
Ebook: Artificial Intelligence Research and Development

Artificial intelligence has become an indispensible part of our lives in recent years, affecting all aspects from business and leisure to transport and health care. This book presents the proceedings of the 23rd edition of the International Conference of the Catalan Association for Artificial Intelligence (CCIA), an annual event that serves as a meeting point for researchers in Artificial Intelligence in the area of the Catalan speaking territories and from around the world. The 2021 edition was held online as a virtual conference from 20 - 22 October 2021 due to the COVID-19 pandemic. The book contains 42 long papers and 9 short papers, carefully reviewed and selected. The papers cover all aspects of artificial intelligence and are divided under six section headings: combinatorial problem solving and logics for artificial intelligence; sentiment analysis and tekst analysis; data science and decision support systems; machine learning; computer vision; and explainability and argumentation. Abstracts of the 2 invited talks delivered at the conference by Prof. Patty Kostkova and Prof. João Marques-Silva are also included.
Offering a state of the art overview of the subject from a regional perspective, the book will be of interest to all those working in the field of artificial intelligence.
The International Conference of the Catalan Association for Artificial Intelligence (CCIA) is an annual event that serves as a meeting point for researchers in Artificial Intelligence based in the area of the Catalan speaking territories (south France, Catalonia, Valencia, Balearic Islands and Alghero in Italy) and from around the world.
This book constitutes the proceedings of the 23rd edition of the CCIA, held in Lleida, in October 2021. The conference was held virtually due to the COVID-19 pandemic. Previous editions of the CCIA have been in Tarragona (1998), Girona (1999), Vilanova i la Geltrú (2000), Barcelona (2001, 2004, 2014, 2016), Castelló de la Plana (2002), Mallorca (2003), L’Alguer (Italy) (2005), Perpinyà (France) (2006), Andorra (2007), Sant Martí d’Empúries (2008), Cardona (2009), L’Espluga de Francolí (2010), Lleida (2011), Alacant (2012), Vic (2013), València (2015), Deltebre (2017), Roses (2018) and Colònia de Sant Jordi (2019). There was no CCIA in 2020 because of the severe restrictions caused by the COVID-19 pandemic.
The 42 long papers and 9 short papers presented in this volume were carefully reviewed and selected from 59 submissions. The reviewing process was made possible thanks to the 93 artificial intelligence experts who make up the program committee, plus some additional referees. We would especially like to thank them for their efforts in this task, as well as thanking the authors of the 59 submissions for their work.
The accepted papers deal with all aspects of artificial intelligence including: Combinatorial Problem Solving and Logics for Artificial Intelligence, Sentiment Analysis and Text Analysis, Data Science and Decision Support Systems, Machine Learning, Computer Vision, and Explainability and Argumentation. This book of proceedings also includes abstracts of the two invited talks, given by Prof. Patty Kostkova and Prof. João Marques-Silva.
We want to express our sincere gratitude to the Catalan Association for Artificial Intelligence (ACIA), the Research Group in Energy and Artificial Intelligence (GREiA), the Universitat de Lleida (UdL), the Universitat de Girona (UdG) and the Universitat Rovira i Virgili (URV) for their support.
Universitat de Lleida (Lleida), October 2021
Mateu Villaret, Universitat de Girona
Aïda Valls, Universitat Rovira i Virgili
Teresa Alsinet, Universitat de Lleida
Cèsar Fernández, Universitat de Lleida
In this paper, we introduce a framework for probabilistic logic-based argumentation inspired on the DeLP formalism and an extensive use of conditional probability. We define probabilistic arguments built from possibly inconsistent probabilistic knowledge bases and study the notions of attack, defeat and preference between these arguments. Finally, we discuss consistency properties of admissible extensions of the Dung’s abstract argumentation graphs obtained from sets of probabilistic arguments and the attack relations between them.
