial configuration contemplates by 8.11% of the tools. As for the tools that have error debugging mode, 19 of the tools in Fig. 5.19 have some debugging, either direct or partial. The partial debugging mode corresponds to the use of logs or visuals, indicating the existence of incorrectly extracted data and the need to reconfigure the process. 126 A. Martínez-Rojas et al. thanks to additional phases of document template configuration and AI retraining. Nevertheless, industrial research shows that the companies are more focused on IDP pipeline phases, such as classification and extraction/detection, which causes some tools to skip some of the phases of the process. All this, combined with the ability to process documents in several languages, aims to reach customers worldwide and process documents on a massive scale to maximize their profitability from implementing the tools. In addition, the current chapter leads to some future works to be explored. First, integrating IDP processes in business process management or RPA methodology would be beneficial due to (1) the strong relations between them and (2) the research gap acknowledged in the current chapter—overall from an industrial perspective. Second, although there are already proposals to facilitate the use of IDP with developments or Application Programming Interfaces [45], it can be investigated how to best integrate the IDP processes in RPA, how to include new techniques with the advance of research, how to integrate different approaches depending on their needs, and how to train a machine learning models for specific use cases. This point may lead to an IDP/RPA framework to facilitate their use. Finally, going beyond IDP, the performance and suitability of different technologies can be evaluated when paralleling or scaling up tasks with RPA. The critical point here is to identify the challenges in integrating these technologies and the possible limitations of such integration. Appendix A: List of Abbreviations RPA Robotic Process Automation IDP Intelligent Document Processing AI Artificial Intelligence OCR Optical Character Recognition SLR Systematic Literature Review RQs Research Questions References 1. Abdallah, A., Berendeyev, A., Nuradin, I., Nurseitov, D.: Tncr: Table net detection and classification dataset. Neurocomputing (2021) 2. Ahmed, R., Gogate, M., Tahir, A., Dashtipour, K., Al-Tamimi, B., Hawalah, A., El-Affendi, M.A., Hussain, A.: Deep neural network-based contextual recognition of Arabic handwritten scripts. Entropy 23(3), 340 (2021) 3. Alaei, A., Conte, D., Martineau, M., Raveaux, R.: Blind document image quality prediction based on modification of quality aware clustering method integrating a patch selection strategy. Exp. Syst. Appl. 108, 183–192 (2018) 4. Annabestani, M., Saadatmand-Tarzjan, M.: A new threshold selection method based on fuzzy expert systems for separating text from the background of document images. Iran. J. Sci. Technol. Trans. Electr. Eng. 43(1), 219–231 (2019) 5 Intelligent Document Processing in End-to-End RPA Contexts ... 127 5. Au, W., Ait-Azzi, A., Kang, J.: Finsbd-2021: the 3rd shared task on structure boundary detection in unstructured text in the financial domain. In: Companion Proceedings of the Web Conference, pp. 276–279 (2021) 6. Baidya, A.: Document analysis and classification: a robotic process automation (RPA) and machine learning approach. In: 2021 4th International Conference on Information and Computer Technologies (ICICT). pp. 33–37. IEEE (2021) 7. Banumathi, K.L., Jagadeesh Chandra, A.P.: An approach to estimate skew angle in printed document images. In: 2019 1st International Conference on Advances in Information Technology (ICAIT), pp. 480–484. IEEE (2019) 8. Baviskar, D., Ahirrao, S., Kotecha, K.: A bibliometric survey on cognitive document processing. Libr. Philos. Pract. 