On the extraction and use of arguments in recommender systems: A case study in the e-participation domain Andrés Segura-Tinoco1 , Iván Cantador2 1 Departamento de Ingeniería Informática, Universidad Autónoma de Madrid, Spain 2 Departamento de Ingeniería Informática, Universidad Autónoma de Madrid, Spain Abstract In this paper, we present ongoing work on the automatic extraction of arguments from textual content, and on the use of interconnected argument structures by recommender systems. Differently to the majority of existing argument mining methods –which only consider ‘premise’ and ‘claim’ as the components of an argument, and ‘support’ and ‘attack’ as the possible relations between argument components–, we propose an argumentation model based on a detailed taxonomy of argumentative relations. Moreover, we provide a lexicon of English and Spanish linguistic connectors categorized in our taxonomy. As a proof of concept, we apply a simple, yet effective method that makes use of the built taxonomy and lexicon to extract argument graphs from citizen proposals and debates of an e-participation platform. We then describe how the extracted graphs could be exploited to generate and explain argument-based recommendations. Keywords argument-based recommender systems, recommendation explanations, argument mining, natural language processing, e-government 1. Introduction statements and assertions, such as those given in social networking and microblogging services [10]. Since the origins of the recommender systems field, in Beyond the benefits of providing recommendations the mid-1990s [1], content-based recommendations have based on opinions, in certain cases, it would be use- received special attention not only to deal with cold-start ful to understand and consider the reasons (arguments) situations [2] and to complement collaborative filtering for given opinions [11, 12]. This would be valuable for techniques [3], but also to address domains character- both traditional recommendation domains, such as e- ized by textual content, such as books [4], scientific pub- commerce, leisure and tourism –-where specific websites lications [5], news articles [6], and online reviews [7]. are plenty of user reviews–-, and less common domains The data sources in these domains are heterogeneous that are rich in argumentative information [13]; in partic- in nature and form –ranging from well-defined cate- ular, web forums and electronic platforms for discussion gories and freely-chosen (social) tags to natural language and debate, and software tools that handle argumentative texts of different length, e.g., titles, summaries, and long content, e.g., legal corpora, educational text resources, descriptions–, and have distinct levels of linguistic for- transcripts of political speeches, and collections of citizen mality and explicit/implicit structure complexity. These proposals. text characteristics, as well as own particularities of nat- In all these domains, argumentative information would ural language (e.g., misspellings, ambiguity, irony) make not only be part of the recommendation explanations, the content-based recommendation a challenging task. but could also be exploited by the recommendation al- In this context, many research efforts have been de- gorithms. For such purpose, it is first necessary to voted to recommendation approaches aimed to exploit automatically identify and extract from text the ex- opinions expressed as natural language in unstructured, isting arguments. Then, it is desirable to represent free-form texts. The opinions can be detailed and fo- the extracted argumentative information in structured, cused on a particular item and its aspects, such as those computer-processable forms, which would allow inter- provided in blogs and reviews [8, 7, 9], or can be short connecting the arguments –e.g., through relationships in favor or against– and even to contrast them with objec- 3rd Edition of Knowledge-aware and Conversational Recommender tive (external) facts. Systems (KaRS) & 5th Edition of Recommendation in Complex Environments (ComplexRec) Joint Workshop @ RecSys 2021, Addressing these goals, in this paper, we present ongo- September 27–1 October 2021, Amsterdam, Netherlands ing work on the automatic extraction of arguments from Envelope-Open andres.segurat@estudiante.uam.es (A. Segura-Tinoco); textual content, and on the use of interconnected argu- ivan.cantador@uam.es (I. Cantador) ment structures by recommender systems. Differently to Orcid 0000-0001-6868-1445 (A. Segura-Tinoco); 0000-0001-6663-4231 methods existing in the argument mining field [14, 13], (I. Cantador) © 2021 Copyright for this paper by its authors. Use permitted under Creative which only consider ‘premise’ and ‘claim’ as the compo- CEUR Workshop Proceedings http://ceur-ws.org ISSN 1613-0073 Commons License Attribution 4.0 International (CC BY 4.0). CEUR Workshop Proceedings (CEUR-WS.org) nents of an argument, and ‘support’ and ‘attack’ (rebuttal) as the possible relations between argument components, The latter works present implementations and evalua- we propose an argumentation model based on a detailed tions of classic content-based, collaborative filtering and taxonomy of argumentative relations. The taxonomy is hybrid recommendation methods that exploit a variety then populated with a lexicon of linguistic connectors for of user-generated content, such as social tags and votes, both English and Spanish, and is preliminary exploited by as well as item (citizen proposal) metadata based on cate- simple, yet effective argument extraction and argument- gories, topics, and geographic locations. Differently to based recommendation methods. As a proof of concept, these works, in this paper, we advocate for recommender we report some results on generated arguments, recom- systems that dig into the semantics underlying the texts mendations, and recommendation explanations for the of the citizens’ proposals and comments. Hence, we aim e-participation domain, in which graphs of arguments to investigate recommendation approaches that exploit exist around citizen proposals and debates. the arguments provided by citizens, in favour or against the created proposals. 