Construct, Merge, Solve & Adapt (CMSA) is a recently developed algorithm for solving combinatorial optimization problems. It combines heuristic elements, such as the probabilistic generation of solutions, with an exact solver that is iteratively applied to sub-instances of the tackled problem instance. In this paper, we present the application of CMSA to an NP-hard problem from the family of dominating set problems in undirected graphs. More specifically, the application in this paper concerns the minimum positive influence dominating set problem, which has applications in social networks. The obtained results show that CMSA outperforms the current state-of-the-art metaheuristics from the literature. Moreover, when instances of small and medium size are concerned CMSA finds many of the optimal solutions provided by CPLEX, while it clearly outperforms CPLEX in the context of the four largest, respectively more complicated, problem instances.
We tackle the problem of solving MinSAT for multisets of propositional formulas that are not necessarily in clausal form. Our approach reduces non-clausal to clausal MinSAT, since this allows us to rely on the much developed clause-based MinSAT solvers. The main contribution of this paper is the definition of several transformations of multisets of propositional formulas into multisets of clauses so that the maximum number of unsatisfied clauses in both multisets is preserved.
Understanding different perceptions of human being when using linguistic terms is a crucial issue in human-machine interaction. In this paper, we propose the concept of perceptual maps to model human opinions in a group decision-making context. The proposed approach considers a multi-granular structure using unbalanced hesitant linguistic term sets. An illustrative case is presented in the location decisions made by multinationals enterprises of the energy sector within the European smart city context.
The remarkable advances in SAT solving achieved in the last years have allowed to use this technology in many real-world applications of Artificial Intelligence, such as planning, formal verification, and scheduling, among others. Interestingly, these industrial SAT problems are commonly believed to be easier than classical random SAT formulas, but estimating their actual hardness is still a very challenging question, which in some cases even requires to solve them. In this context, realistic pseudo-industrial random SAT generators have emerged with the aim of reproducing the main features shared by the majority of these application problems. The study of these models may help to better understand the success of those SAT solving techniques and possibly improve them.
In this work, we present a model to estimate the temperature of real-world SAT instances. This temperature represents the degree of distortion into the expected structure of the formula, from highly structured benchmarks (more similar to real-world SAT instances) to the complete absence of structure (observed in the classical random SAT model). Our solution is based on the Popularity-Similarity (PS) random model for SAT, which has been recently presented to reproduce two crucial features of application SAT benchmarks: scale-free and community structures. The PS model is able to control the hardness of the generated formula by introducing some randomizations in the expected structure. Our solution is a first step towards a hardness oracle based on the temperature of SAT formulas, which may be able to estimate the cost of solving real-world SAT instances without solving them.
In this paper we analyze the effect of selecting the root in a tree decomposition when using decomposition-based backtracking algorithms. We focus on optimization tasks for Graphical Models using the BTD algorithm. We show that the choice of the root typically has a dramatic effect in the solving performance. Then we investigate different simple measures to predict near optimal roots. Our study shows that correlations are often low, so the automatic selection of a near optimal root will require more sophisticated techniques.
State-space planning is the de-facto search method of the automated planning community. Planning problems are typically expressed in the Planning Domain Definition Language (PDDL), where action and variable templates describe the sets of actions and variables that occur in the problem. Typically, a planner begins by generating the full set of instantiations of these templates, which in turn are used to derive useful heuristics that guide the search. Thanks to this success, there has been limited research in other directions.
We explore a different approach, keeping the compact representation by directly reformulating the problem in PDDL into ESSENCE PRIME, a Constraint Programming language with support for distinct solving technologies including SAT and SMT. In particular, we explore two different encodings from PDDL to ESSENCE PRIME, how they represent action parameters, and their performance. The encodings are able to maintain the compactness of the PDDL representation, and while they differ slightly, they perform quite differently on various instances from the International Planning Competition.