1–31 (2020) 9. Baviskar, D., Ahirrao, S., Potdar, V., Kotecha, K.: Efficient automated processing of the unstructured documents using artificial intelligence: a systematic literature review and future directions. IEEE Access (2021) 10. Bhowmik, S., Sarkar, R., Nasipuri, M., Doermann, D.: Text and non-text separation in offline document images: a survey. Int. J. Doc. Anal. Recogn. (IJDAR) 21(1), 1–20 (2018) 11. Biswas, S., Riba, P., Lladós, J., Pal, U.: Beyond document object detection: instance-level segmentation of complex layouts. Int. J. Doc. Anal. Recogn. (IJDAR) 24(3), 269–281 (2021) 12. Boro¸s, E., Romero, V., Maarand, M., Zenklová, K., Kˇreˇcková, J., Vidal, E., Stutzmann, D., Kermorvant, C.: A comparison of sequential and combined approaches for named entity recognition in a corpus of handwritten medieval charters. In: 2020 17th International Conference on Frontiers in Handwriting Recognition (ICFHR), pp. 79–84. IEEE (2020) 13. Bourbakis, N., Mertoguno, S.: A holistic approach for automatic deep understanding and protection of technical documents. Int. J. Artif. Intell. Tools 29(06), 2050007 (2020) 14. Bukhari, S.S., Kadi, A., Jouneh, M.A., Mir, F.M., Dengel, A.: anyocr: An open-source OCR system for historical archives. In: 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR), vol. 1, pp. 305–310. IEEE (2017) 15. Burget, R.: Model-based integration of unstructured web data sources using graph representation of document contents. In: WEBIST, pp. 326–333 (2019) 16. Calvo-Zaragoza, J., Castellanos, F.J., Vigliensoni, G., Fujinaga, I.: Deep neural networks for document processing of music score images. Appl. Sci. 8(5), 654 (2018) 17. Can, Y.S., Kabadayı, M.E.: Line segmentation of individual demographic data from Arabic handwritten population registers of ottoman empire. In: International Conference on Document Analysis and Recognition, pp. 312–321. Springer (2021) 18. Chakraborti, T., Isahagian, V., Khalaf, R., Khazaeni, Y., Muthusamy, V., Rizk, Y., Unuvar, M.: From robotic process automation to intelligent process automation. In: International Conference on Business Process Management, pp. 215–228. Springer (2020) 19. Cristani, M., Bertolaso, A., Scannapieco, S., Tomazzoli, C.: Future paradigms of automated processing of business documents. Int. J. Inform. Manag. 40, 67–75 (2018) 20. Degtyarenko, I., Deriuga, I., Grygoriev, A., Polotskyi, S., Melnyk, V., Zakharchuk, D., Radyvonenko, O.: Hierarchical recurrent neural network for handwritten strokes classification. In: ICASSP 2021–2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2865–2869. IEEE (2021) 21. Enríquez, J.G., Jimenez-Ramirez, A., Dominguez-Mayo, F., Garcia-Garcia, J.: Robotic process automation: a scientific and industrial systematic mapping study. IEEE Access 8, 39113–39129 (2020) 22. Fawzi, A., Pastor, M., Martínez-Hinarejos, C.D.: Baseline detection on Arabic handwritten documents. In: Proceedings of the 2017 ACM Symposium on Document Engineering, pp. 193–196 (2017) 23. Feng, D., Chen, H.: A small samples training framework for deep learning-based automatic information extraction: case study of construction accident news reports analysis. Adv. Eng. Inform. 