2. Related work 2.2. Argument-based recommender In this section, we describe some representative works of systems the topics addressed in our research. Specifically, we sur- vey recommender systems targeting the e-participation Surveying the academic literature on argument-based rec- domain (section 2.1) and exploiting argumentative infor- ommender systems, two main groups of researches can mation (section 2.2), and we provide major references be identified. The first group refers to recommendation on argument mining (section 2.3). methods that are based on Defeasible Logic Program- ming (DeLP) [27]. DeLP is a computational reasoning framework that consists of an argumentation engine op- 2.1. Recommender systems in erating over a knowledge base expressed in a logic pro- e-participation gramming language, which accepts encoded facts, and strict and defeasible rules (constraints). In the context As explained in [15], recommendation solutions are of in- of recommender systems, DeLP allows defining as rules creasing interest and application for numerous problems, user tastes and interests, item features and relations, and tasks and challenges of (smart) cities. In the paper, the contextual conditions [28, 29, 30]. Hence, the engine authors survey the academic literature on recommender reasons over a set of defined rules in order to infer po- systems for the principal six dimensions of smart cities, tential preferences of users for certain items, that is, to namely economy, environment, mobility, governance, provide lists of item recommendations in an argumenta- living and people. tive fashion. Presenting the set of rules applied (satisfied) With respect to the governance dimension, recom- in such process, DeLP enables the explanation of gener- mender systems have been mainly proposed to facilitate ated recommendations. It, however, requires building the the access to government information and increase effi- argument knowledge base, which to date has been done ciency in municipal management –e.g., by providing per- manually [31] or has been limited to simple, automatic sonalized suggestions of electronic government notifica- transformations of relational databases [32]. tions and services [16, 17, 18, 19, 20, 21]–, and to provide The second group is composed of approaches aimed government transparency and accountability, and pro- to provide argumentative explanations of recommenda- mote citizens’ participation and inclusion in public deci- tions, regardless the filtering algorithm used. In this case, sion making –e.g., by assisting voters through the presen- arguments mainly represent relationships between user tation of candidates with similar political views [22, 23]. preferences and item attributes. In [33], the authors pro- In [24], the authors discuss recommender systems for e- pose a framework where different types of justification governance, differentiating nine use cases in government- (e.g., ethical, aesthetic) are given for generated recom- to-citizen (G2C), government-to-business (G2B), and mendations depending on the users’s preferences and government-to-government (G2G) e-services. From according to manually defined rules. In [11], the authors them, we focus on the G2C case where users are assisted address the task of predicting the usefulness of review in finding relevant citizen proposals and debates gener- fragments according to their argumentative content. The ated in e-participation tools. In this case, among other estimated usefulness is used to rank the reviews asso- applications, recommender systems have been used as ciated to recommended items. Also focusing on user information filtering mechanisms for e-participatory bud- reviews, in [34], the authors propose to identify aspects geting (ePB) platforms [25, 26], where citizens propose important for the target user through an attention neural and debate online a large number (hundreds or even thou- network model, extract and summarize relevant argu- sands) of ideas, initiatives and projects aimed to address ments (opinions) about such aspects, and present the municipal issues. arguments as textual explanations of personalized rec- ommendations. A related approach is followed in [35], approaches. On the other hand, a variety of tools are where the authors propose a method that generates ex- available for different purposes, such as argumentative planations in an argumentative manner by presenting modeling (e.g., Agora,1 Argunet,2 DebateGraph3 and Ra- an incremental selection of positive and negative state- tionale Online4 ), and argument-based text annotation ments that support or contradict recommended items (e.g., Araucaria5 and OVA6 ). and their aspects, according to opinions expressed in In this paper, we i) preliminary experiment with a user reviews. Lastly, without taking user reviews into simple, yet effective syntactic pattern-based method to account, in [36], the authors exploit Linked Open Data argument extraction (addressing the three main AM tasks to extract descriptive properties about items, and use the explained above), and ii) provide new resources for the extracted properties to feed graph-based explanations of AM community; specifically, a detailed argument rela- recommended items. These explanations are generated tion taxonomy that goes beyond the premise-claim and through argumentative, natural language templates. support-attack models, and a lexicon of English and Span- For both groups, to the best of our knowledge, and ish linguistic connectors associated to the taxonomy cat- differently to our proposal, published argument-based egories. recommender systems do not make use of argument min- ing methods and resources to automatically extract argu- mentative information from textual content, and exploit 3. Case study such