No description available.

Dans cette thèse, nous étudions la sémantique des relations entre les utilisateurs et des forces antagonistes que nous observons naturellement dans diverses relations sociales, comme hostilité ou méfiance. L'étude de ces relations soulève de nombreux problèmes à la fois techniques, puisque l'arsenal mathématique n'est souvent pas adapté aux liens négatifs, mais aussi pratiques à cause de la difficulté rencontrée pour collecter de telles données (expliciter une relation négative est perçu comme malvenu pour de nombreux utilisateurs). Nous nous intéressons alors aux solutions alternatives de collecte afin d'inférer ces relations négatives à partir d'autres contenus. En particulier, nous allons utiliser les jugements communs que les utilisateurs partagent à propos d'items (données des systèmes de recommandation). Nous apportons trois contributions. Dans la première, nous allons aborder le cas des accords sur les items qui peuvent ne pas avoir la même sémantique selon qu'ils concernent des items appréciés ou non par les utilisateurs. Nous verrons que le fait de ne pas aimer un même produit n'est pas synonyme de similarité. Ensuite, nous allons prendre en compte dans notre seconde contribution les distributions de notes des utilisateurs et des items afin de mesurer si les accords ou les désaccords arrivent par hasard ou non, afin notamment d'éviter les conséquences des différents biais utilisateurs et items présents dans ce type de données. Enfin, notre troisième contribution consistera à exploiter ces différents résultats afin de prédire le signe des liens entre utilisateurs à partir des seuls jugements communs à propos des items et sans aucune information sociale négative. / In this thesis, we study the semantic of relations between users and, in particular, the antagonistic forces we naturally observe in various social relationships, such as hostility or suspicion. The study of these relationships raises many problems both techniques - because the mathematical arsenal is not really adapted to the negative ties - and practical, due to the difficulty of collecting such data (explaining a negative relationship is perceived as intrusive and inappropriate for many users). That’s why we focus on the alternative solutions consisting in inferring these negative relationships from more widespread content. We use the common judgments about items the users share, which are the data used in recommender systems. We provide three contributions, described in three distinct chapters. In the first one, we discuss the case of agreements about items that may not have the same semantics if they involve appreciated items or not by two users. We will see that disliking the same product does not mean similarity. Afterward, we consider in our second contribution the distributions of user ratings and items ratings in order to measure whether the agreements or disagreements may happen by chance or not, in particular to avoid the user and item biases observed in this type of data. Our third contribution consists in using these results to predict the sign of the links between users from the only positive ties and the common judgments about items, and then without any negative social information.

Enterprises use live video streaming as a mean of communication. Streaming high-quality video to thousands of devices in a corporate network is not an easy task; the bandwidth requirements often exceed the network capacity. For that matter, Peer-To-Peer (P2P) networks have been proven beneficial, as peers can exchange content efficiently by utilizing the topology of the corporate network. However, such networks are dynamic and their topology might not always be known. In this project we propose ABD, a new dynamic graph representation learning approach, which aims to estimate the bandwidth capacity between peers in a corporate network. The architecture of ABDis adapted to the properties of corporate networks. The model is composed of an attention mechanism and a decoder. The attention mechanism produces node embeddings, while the decoder converts those embeddings into bandwidth predictions. The model aims to capture both the dynamicity and the structure of the dynamic network, using an advanced training process. The performance of ABD is tested with two dynamic graphs which were produced by real corporate networks. Our results show that ABD achieves better results when compared to existing state-of-the-art dynamic graph representation learning models. / Företag använder live video streaming för både intern och extern kommunikation. Strömmning av hög kvalitet video till tusentals tittare i ett företagsnätverk är inte enkelt eftersom bandbreddskraven ofta överstiger kapaciteten på nätverket. För att minska lasten på nätverket har Peer-to-Peer (P2P) nätverk visat sig vara en lösning. Här anpassar sig P2P nätverket efter företagsnätverkets struktur och kan därigenom utbyta video data på ett effektivt sätt. Anpassning till ett företagsnätverk är ett utmanande problem eftersom dom är dynamiska med förändring över tid och kännedom över topologin är inte alltid tillgänglig. I det här projektet föreslår vi en ny lösning, ABD, en dynamisk approach baserat på inlärning av grafrepresentationer. Vi försöker estimera den bandbreddskapacitet som finns mellan två peers eller tittare. Architekturen av ABD anpassar sig till egenskaperna av företagsnätverket. Själva modellen bakom ABD använder en koncentrationsmekanism och en avkodare. Attention mekanismen producerar node embeddings, medan avkodaren konverterar embeddings till estimeringar av bandbredden. Modellen fångar upp dynamiken och strukturen av nätverket med hjälp av en avancerad träningsprocess. Effektiviteten av ABD är testad på två dynamiska nätverksgrafer baserat på data från riktiga företagsnätverk. Enligt våra experiment har ABD bättre resultat när man jämför med andra state-of the-art modeller för inlärning av dynamisk grafrepresentation.