In this paper, we present a model of the sense-making process for diagrams, and describe it for the case of Hasse diagrams. Sense-making is modeled as the construction of networks of conceptual blends among image schemas and the diagram’s geometric configuration. As a case study, we specify four image schemas and the geometric configuration of a Hasse diagram, with typed FOL theories. In addition, for the diagram geometry, we utilise Qualitative Spatial Reasoning formalisms. Using an algebraic specification language, we can compute conceptual blends as category-theoretic colimits. Our model approaches sense-making as a process where the image schemas and the diagram geometry both structure each other through a complex network of conceptual blends. This yields a final blend in which the sort of inferences we confer to diagrammatic representations emerge. We argue that this approach to sense-making in diagrams is more cognitively apt than the mainstream view of a diagram being a syntactic representation of some underlying logical semantics. Moreover, our model could be applied to various types of stimuli and is thus valuable for the general field of AI.
The law of importation has attracted the interest of many researchers devoted to fuzzy implication functions in the last decades. This property has several important applications, especially in approximate reasoning and image processing. Several generalizations of this property have been proposed. Specifically, one generalization related to the law of migrativity was recently introduced by Baczyński et al. in which two fuzzy implication functions are involved. In this paper, some advances on the solution of this functional equation for the particular case where the involved fuzzy conjunction is a t-norm are presented. Indeed, a complete characterization of all those pairs of fuzzy implication functions with a strict natural fuzzy negation satisfying the generalized law of importation is achieved.
In the field of normative multiagent systems, the relationship between a game structure and its underpinning agent interaction rules is hardly ever addressed in a systematic manner. In this work, we introduce the Action Situation Language (ASL), inspired by Elinor Ostrom’s Institutional Analysis and Development framework, to bridge the gap between games and rules. The ASL provides a syntax for the description of agent interactions, and is complemented by an engine that automatically provides semantics for them as extensive-form games. The resulting games can then be analysed using standard game-theoretical solution concepts, hence allowing any community of agents to automatically perform what-if analysis of potential new interaction rules.
Hate speech expresses prejudice and discrimination based on personal characteristics such as race or gender. Research has proven that the amount of hateful messages increases on online social media. If not countered properly, the spread of hatred can overwhelm entire societies. This paper proposes a multi-agent model of the spread of hatred. We reuse insights from previous research to construct and validate a baseline model. From this, three countermeasures are modelled and simulated to investigate their effectiveness in containing the spread of hatred: (1) The long-term measure of education is very successful, but it still cannot eliminate hatred completely; (2) Deferring hateful content has a similar positive effect with the advantage of being a short-term countermeasure; (3) Extreme cyber activism against hatred can worsen the situation and even increase the likelihood of high polarisation within societies.
Reddit is a social news aggregation and discussion website. Users submit content to the site such as links to news, which are then voted up or down by other members who in turn, can comment on others’ posts to continue the conversation. In this work, we are interested in modeling how users interact with each other in Reddit debates, to discover the most dominant opinions in a debate. To this end, we introduce a user-based model for analysis of Reddit debates. In this model, comments by users are grouped per user, describing their opinion in relation to the root comment of the debate, and users are represented with a single node in a weighted graph, where node’s weights represent relevance of user’s opinions and edges represent agreement or disagreement relationships between users throughout the debate. In this model, agreement or disagreement between the opinions of two users is defined by aggregating the set of single interactions that have occurred between them during the debate. In this work we present a skeptical aggregation model for this task. For measuring the relevance of user’s opinions, we consider two models: one based on the score of all the user’s comments and other based on the user’s karma, as computed by the Reddit platform. We characterize the set of most dominant opinions with an argumentative-based model, using the information of disagreement between opinions and relevance of opinions.