information during the item filtering process. In this section, we introduce Decide Madrid,7 an e- participation platform for which we have preliminary 2.3. Argument mining tested our argument mining and argument-based recom- Emerged from the confluence of the Computational Lin- mendation methods. guistics (CL) and Natural Language Processing (NLP) Among other citizen participation methods, Decide areas, Argument Mining (AM) [13] is a relatively young Madrid is an online website used by the Madrid City field that dates back to the late 2000s. In [37], it was for- Council for its annual participatory budgets. Since mulated with the general aim of automatically extracting September 2015, every year, city residents are allowed structured, argumentative information from text. to freely upload, comment and vote for proposals aimed This research challenge has been commonly mod- to address city problems and initiatives. A citizen pro- eled as a pipeline of three (consecutive) tasks: argu- posal is composed of the following data: title, description, ment detection [38, 39, 40], argument component identifi- author, date, tags, multimedia elements (i.e., pictures, cation [41, 42, 43], and argument relation recognition [40]. photos, videos, maps), comment threads, and supports Argument detection refers to the segmentation of a text (votes). into argumentative and non-argumentative units. Argu- Those proposals that receive a minimum number of ment component identification refers to the classification supports (around 22,000) are analyzed by experts in of argumentative units according to their role within order to check their feasibility. At the end of each the underlying arguments: ‘premise’ or ‘conclusion’, in yearly proposing period, the accepted, feasible propos- general. Lastly, argument relation recognition refers to als (around 300) receive funding and are implemented. the classification of the semantic relationships between Accessible as Open Data,8 every year, around 4,000 pro- pairs of argument fragments, such as ‘supporting’ and posals are created by city residents with the aim of re- ‘attacking.’ ceiving enough citizens’ supports and consequently the To date, these tasks have been mostly addressed sep- government’s approval. arately through machine learning methods [39, 42], but The large number of proposals, which also occurs in recently, they have been jointly treated as sequence la- e-participatory budgeting processes of other big cities belling tasks of NLP, addressed by specialized neural worldwide, has motivated the investigation of recom- network models [44]. In both cases, the desired, final mender systems to assist on the exploration of propos- outcome of the AM process is a tree or graph structure als [24, 26]. Published recommenders have exploited that semantically interconnects the arguments existing content-based (e.g., topics, categories) and collaborative in an input text. (e.g., supports/votes) data of the proposals. However, Additionally to algorithmic solutions, significant ad- 1 http://agora.gatech.edu vances have been made on the development of linguistic 2 https://sourceforge.net/projects/argunet resources. On the one hand, there are a number of cor- 3 https://debategraph.org pora annotated with structured argument information 4 https://www.rationaleonline.com 5 from different sources –such as persuasive essays, online http://staff.computing.dundee.ac.uk/creed/araucaria 6 http://ova.arg-tech.org debates, and news media items (cf. [13] for a detailed 7 https://decide.madrid.es survey)–, which can be used to build and evaluate AM 8 https://datos.madrid.es they have not considered the textual content of the pro- 4.2. Argument relation taxonomy and posals’ descriptions and comments. In the ongoing work lexicon presented in this paper, by contrast, we advocate for the use of such content, in particular, its underlying argu- As introduced in the previous section, the argument mentative information. model that we propose to follow aims to consider a vari- ety of relations that go beyond the support-attack schema. Surveying the academic literature, we find studies that 4. Argument mining framework have have presented distinct types of relations, and have compiled sets of linguistic connectors (or indicators) as- In this section, we present our framework to automati- sociated to such types. cally identify arguments in textual content, split them For instance, in [46], the authors provide an exhaus- into premise and claim components, and categorize the tive corpus of relational phrases, categorized in a taxon- relation between such components. The framework is omy based on discourse functions: expressing sequences built upon a well known argument model (section 4.1) (e.g., to start with, then, in addition), situating an event and novel argument relation taxonomy and lexicon (sec- in time (e.g., before, while, after) and space (e.g., where, tion 4.2). It is preliminary implemented through an argu- wherever), providing causal or purpose relations (e.g., so, ment extraction method based on simple syntactic rules in case, therefore), giving similarities (e.g., also, likewise, (section 4.3). correspondingly), showing contrast and choice (e.g., by contrast, although, whereas), and clarifying statements 4.1. Argument model (e.g., that is, for example, to sum up). In [40], the authors describe a number of rhetorical relations related to ar- The academic literature on argumentation and discourse gumentative explanation, given examples of sentences is extensive and multidisciplinary. In fact, the under- and connectors for each relation. More specifically, they standing and modeling of arguments are topics of human consider the following relations: justification, reformula- concern and thought in philosophy since the Ancient tion, elaboration by illustration (or enumeration), elabo- Greece [14]. ration by precision, elaboration via comparison, elabora- The Toulmin’s model [45] is one of the most popular tion via