47, 101256 (2021) 24. Fenton, K., Simske, S.: Engineering of an artificial intelligence safety data sheet document processing system for environmental, health, and safety compliance. In: Proceedings of the 21st ACM Symposium on Document Engineering, pp. 1–4 (2021) 128 A. Martínez-Rojas et al. 25. Fernandes, J., Simsek, M., Kantarci, B., Khan, S.: Tabledet: an end-to-end deep learning approach for table detection and table image classification in data sheet images. Neurocomputing 468, 317–334 (2022) 26. Fugini, M., Finocchi, J.: Quality evaluation for documental big data. In: Proceedings of the 22nd International Conference on Enterprise Information Systems—Volume 1: ICEIS, pp. 132–139. INSTICC, SciTePress (2020). https://doi.org/10.5220/0009394301320139 27. Gatos, B., Louloudis, G., Stamatopoulos, N., Sfikas, G.: Historical document processing. In: Proceedings of the 2017 ACM Symposium on Document Engineering, pp. 1–2 (2017) 28. Gómez-Pérez, P., Phan, T.N., Küeng, J.: Agricultural knowledge extraction from text sources using a distributed mapreduce cluster. In: 2016 27th International Workshop on Database and Expert Systems Applications (DEXA), pp. 29–33. IEEE (2016) 29. Gorai, M., Nene, M.J.: Layout and text extraction from document images using neural networks. In: 2020 5th International Conference on Communication and Electronics Systems (ICCES), pp. 1107–1112. IEEE (2020) 30. Grygoriev, A., Degtyarenko, I., Deriuga, I., Polotskyi, S., Melnyk, V., Zakharchuk, D., Radyvonenko, O.: Hcrnn: a novel architecture for fast online handwritten stroke classification. In: International Conference on Document Analysis and Recognition, pp. 193–208. Springer (2021) 31. Guerry, C., Coüasnon, B., Lemaitre, A.: Combination of deep learning and syntactical approaches for the interpretation of interactions between text-lines and tabular structures in handwritten documents. In: 2019 International Conference on Document Analysis and Recognition (ICDAR), pp. 858–863. IEEE (2019) 32. Ha, H., Horák, A.: Information extraction from scanned invoice images using text analysis and layout features. Signal Proc.: Image Commun., 116601 (2021) 33. Hadjadji, B., Chibani, Y., Nemmour, H.: An efficient open system for offline handwritten signature identification based on curvelet transform and one-class principal component analysis. Neurocomputing 265, 66–77 (2017) 34. Hammarström, H.: Inventory and content separation in grammatical descriptions of languages of the world. In: International Conference on Theory and Practice of Digital Libraries, pp. 29–40. Springer (2021) 35. Holeˇcek, M.: Learning from similarity and information extraction from structured documents. Int. J. Doc. Anal. Recogn. (IJDAR), 1–17 (2021) 36. Ivanˇci´c, L., Suša Vugec, D., Bosilj Vukši´c, V.: Robotic process automation: systematic literature review. In: International Conference on Business Process Management, pp. 280–295. Springer (2019) 37. Jalali, F., Ebrahimi, A.: A novel mixed approach for detecting overlap in document images. In: 2017 Iranian Conference on Electrical Engineering (ICEE), pp. 1701–1707. IEEE (2017) 38. Jiang, J., Simsek, M., Kantarci, B., Khan, S.: High precision deep learning-based tabular position detection. In: 2020 IEEE Symposium on Computers and Communications (ISCC), pp. 1–7. IEEE (2020) 39. Jun, C., Suhua, Y., Shaofeng, J.: Automatic classification and recognition of complex documents based on faster RCNN. In: 2019 14th IEEE International Conference on Electronic Measurement & Instruments (ICEMI), pp. 573–577. IEEE (2019) 40. Kajla, N.I., Missen, M.M.S., Luqman, M.M., Coustaty, M., Mehmood, A., Choi, G.S.: Additive angular margin loss in deep graph neural network classifier for learning graph edit distance. IEEE Access 8, 201752–201761 (2020) 41. Kara, E., Traquair, M., Kantarci, B., Khan, S.: Deep learning for recognizing the anatomy of tables on datasheets. In: 2019 IEEE Symposium on Computers and Communications (ISCC), pp. 1–6. IEEE (2019) 42. Kara, E., Traquair, M., Simsek, M., Kantarci, B., Khan, S.: Holistic design for deep learningbased discovery of tabular structures in datasheet images. Eng. Appl. Artif. Intell. 