No description available.

<p dir="ltr">Recent advancements in representation learning have significantly enhanced the analysis of relational data across various domains, including social networks, bioinformatics, and recommendation systems. In general, these methods assume that the training and test datasets come from the same distribution, an assumption that often fails in real-world scenarios due to evolving data, privacy constraints, and limited resources. The task of out-of-distribution (OOD) extrapolation emerges when the distribution of test data differs from that of the training data, presenting a significant, yet unresolved challenge within the field. This dissertation focuses on developing robust representations for effective OOD extrapolation, specifically targeting relational data types like graphs and sets. For successful OOD extrapolation, it's essential to first acquire a representation that is adequately expressive for tasks within the distribution. In the first work, we introduce Set Twister, a permutation-invariant set representation that generalizes and enhances the theoretical expressiveness of DeepSets, a simple and widely used permutation-invariant representation for set data, allowing it to capture higher-order dependencies. We showcase its implementation simplicity and computational efficiency, as well as its competitive performances with more complex state-of-the-art graph representations in several graph node classification tasks. Secondly, we address OOD scenarios in graph classification and link prediction tasks, particularly when faced with varying graph sizes. Under causal model assumptions, we derive approximately invariant graph representations that improve extrapolation in OOD graph classification task. Furthermore, we provide the first theoretical study of the capability of graph neural networks for inductive OOD link prediction and present a novel representation model that produces structural pairwise embeddings, maintaining predictive accuracy for OOD link prediction as the test graph size increases. Finally, we investigate the impact of environmental data as a confounder between input and target variables, proposing a novel approach utilizing an auxiliary dataset to mitigate distribution shifts. This comprehensive study not only advances our understanding of representation learning in OOD contexts but also highlights potential pathways for future research in enhancing model robustness across diverse applications.</p>