consequence, contrast, and concession. Lastly, argument models. It structures an argument into six com- in [47], the authors consider a total of 115 lexical indi- ponents: the claim (i.e., the conclusion of the argument), cators categorized as ‘forward’ (e.g., as a result, because, the ground (i.e., the premise, foundation or basis for the thus), ‘backward’ (e.g., additionally, besides, moreover), claim), the warrant (i.e., the reasoning that legitimizes ‘thesis’ (e.g., all in all, finally, in conclusion), and ‘rebut- the claim by showing the relevance of the ground), the tal’ (e.g., but, however, though) indicators. Regardless backing (i.e., the support for the warrant), the qualifier these taxonomies, one can find works (e.g., [42, 48, 49]) (i.e., the degree of certainty of the claim), and the rebuttal that also provide lists of connectors used as features of (i.e., an exception that may apply to the claim). machine learning models for AM tasks. In CL in general and in AM in particular, however, the Carefully revising and jointly considering all these majority of existing computational methods and tools references, we have developed a two-level taxonomy of to design, extract and share arguments follow simpler argument relations, and have gathered a relatively large argument models [13]. Specifically, most of them only set of linguistic connectors classified with the taxonomy. consider premises and claims as argumentative units, and The taxonomy and the set of connectors, referred as an support and attack (rebuttal) as argument relations. ‘argument relation lexicon,’ are made accessible online9 Our argument model extends this basic representa- in English and Spanish. tion as follows. First, as done in some works [13], in Table 1 shows the categories and subcategories of the addition to premises and claims, we also consider major proposed taxonomy, with their primary intents (i.e., sup- claims as fundamental argument units. They refer to port, attack, qualifier), and gives some examples of En- the principal, resultant parts of argumentative chains glish and Spanish connectors of each (sub)category. within a discourse. Hence, other claims (and premises) As it can be seen, our taxonomy includes the following relate or depend on major claims. Second, instead of types of argument (component) relations: narrowing the scope to support and attack relations, we take more fine-grained relation types into account, e.g., • Cause. This relation links an argument that re- by distinguishing whether an attack really represents flects the reason or condition for another argu- an opposition or, on the contrary, it suggests an alterna- ment. tive, a comparison or a concession for an argument. The considered argument relation types form a taxonomy, as 9 explained next. Developed taxonomy and lexicon, https://github.com/ argrecsys/connectors • Clarification. This relation introduces a conclu- people, organizations, places) of the sentence are also rec- sion, exemplification, restatement or summary of ognized to enrich the underlying arguments, since they an argument. could be considered to relate the different arguments, in • Consequence. This relation evidences an explana- addition to their topics. tion, goal or result of a previous argument. On the sentence, constituency parsing is finally con- • Contrast. This relation links attacking arguments, ducted to extract a parse tree that represents the syn- distinguishing between several types of attack: tactic structure (i.e., interconnected phrases) of the sen- giving alternatives, doing comparisons, making tence. This structure will be used to recursively group the concessions, and providing oppositions. phrases of the sentence. Within the built phrase groups, • Elaboration. This relation introduces an argument syntactic patterns –e.g., in the premise-connector-claim that provides details about another one. The de- form– will be searched, thus identifying the existing ar- tails can entail addition, precision or similarity guments. issues about the target argument. All these NLP tasks are performed using the Stanford CoreNLP [50] library, both for English and Spanish. The lexicon is composed of 248 English connectors and 384 Spanish connectors. As shown in the table, the 4.3.2. Identifying arguments English connectors are evenly distributed into the tax- onomy categories (with an average of 44.5 connectors This phase aims to automatically identify arguments in per category), except the contrast category, which is the sentences that have connectors, using the outputs gener- only one with (70) connectors whose primary intent is ated in the previous phase. ‘attack.’ Specifically, the grouped phrases (obtained from the constituency parsing process) are traversed from the bot- tom to the top of the sentence constituency tree, and 4.3. Argument extraction method are matched with predefined syntactic patterns. For the This method is a simple heuristic approach that aims moment, arguments are recognized as matches with any to automatically identify and extract arguments from of the following two patterns: textual content using basic syntactic patterns. It per- forms in a simple but effective way the three basic tasks [𝑐𝑙𝑎𝑖𝑚{𝑚𝑎𝑖𝑛_𝑣𝑒𝑟𝑏} + 𝑐𝑜𝑛𝑛𝑒𝑐𝑡𝑜𝑟 + 𝑝𝑟𝑒𝑚𝑖𝑠𝑒{𝑚𝑎𝑖𝑛_𝑣𝑒𝑟𝑏}] of argument mining, namely: argument detection (from formed by three grouped phrases. citizen proposals), argument component identification [𝑐𝑙𝑎𝑖𝑚{𝑚𝑎𝑖𝑛_𝑣𝑒𝑟𝑏} + [𝑐𝑜𝑛𝑛𝑒𝑐𝑡𝑜𝑟 + 𝑝𝑟𝑒𝑚𝑖𝑠𝑒{𝑚𝑎𝑖𝑛_𝑣𝑒𝑟𝑏}]] (i.e., claims and premises linked through a connector), formed by two grouped phrases. and argument relation recognition using the proposed tax- onomy and lexicon. For such purpose, the method is In the patterns, both the claim and the premise can divided into two (consecutive) phases, where the output contain the main verb of the sentence. The verb is first of the first phase serves as input for the second phase. searched within the claim (it is more likely to be found In particular, the phases are processing natural language here), and then within the premise. In the future, other and identifying arguments (and their relations). more complex syntactic patterns could be considered, since they are easy to integrate into our method. 