The automated analysis of different trends in online debating forums is an interesting tool for sampling the agreement between citizens in different topics. In these online debating forums, users post different comments and answers to previous comments of other users. In previous work, we have defined computational models to measure different values in these online debating forums. A main ingredient in these models has been the identification of the set of winning posts trough an argumentation problem that characterizes this winning set trough a particular argumentation acceptance semantics. In the argumentation problem we first associate the online debate to analyze as a debate tree. Then, comments are divided in two groups, the ones that agree with the root comment of the debate, and the ones that disagree with it, and we extract a bipartite graph where the unique edges are the disagree edges between comments of the two different groups. Once we compute the set of winning posts, we compute the different measures we are interested to get from the debate, as functions defined over the bipartite graph and the set of winning posts. In this work, we propose to explore the use of graph neural networks to solve the problem of computing these measures, using as input the debate tree, instead of our previous argumentation reasoning system that works with the bipartite graph. We focus on the particular online debate forum Reddit, and on the computation of a measure of the polarization in the debate. Our results over a set of Reddit debates, show that graph neural networks can be used with them to compute the polarization measure with an acceptable error, even if the number of layers of the network is bounded by a constant.
This paper presents an ongoing project about the implementation of digital twins (DT) for simulating cognitive-affective behaviours in social networks. Our approach relies on a pure data-driven solution, which takes existing public data from social networks to learn cognitive models according to the profile, posts and interactions of the social network users. The final aim is that the learned cognitive models can be parameterised according to existing classifications of traits and emotions so that different behaviours can be eventually simulated with the resulting DTs. In this work, we propose the use of the Transformers neural-network architectures to both interpret incoming messages according to cognitive contexts, and to generate responses to these messages. The first experiments are aimed at integrating and measuring existing approaches for emotion recognition from texts.
Topic modelling is nowadays one of the most popular techniques used to extract knowledge from texts. There are several families of methods related to this problem, among them 1) Factorial methods, 2) Probabilistic methods and 3) Natural Language Processing methods. In this paper a common conceptual framework is provided for Factorial and probabilistic methods by identifying common elements and describing them with common and homogeneous notation and 7 different methods are described accordingly. Under a common notation it is easy to make a comparative analysis and see how flexible or more or less realistic assumptions are made by the different methods. This is the first step to a wider analysis where all families can be related to this common conceptual framework and to go in depth in the understanding of stengths and weakenesses of each method and ellaboration of general guidelines to provide application criteria. The paper ends with a discussion comparing the presented methods and future research lines.
Research suggests that colour plays an important role in creating wellness emotions in hotel customers. This paper considers that tourists’ needs for wellness may be satisfied by manipulating existing elements of a hotel, such as the colour of a hotel room. The paper studies the relationship between tourists’ emotions and the main colour of a hotel room, and also the relationship between that emotion and their intention to stay in the hotel, and even the price that the tourists are willing to pay. Also, the paper studies the role of cultural differences in these relationships, specifically between Spanish and Equatorian tourists.
The business environment today is characterized by high competition and saturated markets. Pay-tv platforms there are not an exception. Because of that, the cost to acquire new customers is much higher than the cost of retaining the existing customers. Therefore, it is important for Pay-TV platforms to keep controlled the Customer Churn. Therefore, the paper studies existing models used to predict Customer Churn in other context -like telecommunication companies customer Churn-and adapts them to the Pay-TV context. Another big problem faced in the paper is the fact that, in the data set udes in the paper there are not personal metrics, which are indispensables to solve the problem. Therefore this approach has defined new metrics in order to be able to predict customer churn.
Financial networks represent the daily business interactions of customers and suppliers. Research in this domain has mainly focused on characterizing different network structures and studying dynamical processes over them. These two aspects, structure and dynamics, play a key role in understanding how emergent collective behaviors, such as those that arise during economic crises, propagate through networks. Business interactions between companies form a direct and weighted network, where the financial distress of a node depends on the ability of its customers to fulfill payments. In situations where there is no such inbound cash flow, a company may have to close down due to a lack of liquidity. Interconnection therefore seems to be at the core of systemic fragility. Whether the nature and form of this connection may have an impact on how distress is propagated is still an open question. In this paper, we study how disruptive events propagate through different network structures, under different scenarios. For this purpose, we use a liquidity model that describes how the economy of nodes evolves from a given initial state in terms of their interactions. From our experiments, we empirically conclude that most of the studied network dynamics reach a steady-state, even in the presence of large noise values.