Online social networks (OSNs) are Web platforms providing different services to facilitate social interaction among their users. A particular kind of OSNs is the location-based social network (LBSN), which adds services based on location. One of the most important challenges in LBSNs is the link prediction problem. Link prediction problem aims to estimate the likelihood of the existence of future friendships among user pairs. Most of the existing studies in link prediction focus on the use of a single information source to perform predictions, i.e. only social information (e.g. social neighborhood) or only location information (e.g. common visited places). However, some researches have shown that the combination of different information sources can lead to more accurate predictions. In this sense, in this thesis we propose different link prediction methods based on the use of different information sources naturally existing in these networks. Thus, we propose seven new link prediction methods using the information related to user membership in social overlapping groups: common neighbors within and outside of common groups (WOCG), common neighbors of groups (CNG), common neighbors with total and partial overlapping of groups (TPOG), group naïve Bayes (GNB), group naïve Bayes of common neighbors (GNB-CN), group naïve Bayes of Adamic-Adar (GNB-AA) and group naïve Bayes of Resource Allocation (GNB-RA). Due to that social groups exist naturally in networks, our proposals can be used in any type of OSN.We also propose new eight link prediction methods combining location and social information: Check-in Observation (ChO), Check-in Allocation (ChA), Within and Outside of Common Places (WOCP), Common Neighbors of Places (CNP), Total and Partial Overlapping of Places (TPOP), Friend Allocation Within Common Places (FAW), Common Neighbors of Nearby Places (CNNP) and Nearby Distance Allocation (NDA). These eight methods are exclusively for work in LBSNs. Obtained results indicate that our proposals are as competitive as state-of-the-art methods, or better than they in certain scenarios. Moreover, since our proposals tend to be computationally more efficient, they are more suitable for real-world applications. / Redes sociais online (OSNs) são plataformas Web que oferecem serviços para promoção da interação social entre usuários. OSNs que adicionam serviços relacionados à geolocalização são chamadas redes sociais baseadas em localização (LBSNs). Um dos maiores desafios na análise de LBSNs é a predição de links. A predição de links refere-se ao problema de estimar a probabilidade de conexão futura entre pares de usuários que não se conhecem. Grande parte das pesquisas que focam nesse problema exploram o uso, de maneira isolada, de informações sociais (e.g. amigos em comum) ou de localização (e.g. locais comuns visitados). Porém, algumas pesquisas mostraram que a combinação de diferentes fontes de informação pode influenciar o incremento da acurácia da predição. Motivado por essa lacuna, neste trabalho foram desenvolvidos diferentes métodos para predição de links combinando diferentes fontes de informação. Assim, propomos sete métodos que usam a informação relacionada à participação simultânea de usuários en múltiples grupos sociais: common neighbors within and outside of common groups (WOCG), common neighbors of groups (CNG), common neighbors with total and partial overlapping of groups (TPOG), group naïve Bayes (GNB), group naïve Bayes of common neighbors (GNB-CN), group naïve Bayes of Adamic-Adar (GNB-AA), e group naïve Bayes of Resource Allocation (GNB-RA). Devido ao fato que a presença de grupos sociais não está restrita a alguns tipo de redes, essas propostas podem ser usadas nas diversas OSNs existentes, incluindo LBSNs. Também, propomos oito métodos que combinam o uso de informações sociais e de localização: Check-in Observation (ChO), Check-in Allocation (ChA), Within and Outside of Common Places (WOCP), Common Neighbors of Places (CNP), Total and Partial Overlapping of Places (TPOP), Friend Allocation Within Common Places (FAW), Common Neighbors of Nearby Places (CNNP), e Nearby Distance Allocation (NDA). Tais propostas são para uso exclusivo em LBSNs. Os resultados obtidos indicam que nossas propostas são tão competitivas quanto métodos do estado da arte, podendo até superá-los em determinados cenários. Ainda mais, devido a que na maioria dos casos nossas propostas são computacionalmente mais eficientes, seu uso resulta mais adequado em aplicações do mundo real.

Atualmente, as redes sociais estão ganhando cada vez mais destaque no dia-a-dia das pessoas. Nessas redes são estabelecidos diferentes relacionamentos entre entidades que compartilham alguma característica ou objetivo em comum. Diversas informações sobre a produção científica nacional podem ser encontradas na Plataforma Lattes, que é um sistema utilizado para o registro dos currículos dos pesquisadores no Brasil. A partir dessas informações é possível construir uma rede social acadêmica, na qual as relações entre os pesquisadores representam uma parceria na produção de uma publicação (coautoria) - um link. Na análise de redes sociais existe uma linha de pesquisa conhecida como predição de link ou de relacionamentos, que tem como objetivo identificar relacionamentos futuros. Essa tarefa pode favorecer a comunicação entre os usuários e otimizar o processo de produção científica identificando possíveis colaboradores. Este projeto analisou a influência de diferentes atributos encontrados na literatura e filtros de dados para prever relações de coautoria nas redes sociais acadêmicas. Foi abordado dois tipos de problemas na predição de relacionamentos, o problema geral que analisa todos os possíveis relacionamentos de coautoria e o problema de novas coautoria que refere-se aos relacionamentos de coautorias inéditas na rede. Os resultados dos experimentos foram promissores para o problema geral de predição com a combinação de atributos e filtros utilizados. Contudo, para o problema de novas coautorias, devido à sua maior complexidade, os resultados não foram tão bons. Os experimentos apresentados avaliaram diferentes estratégias e analisaram o custo e benefício de cada uma. Conclui-se que para lidar com o problema de predição de coautorias em redes sociais acadêmicas é necessário analisar as vantagens e desvantagens entre as estratégias, encontrando um equilíbrio entre a revocação da classe positiva e a acurácia geral / Nowadays, social networks are gaining prominence in the day-to-day lives. In these networks, different relationships are established between entities that share some characteristic or common goal. A huge amount of information about the Brazilian national scientific production can be found in the Lattes Platform, which is a system used to record the curricula of researchers in Brazil. From this information, it is possible to build an academic social network, where relations between researchers represent a partnership in the production of a publication - a link. In social network analysis there is a research area known as link prediction, which aims to identify future relationships. This task may facilitate communication among researchers and optimize the scientific production process identifying possible collaborators. This project analyzed the influence of different attributes found in the literature and data filters to predict co-authorship relationships in academic social networks. Was approached two types of problems in predicting relationships, the general problem that analyzes all possible co-authoring relationships and the problem of new co-authoring that relates to novel co-authorships relationships in the network. The experimental results were promising to the prediction general problem, combining attributes and using filters. However, for the new co-authorships problem the results were not as good. The experiments evaluated different strategies and analyzed the costs and benefits of each. We concluded that to deal with the co-authorships prediction problem in academic social networking it is necessary to analyze the advantages and disadvantages among the strategies, finding a balance between the recall of the positive class and the overall accuracy