4.3.1. Processing natural language Once one of the two aforementioned patterns is matched, the sentence is split into claim and premise In this phase, the source text –i.e., a citizen proposal according to and linked with the sentence connector (ex- description– is first split into sentences, where arguments isting in the lexicon). The identified argument structure is will be searched (isolatedly in this stage of our work). finally stored into a JSON data type along with: i) the con- Only those sentences that contain at least one of the nector and its argument relation category, sub-category connectors in the proposed lexicon are then taken into and primary intent, ii) the sentence lists of nouns, verbs account. and named entities, iii) the main verb of the argument, For a given sentence, part-of-speech (PoS) tags are ex- and iv) the identifier of the citizen proposal where the tracted, identifying the grammatical category (i.e., noun, argument was found. verb, adjective, adverb, etc.) of each word. In this process, Figure 1 shows an example in JSON format of an ar- the identified verbs are stored into a list, which will be gument extracted from a citizen proposal on a specific used to establish the main verb (action) of an argument, topic: public transportation. In this example, the premise and the nouns are stored in another list, which will be directly attacks the claim of the argument, in order to used to set the possible topics or aspects the argument support (by contrast) the major claim, extracted from the refers to. All this information could be exploited by a rec- citizen proposal title. ommendation method as well. The named entities (e.g., Table 1 Categories and subcategories of the proposed argument type taxonomy, and some categorized examples of English and Spanish argument connectors from the built lexicon. Words in brackets are optional. Primary English connectors Spanish connectors Category Subcategory intent Num. Examples Num. Examples if [ever/so], in case of/that, si [alguna vez/es así], en caso de/que Condition qualifier 34 on the condition [that], unless 35 con/bajo la condición de [que], a no ser que Cause because [of], due to, since porque, ya que, debido a [que], pues, Reason support 14 given that, based on, forasmuch as 21 dado que, basándose en [que], puesto que 48 56 para concluir, en/como conclusión, to conclude, in/as conclusion, Conclusion support 17 19 en definitiva, atendiendo a/con [todo] all in all, all things considered lo considerado for [example/instance], as an example [of] por ejemplo, como ejemplo [de], Exemplification support 9 like, such as, to take/give an example [of] 14 tales como, por dar/poner un ejemplo [de] Clarification in other words, that is [to say], en otras palabras, es decir, esto es, Restatement support 6 put differently, to put it another way 34 mejor dicho, dicho de otro modo summarizing, summing up, to sum up, resumiendo, concluyendo, para acabar, Summary support 14 in summary/short, in a few words 12 por resumir/concluir, en pocas palabras 46 79 actually, in [actual] fact, indeed, realmente, de hecho, en realidad, Explanation support 6 of course, for that matter 8 por supuesto, en efecto, para el caso for, to, in order to, aimed/aiming to, para, por, con el fin de, Goal support 19 that/which allows/entails/implies 18 lo que/cual permite/conlleva/implica Consequence therefore, thus, hence, then, so [that] por [lo] tanto, por consiguiente/ende Result support 21 as a result [of], this/that/such reason, 44 como resultado, por esta/esa razón, accordingly, in/as a consequence así que, es por ello que, de este/ese modo 46 70 on the other hand, in another case, por otra parte, por otro lado, en otro caso, Alternative support/attack 21 if not, instead [of], rather than, 29 si no, en vez/lugar de, en cambio/su defecto, alternatively [to], otherwise, else alternativamente [a], de otro modo while, whereas, compared [to/with], mientras [que], comparado con, Comparison support/attack 7 in comparison to/with, as long as 20 en comparación a/con, a la vez de/que Contrast although, [even] though, despite [that], aunque, aún/incluso [si/así], a pesar de/del, Concession support/attack 20 in spite/despite of, regardless [of] 38 a pesar de que, pese a [que], pese al but, however, nonetheless, albeit, pero, sin embargo, no obstante, Opposition attack 22 nevertheless, in contrast [to/with] 46 en contraste a/con, en contra [de/del] 70 133 also, besides, as well, too, moreover, también, además/aparte [de], [lo que] es más, Addition support 18 furthermore, additionally, in addition [to] 22 asímismo, encima de, adicionalmente [a] en particular, particularmente, especialmente, in particular, particularly, especially, Elaboration Precision support 11 13 principalmente, [más] especificamente/ mainly, [more] specifically/precisely precisamente similarmente/analogamente [a], como, al similarly/analogously [to], like, likewise, Similarity support 9 11 igual que, del mismo modo [que], de la misma in the same way, correspondingly manera [que] 38 46 248 384 To conclude, we present some statistics from a prelim- • Of the 1,379 arguments extracted (some proposals inary offline test (with a subset of lexicon connectors) on had more than one argument), 1,034 were identi- the automatic identification and extraction of arguments fied with connectors from the CAUSE category from the citizen proposals available in the Decide Madrid and 345 from the CONTRAST category. database: • An accuracy of 78.8% was achieved in a manual evaluation of 47 arguments about public trans- • From a reduced list of 10 connectors (belonging portation. to the CAUSE and CONTRAST categories), 1,744 proposals with possible arguments were identi- fied out of the 21,744 proposals available. • Arguments were automatically extracted in 1,362 of the 1,744 proposals