90, 103551 (2020) 43. Keeling, R., Chhatwal, R., Huber-Fliflet, N., Zhang, J., Wei, F., Zhao, H., Shi, Y., Qin, H.: Empirical comparisons of CNN with other learning algorithms for text classification in legal document review. In: 2019 IEEE International Conference on Big Data (Big Data), pp. 2038– 2042. IEEE (2019) 5 Intelligent Document Processing in End-to-End RPA Contexts ... 129 44. Kitchenham, B., Charters, S.: Guidelines for performing systematic literature reviews in software engineering (2007) 45. Lenc, L., Martínek, J., Král, P., Nicolao, A., Christlein, V.: Hdpa: historical document processing and analysis framework. Evol. Syst. 12, 177–190 (2021) 46. Li, D., Wu, Y., Zhou, Y.: Linecounter: Learning handwritten text line segmentation by counting (2021). arXiv preprint arXiv:2105.11307 47. Li, J., Lin, C.M., Hu, S.x.: Intelligent document processing method based on robot process automation. In: 2021 Global Reliability and Prognostics and Health Management (PHM- Nanjing), pp. 1–6. IEEE (2021) 48. Ljaji´c, A., Stankovi´c, M., Marovac, U.: Detection of negation in the Serbian language. In: Proceedings of the 8th International Conference on Web Intelligence, Mining and Semantics, pp. 1–6 (2018) 49. Long, S., He, X., Yao, C.: Scene text detection and recognition: the deep learning era. Int. J. Comput. Vis. 129(1), 161–184 (2021) 50. Mansar, Y., Kang, J., Maarouf, I.E.: The finsim-2 2021 shared task: learning semantic similarities for the financial domain. In: Companion Proceedings of the Web Conference 2021, pp. 288–292 (2021) 51. Martínez-Rojas, A., Sánchez-Oliva, J., López-Carnicer, J., Jiménez-Ramírez, A.: Airpa: An architecture to support the execution and maintenance of AI-powered RPA robots. In: International Conference on Business Process Management, pp. 38–48. Springer (2021) 52. Memon, J., Sami, M., Khan, R.A., Uddin, M.: Handwritten optical character recognition (OCR): a comprehensive systematic literature review (SLR). IEEE Access 8, 142642–142668 (2020) 53. Mercier, D., Rizvi, S.T.R., Rajashekar, V., Dengel, A., Ahmed, S.: Impactcite: An xlnet-based solution enabling qualitative citation impact analysis utilizing sentiment and intent. In: ICAART (2), pp. 159–168 (2021) 54. Mijangos, V., Sierra, G., Montes, A.: Sentence level matrix representation for document spectral clustering. Pattern Recogn. Lett. 85, 29–34 (2017) 55. Mittal, R., Garg, A.: Text extraction using OCR: a systematic review. In: 2020 Second International Conference on Inventive Research in Computing Applications (ICIRCA), pp. 357–362. IEEE (2020) 56. Moreno, M., Schirmer, L., Bayser, M., Brandão, R., Cerqueira, R.: Understanding documents with hyperknowledge specifications. In: Proceedings of the ACM Symposium on Document Engineering 2018, pp. 1–4 (2018) 57. Neji, H., Halima, M.B., Hamdani, T.M., Nogueras-Iso, J., Alimi, A.M.: Blur2sharp: a GAN- based model for document image deblurring. Int. J. Comput. Intell. Syst. 14(1), 1315–1321 (2021) 58. Ng, K.K., Chen, C.H., Lee, C.K., Jiao, J.R., Yang, Z.X.: A systematic literature review on intelligent automation: aligning concepts from theory, practice, and future perspectives. Adv. Eng. Inform. 47, 101246 (2021) 59. Nguyen, M.T., Le, D.T., Le, L.: Transformers-based information extraction with limited data for domain-specific business documents. Eng. Appl. Artif. Intell. 