Atualmente, as redes sociais estão ganhando cada vez mais destaque no dia-a-dia das pessoas. Nessas redes são estabelecidos diferentes relacionamentos entre entidades que compartilham alguma característica ou objetivo em comum. Diversas informações sobre a produção científica nacional podem ser encontradas na Plataforma Lattes, que é um sistema utilizado para o registro dos currículos dos pesquisadores no Brasil. A partir dessas informações é possível construir uma rede social acadêmica, na qual as relações entre os pesquisadores representam uma parceria na produção de uma publicação (coautoria) - um link. Na análise de redes sociais existe uma linha de pesquisa conhecida como predição de link ou de relacionamentos, que tem como objetivo identificar relacionamentos futuros. Essa tarefa pode favorecer a comunicação entre os usuários e otimizar o processo de produção científica identificando possíveis colaboradores. Este projeto analisou a influência de diferentes atributos encontrados na literatura e filtros de dados para prever relações de coautoria nas redes sociais acadêmicas. Foi abordado dois tipos de problemas na predição de relacionamentos, o problema geral que analisa todos os possíveis relacionamentos de coautoria e o problema de novas coautoria que refere-se aos relacionamentos de coautorias inéditas na rede. Os resultados dos experimentos foram promissores para o problema geral de predição com a combinação de atributos e filtros utilizados. Contudo, para o problema de novas coautorias, devido à sua maior complexidade, os resultados não foram tão bons. Os experimentos apresentados avaliaram diferentes estratégias e analisaram o custo e benefício de cada uma. Conclui-se que para lidar com o problema de predição de coautorias em redes sociais acadêmicas é necessário analisar as vantagens e desvantagens entre as estratégias, encontrando um equilíbrio entre a revocação da classe positiva e a acurácia geral / Nowadays, social networks are gaining prominence in the day-to-day lives. In these networks, different relationships are established between entities that share some characteristic or common goal. A huge amount of information about the Brazilian national scientific production can be found in the Lattes Platform, which is a system used to record the curricula of researchers in Brazil. From this information, it is possible to build an academic social network, where relations between researchers represent a partnership in the production of a publication - a link. In social network analysis there is a research area known as link prediction, which aims to identify future relationships. This task may facilitate communication among researchers and optimize the scientific production process identifying possible collaborators. This project analyzed the influence of different attributes found in the literature and data filters to predict co-authorship relationships in academic social networks. Was approached two types of problems in predicting relationships, the general problem that analyzes all possible co-authoring relationships and the problem of new co-authoring that relates to novel co-authorships relationships in the network. The experimental results were promising to the prediction general problem, combining attributes and using filters. However, for the new co-authorships problem the results were not as good. The experiments evaluated different strategies and analyzed the costs and benefits of each. We concluded that to deal with the co-authorships prediction problem in academic social networking it is necessary to analyze the advantages and disadvantages among the strategies, finding a balance between the recall of the positive class and the overall accuracy