identified, entailing a cov- erage of 78.0%. Figure 1: Example in JSON format of an argument extracted from a citizen proposal about public transportation. ”5717-1”: { ”proposalID”: 5717, ”sentence”: ”The use of public transport in the city is almost forced but in EMT pets are not allowed”, ”mainVerb”: ”is forced”, ”connector”: { ”value”: ”but”, ”intent”: ”attack”, ”category”: ”CONTRAST”, ”subCategory”: ”OPPOSITION” }, ”premise”: { ”entities”: ”[EMT]”, ”text”: ”in EMT pets are not allowed”, ”nouns”: ”[pets]” }, ”claim”: { ”entities”: ”[]”, ”text”: ”The use of public transport in the city is almost forced”, ”nouns”: ”[use, transport, city]” }, ”majorClaim”: { ”entities”: ”[]”, ”text”: ”Allowing pets on public transport”, ”nouns”: ”[pets, transport]” }, ”pattern”: ”P1 -> CLAIM + CONNECTOR + PREMISE” } 5. Argument-based for the target topic, and the arguments that support or attack these proposals grouped by topics and aspects. recommendations A contribution of our work is the proposal of this new Once the arguments are automatically identified and recommendation paradigm, which is based on the min- extracted from a set of citizen proposals, they can be ing of arguments, that is, instead of just recommending exploited as complex inputs of an argument-based rec- proposals that satisfy a user’s information needs (𝑢𝑠𝑒𝑟 → ommendation method. As a proof of concept, given a 𝑡𝑜𝑝𝑖𝑐𝑠 → 𝑝𝑟𝑜𝑝𝑜𝑠𝑎𝑙𝑠), we propose to recommend proposals particular topic –e.g., public transportation–, we consider that have arguments concerning the user’s topics of inter- a recommender that, via content-based filtering, first re- est (𝑢𝑠𝑒𝑟 → 𝑡𝑜𝑝𝑖𝑐𝑠/𝑎𝑠𝑝𝑒𝑐𝑡𝑠 → 𝑎𝑟𝑔𝑢𝑚𝑒𝑛𝑡𝑠 → 𝑝𝑟𝑜𝑝𝑜𝑠𝑎𝑙𝑠), in trieves and filters proposals about the topic, and then con- order to not only filter relevant information for the user, siders the arguments given in such proposals to rerank but also to assist her on decision making tasks. Moreover, and present recommended proposals (and arguments). the proposed approach allows creating in a direct and pre- More specifically, from the selected proposals and their cise way explanations of the generated argument-based associated arguments, the recommender identifies the recommendations. The following are possible explana- discussed aspects of the topic of interest (e.g., price, lo- tion templates: cation, quantity) for which there are arguments in favor • “[These] citizen proposals about [this] topic are or against. With these aspects, the recommender builds recommended because they have the following a graph that relates proposals, topics, aspects and ar- supporting (attacking) arguments...” guments, and exploits such graph to find relevant (i.e., • “Regarding [these] aspects on [this] topic of in- highly connected) proposals which are recommended to terest, the following proposals are recommended the user. since they have more arguments in favor” These proposals are presented along with their respec- tive arguments in the form, claim-connector-premise for We believe that these types of recommendations and each aspect. Figure 2 shows a subset of recommended explanations not only may help improving the effective- proposals about public transport in the context of the ness of the system, but also may increase its transparency Decide Madrid e-participation platform. The output of and foster the user’s trust. the argument-based recommender is an XML file which is composed of two blocks: the recommended proposals Figure 2: Example in XML format of recommendations of citizen proposals and arguments about public transportation. Urban buses connecting San Chinarro and Las Tablas with Cuatro Caminos Public transportation in Madrid Río Allowing pets on public transport Public transport price The Transport Pass should expire in one month The PAU of Norte Sanchinarro Las Tablas are poorly served by public transport due to the ineffectiveness of light subway The Madrid Rio park was created promising that public transport would reach there but it is false, the Legazpi subway is far away and buses are non-existent The use of public transport in the city is almost forced but in EMT pets are not allowed Lower the price of transportation because it is very expensive The Madrid Transport Pass expires in 30 days but not all months have 30 days, there are several months that have 31 days 6. Conclusions and future work beyond the commonly adopted support-attack schema, and a rich lexicon of argument connectors for both En- The ongoing work presented in this paper has resulted in glish and Spanish. The use of these resources has been a novel taxonomy of argumentative relations that goes preliminary exemplified through the automatic extrac- tion of arguments from text contents, and the genera- and Development in Information Retrieval, ACM, tion of argument-based recommendations in a real e- 2002, pp. 253–260. participation case study, where graphs of interconnected [3] P. Lops, M. De Gemmis, G. Semeraro, Content- topics, premises and claims underlay citizen proposals based recommender systems: State of the art and and debates. trends, Recommender Systems Handbook (1st edi- We believe that this new paradigm of argument-based tion) (2011) 73–105. recommendation, which provides transparency in the [4] H. Alharthi, D. Inkpen, S. Szpakowicz, A survey of form of intuitive, justified explanations, can not only book recommender systems, Journal of Intelligent be applicable to other e-government contexts –such as Information Systems 51 (2018) 139–160. parliamentary debate [51] and political discussion in so- [5] J. Beel, B. Gipp, S. Langer, C. Breitinger, Paper cial networks [52]–, but also to other domains rich in recommender systems: A literature survey, In- argumentative information, such as law, education, and ternational Journal on Digital Libraries 17 (2016) e-commerce. 