97, 104100 (2021) 60. Obukhov, A., Krasnyanskiy, M.: Application of machine learning for document classification and processing in adaptive information systems. In: Computer Science On-line Conference, pp. 291–300. Springer (2020) 61. Oliveira, S.A., Seguin, B., Kaplan, F.: dhsegment: A generic deep-learning approach for document segmentation. In: 2018 16th International Conference on Frontiers in Handwriting Recognition (ICFHR), pp. 7–12. IEEE (2018) 62. Oral, B., Emekligil, E., Arslan, S., Eryiˇgit, G.: Information extraction from text intensive and visually rich banking documents. Inform. Proc. Manag. 57(6), 102361 (2020) 63. Pappu, A., Blanco, R., Mehdad, Y., Stent, A., Thadani, K.: Lightweight multilingual entity extraction and linking. In: Proceedings of the Tenth ACM International Conference on Web Search and Data Mining, pp. 365–374 (2017) 64. Park, C., Shin, J., Park, S., Lim, J., Lee, C.: Fast end-to-end coreference resolution for Korean. In: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Findings, pp. 2610–2624 (2020) 130 A. Martínez-Rojas et al. 65. Pastor, M.: Text baseline detection, a single page trained system. Pattern Recogn. 94, 149–161 (2019) 66. Philips, J., Tabrizi, N.: Historical document processing: A survey of techniques, tools, and trends. In: KDIR, pp. 341–349 (2020) 67. Poddar, A., Chakraborty, A., Mukhopadhyay, J., Biswas, P.K.: Detection and localisation of struck-out-strokes in handwritten manuscripts. In: International Conference on Document Analysis and Recognition, pp. 98–112. Springer (2021) 68. Poddar, A., Chakraborty, A., Mukhopadhyay, J., Biswas, P.K.: Texrgan: a deep adversarial framework for text restoration from deformed handwritten documents. In: Proceedings of the Twelfth Indian Conference on Computer Vision, Graphics and Image Processing, pp. 1–9 (2021) 69. Qasim, S.R., Mahmood, H., Shafait, F.: Rethinking table recognition using graph neural networks. In: 2019 International Conference on Document Analysis and Recognition (ICDAR), pp. 142–147. IEEE (2019) 70. Qurashi, A.W., Holmes, V., Johnson, A.P.: Document processing: methods for semantic text similarity analysis. In: 2020 International Conference on INnovations in Intelligent SysTems and Applications (INISTA), pp. 1–6. IEEE (2020) 71. Rabby, A.S.A., Islam, M.M., Hasan, N., Nahar, J., Rahman, F.: A novel deep learning characterlevel solution to detect language and printing style from a bilingual scanned document. In: 2020 IEEE International Conference on Big Data (Big Data), pp. 5218–5226. IEEE (2020) 72. Research, E.G.: Everest group peak matrix tm for intelligent document processing (IDP) (2021) 73. Ribeiro, J., Lima, R., Paiva, S.: Document classification in robotic process automation using artificial intelligence—a preliminary literature review. Commun. Intell. Syst., 211–221 (2021) 74. Rusticus, D., Goldmann, L., Reisser, M., Villegas, M.: Document domain adaptation with generative adversarial networks. In: 2019 International Conference on Document Analysis and Recognition (ICDAR), pp. 1432–1437. IEEE (2019) 75. Santoro, A., De Stefano, C., Marcelli, A.: Assisted transcription of historical documents by keyword spotting: a performance model. In: 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR), vol. 1, pp. 971–976. IEEE (2017) 76. Shen, Z., Zhang, R., Dell, M., Lee, B.C.G., Carlson, J., Li, W.: Layoutparser: A unified toolkit for deep learning based document image analysis (2021). arXiv preprint arXiv:2103.15348 77. Shidaganti, G., Salil, S., Anand, P., Jadhav, V.: Robotic process automation with