En kunskapsgraf lagrar information från webben i form av relationer mellan olika entiteter. En kunskapsgrafs kvalité bestäms av hur komplett den är och dess noggrannhet. Dessvärre har många nuvarande kunskapsgrafer brister i form av saknad fakta och inkorrekt information. Nuvarande lösningar av länkförutsägelser mellan entiteter har problem med skalbarhet och hög arbetskostnad. Denna uppsats föreslår ett deklarativt regelbaserat probabilistiskt ramverk för att utföra länkförutsägelse. Systemet involverar en regelutvinnande modell till ett “hinge-loss Markov random fields” för att föreslå länkar. Vidare utvecklades tre strategier för regeloptimering för att förbättra reglernas kvalité. Jämfört med tidigare lösningar så bidrar detta arbete till att drastiskt reducera arbetskostnader och en mer spårbar modell. Varje metod har utvärderas med precision och F-värde på NELL och Freebase15k. Det visar sig att strategin för regeloptimering presterade bäst. MAP-uppskattningen för den bästa modellen på NELL är 0.754, vilket är bättre än en nuvarande spjutspetsteknologi graphical model(0.306). F-värdet för den bästa modellen på Freebase15k är 0.709. / The knowledge graph stores factual information from the web in form of relationships between entities. The quality of a knowledge graph is determined by its completeness and accuracy. However, most current knowledge graphs often miss facts or have incorrect information. Current link prediction solutions have problems of scalability and high labor costs. This thesis proposed a declarative rule-based probabilistic framework to perform link prediction. The system incorporates a rule-mining model into a hingeloss Markov random fields to infer links. Moreover, three rule optimization strategies were developed to improve the quality of rules. Compared with previous solutions, this work dramatically reduces manual costs and provides a more tractable model. Each proposed method has been evaluated with Average Precision or F-score on NELL and Freebase15k. It turns out that the rule optimization strategy performs the best. The MAP of the best model on NELL is 0.754, better than a state-of-the-art graphical model (0.306). The F-score of the best model on Freebase15k is 0.709.

During the last years, we have observed a progressive and continuous expansion of human-centered mobile wireless networks. The advent of these networks has encouraged the researchers to think about new solutions in order to ensure efficient evaluation and design of communication protocols. In fact, these networks are faced to several constraints as the lack of infrastructure, the dynamic topology, the limited resources and the deficient quality of service and security. We have been interested in the dynamicity of the network and in particular in human mobility. The human mobility has been widely studied in order to extract its intrinsic properties and to harness them to propose more accurate approaches. Among the prominent properties depicted in the literature, we have been specially attracted by the impact of the social interactions on the human mobility and consequently on the structure of the network. To grasp structural information of such networks, many metrics and techniques have been borrowed from the Social Network Analysis (SNA). The SNA can be seen as another network measurement task which extracts structural information of the network and provides useful feedback for communication protocols. In this context, the SNA has been extensively used to perform link prediction in social networks relying on their structural properties. Motivated by the importance of social ties in human-centered mobile wireless networks and by the possibilities that are brought by SNA to perform link prediction, we are interested by designing the first link prediction framework adapted for mobile wireless networks as Mobile Ad-hoc Networks (MANETs) and Delay/Disruption Tolerant Networks (DTN). Our proposal tracks the evolution of the network through a third-order tensor over T periods and computes the sociometric Katz measure for each pair of nodes to quantify the strength of the social ties between the network entities. Such quantification gives insights about the links that are expected to occur in the period T+1 and the new links that are created in the future without being observed during the tracking time. To attest the efficiency of our framework, we apply our link prediction technique on three real traces and we compare its performance to the ones of other well-known link prediction approaches. The results prove that our method reaches the highest level of accuracy and outperforms the other techniques. One of the major contributions behind our proposal highlights that the link prediction in such networks can be made in a distributed way. In other words, the nodes can predict their future links relying on the local information (one-hop and two-hop neighbors) instead of a full knowledge about the topology of the network. Furthermore, we are keen to improve the link prediction performance of our tensor-based framework. To quantify the social closeness between the users, we take into consideration two aspects of the relationships: the recentness of the interactions and their frequency. From this perspective, we wonder if we can consider a third criterion to improve the link prediction precision. Asserting the heuristic that stipulates that persistent links are highly predictable, we take into account the stability of the relationships (link and proximity stabilities). To measure it, we opt for the entropy estimation of a time series proposed in the Lempel-Ziv data compression algorithm. As we think that our framework measurements and the stability estimations complement each other, we combine them in order to provide new link prediction metrics. The simulation results emphasize the pertinence of our intuition. Providing a tensor-based link prediction framework and proposing relative enhancements tied to stability considerations represent the main contributions of this thesis. Along the thesis, our concern was also focused on mechanisms and metrics that contribute towards improving communication protocols in these mobile networks [...]