305–338. There are, however, many research lines that should be [6] M. Karimi, D. Jannach, M. Jugovac, News recom- addressed before. First, we have to conduct more sophis- mender systems – survey and roads ahead, In- ticated text processing, e.g., by correcting misspellings, formation Processing & Management 54 (2018) dealing with lexical and syntactic variations, and bet- 1203–1227. ter identifying named entities. We then have to extend [7] L. Chen, G. Chen, F. Wang, Recommender systems our extraction method with additional syntactic argu- based on user reviews: The state of the art, User ment patterns, and other argument features different Modeling and User-Adapted Interaction 25 (2015) to linguistic connectors, as done by other methods in 99–154. the argument mining field [14, 13]. We also have to [8] K. Bauman, B. Liu, A. Tuzhilin, Aspect based recom- formally define recommendation methods that exploit mendations: Recommending items with the most argument structures in both the filtering and explanation valuable aspects based on user reviews, in: Pro- phases. These methods could be empirically compared ceedings of the 23rd ACM SIGKDD International with existing argument-based recommenders, e.g., based Conference on Knowledge Discovery and Data Min- on DeLP frameworks [27, 28], since they could be built ing, ACM, 2017, pp. 717–725. on knowledge bases generated with argument extrac- [9] M. Hernández-Rubio, I. Cantador, A. Bellogín, A tion methods. In this context, we should explore and comparative analysis of recommender systems evaluate recommendation explanations in a more natural based on item aspect opinions extracted from user language form [36]. For such purpose, we expect that reviews, User Modeling and User-Adapted Interac- user studies will be needed. In these studies, we may also tion 29 (2019) 381–441. consider evaluating potential benefits of argument-based [10] J. Sirrianni, X. Liu, M. M. Rahman, D. Adams, An recommendations, such as transparency, fairness, and opinion diversity enhanced social connection rec- accountability [53]. These issues are of special interest ommendation re-ranking method based on opinion in e-participation contexts such as the one addressed in distance in cyber argumentation with social net- this work, where, among others, controversy topics and working, in: Proceedings of the 2019 IEEE Interna- minority groups are of high relevance [54]. tional Conference on Cognitive Computing, IEEE, 2019, pp. 106–113. [11] M. Passon, M. Lippi, G. Serra, C. Tasso, Predicting Acknowledgments the usefulness of Amazon reviews using off-the- shelf argumentation mining, in: Proceedings of the This work was supported by the Spanish Ministry of 5th Workshop on Argument Mining, ACL, 2018, pp. Science and Innovation (PID2019-108965GB-I00). 35–39. [12] A. Rago, O. Cocarascu, F. Toni, Argumentation- References based recommendations: Fantastic explanations and how to find them, in: Proceedings of the 27th [1] D. Billsus, M. J. Pazzani, A personal news agent International Joint Conference on Artificial Intelli- that talks, learns and explains, in: Proceedings of gence, IJCAI, 2018, pp. 1949–1955. the 3rd Annual Conference on Autonomous Agents, [13] J. Lawrence, C. Reed, Argument mining: A survey, ACM, 1999, pp. 268–275. Computational Linguistics 45 (2020) 765–818. [2] A. I. Schein, A. Popescul, L. H. Ungar, D. M. Pen- [14] M. Janier, P. Saint-Dizier, Argument mining: Lin- nock, Methods and metrics for cold-start recom- guistic foundations, John Wiley & Sons, 2019. mendations, in: Proceedings of the 25th Annual [15] L. Quijano-Sánchez, I. Cantador, M. E. Cortés- International ACM SIGIR Conference on Research Cediel, O. Gil, Recommender systems for smart cities, Information Systems 92 (2020) 101545. H. Alani, What’s going on in my city? Recom- [16] M. Eirinaki, S. Dhar, S. Mathur, A. Kaley, A. Patel, mender systems and electronic participatory bud- A. Joshi, D. Shah, A building permit system for geting, in: Proceedings of the 12th ACM Confer- smart cities: A cloud-based framework, Computers, ence on Recommender Systems, 2018, pp. 219–223. Environment and Urban Systems 70 (2018) 175–188. [27] A. J. García, G. R. Simari, Defeasible logic program- [17] M. Janssen, R. Wagenaar, J. Beerens, Towards ming: An argumentative approach, Theory and a flexible ICT-architecture for multi-channel e- Practice of Logic Programming 4 (2004) 95–138. government service provisioning, in: Proceedings [28] C. I. Chesñevar, A. G. Maguitman, ArgueNet: An of the 36th Annual Hawaii International Confer- argument-based recommender system for solving ence on System Sciences, IEEE, 2003, pp. 10–pp. web search queries, in: Proceedings of the 2nd In- [18] J. D. Martín-Guerrero, A. Palomares, E. Balaguer- ternational IEEE Conference on Intelligent Systems, Ballester, E. Soria-Olivas, J. Gómez-Sanchis, volume 1, IEEE, 2004, pp. 282–287. A. Soriano-Asensi, Studying the feasibility of a [29] C. I. Chesñevar, A. G. Maguitman, M. P. González, recommender in a citizen web portal based on user Empowering recommendation technologies modeling and clustering algorithms, Expert Sys- through argumentation, in: Argumentation in tems with Applications 30 (2006) 299–312. Artificial Intelligence, Springer, 2009, pp. 403–422. [19] C. Baldassarre, M. Cremaschi, M. Palmonari, Bridg- [30] C. E. Briguez, M. Capobianco, A. G. Maguitman, ing the gap between citizens and local administra- A theoretical framework for trust-based news rec- tions with knowledge-based service bundle recom- ommender systems and its implementation using mendations, in: Proc. of the 24th International defeasible argumentation, International Journal on Workshop on Database and Expert Systems Appli- Artificial Intelligence Tools 22 (2013) 1350021. cations, IEEE, 2013, pp. 157–161. [31] C. I. Chesñevar, A. G. Maguitman, G. R. Simari, [20] R. Ayachi, I. Boukhris, S. Mellouli, N. B. Amor, Recommender system technologies based on argu- Z. Elouedi, Proactive and reactive e-government mentation, in: Proceedings of the 2007 Conference services recommendation, Universal Access in the on Emerging Artificial Intelligence Applications in Information Society 15 (2016) 681–697. Computer Engineering, IOS