AI and OCR to improve business process. In: 2021 Second International Conference on Electronics and Sustainable Communication Systems (ICESC), pp. 1612–1618. IEEE (2021) 78. Sirajudeen, M., Anitha, R.: Forgery document detection in information management system using cognitive techniques. J. Intell. Fuzzy Syst. 39(6), 8057–8068 (2020) 79. Syed, R., Suriadi, S., Adams, M., Bandara, W., Leemans, S.J., Ouyang, C., ter Hofstede, A.H., van de Weerd, I., Wynn, M.T., Reijers, H.A.: Robotic process automation: contemporary themes and challenges. Comput. Indus. 115, 103162 (2020) 80. Tafti, A.P., Baghaie, A., Assefi, M., Arabnia, H.R., Yu, Z., Peissig, P.: OCR as a service: an experimental evaluation of Google Docs OCR, Tesseract, ABBYY Finereader, and Transym. In: International Symposium on Visual Computing, pp. 735–746. Springer (2016) 81. Tensmeyer, C., Martinez, T.: Document image binarization with fully convolutional neural networks. In: 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR), vol. 1, pp. 99–104. IEEE (2017) 82. Tensmeyer, C., Martinez, T.: Confirm-clustering of noisy form images using robust matching. Pattern Recogn. 87, 1–16 (2019) 83. Tomoiaga, C., Feng, P., Salzmann, M., Jayet, P.: Field typing for improved recognition on heterogeneous handwritten forms. In: 2019 International Conference on Document Analysis and Recognition (ICDAR), pp. 487–493. IEEE (2019) 84. Toselli, A.H., Romero, V., Vidal, E.: Word graphs size impact on the performance of handwriting document applications. Neural Comput. Appl. 28(9), 2477–2487 (2017) 85. Tran, M.T., Trieu, L.Q., Tran, H.Q.: Document representation and classification with twitterbased document embedding, adversarial domain-adaptation, and query expansion. J. Heurist., 1–23 (2019) 5 Intelligent Document Processing in End-to-End RPA Contexts ... 131 86. Ubul, K., Tursun, G., Aysa, A., Impedovo, D., Pirlo, G., Yibulayin, T.: Script identification of multi-script documents: a survey. IEEE Access 5, 6546–6559 (2017) 87. Vinjit, B., Bhojak, M.K., Kumar, S., Chalak, G.: A review on handwritten character recognition methods and techniques. In: 2020 International Conference on Communication and Signal Processing (ICCSP), pp. 1224–1228. IEEE (2020) 88. Wang, J., Si, S., Hong, Z., Qu, X., Zhu, X., Xiao, J.: Case study of few-shot learning in text recognition models. In: International Conference on Web Information Systems Engineering, pp. 394–401. Springer (2021) 89. Würsch, M., Ingold, R., Liwicki, M.: Divaservices-a restful web service for document image analysis methods. Digit. Scholarship Human. 32(suppl_1), i150–i156 (2017) 90. Xiong, Z., Shen, Q., Wang, Y., Zhu, C.: Paragraph vector representation based on word to vector and CNN learning. Comput. Mater. Continua 55(2), 213–227 (2018) 91. Yamazaki, A., Sando, K., Suzuki, T., Aiba, A.: A handwritten Japanese historical kana reprint support system: Development of a graphical user interface. In: Proceedings of the ACM Symposium on Document Engineering 2018, pp. 1–4 (2018) 92. Yang, Y., Feng, Y., Ge, J., Zhou, Y., Zeng, J., Li, C., Luo, B.: Checking the statutes in Chinese judgment document based on editing distance algorithm. In: 2017 14th Web Information Systems and Applications Conference (WISA), pp. 197–200. IEEE (2017) 93. Zhao, M., Hochuli, A.G., Cheddad, A.: End-to-end approach for recognition of historical digit strings (2021). arXiv preprint arXiv:2104.13666 94. Zhu, X., Wang, J., Hong, Z., Xia, T., Xiao, J.: Federated learning of unsegmented chinese text recognition model. In: 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI), pp. 1341–1345. IEEE (2019)