Press, 2007, pp. 50–73. [21] L. Á. Sabucedo, R. S. Barreiros, J. M. S. Gago, M. F. [32] C. E. Briguez, M. C. D. Budán, C. A. D. Deagus- Iglesias, A hybrid semantic driven recommender for tini, A. G. Maguitman, M. Capobianco, G. R. Simari, services in the egovernment domain, in: Proceed- Argument-based mixed recommenders and their ings of the 2nd International Conference on Digital application to movie suggestion, Expert Systems Information and Communication Technology and with Applications 41 (2014) 6467–6482. its Applications, IEEE, 2012, pp. 409–414. [33] F. Yetim, A framework for organizing justifications [22] L. Terán, A. Meier, A fuzzy recommender system for for strategic use in adaptive interaction contexts, in: eElections, in: Proceedings of the 1st International Proceedings of the 16th European Conference on Conference on Electronic Government and the In- Information Systems, AIS Electronic Library, 2008, formation Systems Perspective, Springer, 2010, pp. pp. 815–825. 62–76. [34] T. Donkers, B. Loepp, J. Ziegler, Explaining rec- [23] K. Dyczkowski, A. Stachowiak, A recommender ommendations by means of user reviews, in: Joint system with uncertainty on the example of political Proceedings of the ACM IUI 2018 Workshops, ACM, elections, in: Proceedings of the 14th International 2018. Conference on Information Processing and Manage- [35] S. Naveed, T. Donkers, J. Ziegler, Argumentation- ment of Uncertainty in Knowledge-Based Systems, based explanations in recommender systems: Con- Springer, 2012, pp. 441–449. ceptual framework and empirical results, in: Ad- [24] M. E. Cortés-Cediel, I. Cantador, O. Gil, Recom- junct Publication of the 26th Conference on User mender systems for e-governance in smart cities: Modeling, Adaptation and Personalization, ACM, State of the art and research opportunities, in: Pro- 2018, pp. 293–298. ceedings of the 1st International Workshop on Rec- [36] C. Musto, F. Narducci, P. Lops, M. de Gemmis, G. Se- ommender Systems for Citizens, 2017, pp. 1–6. meraro, Linked open data-based explanations for [25] I. Cantador, A. Bellogín, M. E. Cortés-Cediel, O. Gil, transparent recommender systems, International Personalized recommendations in e-participation: Journal of Human-Computer Studies 121 (2019) Offline experiments for the ’Decide Madrid’ plat- 93–107. form, in: Proceedings of the International Work- [37] R. Mochales Palau, M.-F. Moens, Argumentation shop on Recommender Systems for Citizens, 2017, mining, Artificial Intelligence and Law 19 (2011) pp. 1–6. 1–22. [26] I. Cantador, M. E. Cortés-Cediel, M. Fernández, [38] B. Hachey, C. Grover, Automatic legal text summari- sation: Experiments with summary structuring, in: 52nd Annual Meeting of the Association for Compu- Proceedings of the 10th International Conference tational Linguistics: System Demonstrations, 2014, on Artificial intelligence and Law, Springer, 2005, pp. 55–60. pp. 75–84. [51] I. Cantador, L. Q. Sánchez, Semantic annotation and [39] M.-F. Moens, E. Boiy, R. Mochales Palau, C. Reed, retrieval of parliamentary content: A case study on Automatic detection of arguments in legal texts, in: the Spanish Congress of Deputies, in: Proceed- Proceedings of the 11th International Conference ings of the 1st Joint Conference of the Information on Artificial Intelligence and Law, ACM, 2007, pp. Retrieval Communities in Europe, 2020. 225–230. [52] J. L. Lavado, I. Cantador, M. E. Cortés-Cediel, M. Fer- [40] M. P. García Villalba, P. Saint-Dizier, Some facets of nández, Automatic intent-based classification of argument mining for opinion analysis, in: Proceed- citizen-to-government tweets, in: Proceedings of ings of the 4th International Conference on Compu- the 19th IFIP WG 8.5 International Conference on tational Models of Argument, Liverpool University Electronic Government, 2021. Press, 2012, pp. 23–34. [53] D. Shin, Y. J. Park, Role of fairness, accountability, [41] N. Kwon, L. Zhou, E. Hovy, S. W. Shulman, Identi- and transparency in algorithmic affordance, Com- fying and classifying subjective claims, in: Proceed- puters in Human Behavior 98 (2019) 277–284. ings of the 8th Annual International Conference [54] I. Cantador, M. E. Cortés-Cediel, M. Fernández, Ex- on Digital Government Research, ACM, 2007, pp. ploiting Open Data to analyze discussion and con- 76–81. troversy in online citizen participation, Information [42] R. Mochales Palau, M.-F. Moens, Argumentation Processing & Management 57 (2020) 102301. mining: The detection, classification and structure of arguments in text, in: Proceedings of the 12th International Conference on Artificial Intelligence and Law, ACM, 2009, pp. 98–107. [43] R. Rinott, L. Dankin, C. Alzate, M. M. Khapra, E. Aharoni, N. Slonim, Show me your evidence - An automatic method for context dependent evi- dence detection, in: Proceedings of the 2015 Con- ference on Empirical Methods in Natural Language Processing, ACL, 2015, pp. 440–450. [44] S. Eger, J. Daxenberger, I. Gurevych, Neural end- to-end learning for computational argumentation mining, in: Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, ACL, 2017, pp. 11–22. [45] S. E. Toulmin, The uses of argument, Cambridge University Press, 2003. [46] A. Knott, R. Dale, Using linguistic phenomena to motivate a set of coherence relations, Discourse Processes 18 (1994) 35–62. [47] C. Stab, I. Gurevych, Parsing argumentation struc- tures in persuasive essays, Computational Linguis- tics 43 (2017) 619–659. [48] C. Stab, I. Gurevych, Identifying argumentative discourse structures in persuasive essays, in: Pro- ceedings of the 2014 Conference on Empirical Meth- ods in Natural Language Processing, ACL, 2014, pp. 46–56. [49] V. Wei Feng, G. Hirst, Classifying arguments by scheme, in: Proceedings of the 49th Annual Meet- ing of the Association for Computational Linguis- tics, ACL, 2011, pp. 987–996. [50] C. D. Manning, M. Surdeanu, J. Bauer, J. R. Finkel, S. Bethard, D. McClosky, The stanford corenlp nat- ural language processing toolkit, in: Proceedings of