Graphs have attracted researchers in various fields in that many different kinds of real-world entities and relationships between them can be represented and analyzed effectively and efficiently using graphs. In particular, researchers in data mining and machine learning areas have developed algorithms and models to understand the complex graph data better and perform various data mining tasks. While a large body of work exists on graph mining, most existing work does not fully exploit attributes attached to graph nodes or edges.<div><br></div><div>In this dissertation, we exploit node attributes to generate better solutions to several graph data mining problems addressed in the literature. First, we introduce the notion of statistically significant attribute associations in attribute graphs and propose an effective and efficient algorithm to discover those associations. The effectiveness analysis on the results shows that our proposed algorithm can reveal insightful attribute associations that cannot be identified using the earlier methods focused solely on frequency. Second, we build a probabilistic generative model for observed attributed graphs. Under the assumption that there exist hidden communities behind nodes in a graph, we adopt the idea of latent topic distributions to model a generative process of node attribute values and link structure more precisely. This model can be used to detect hidden communities and profile missing attribute values. Lastly, we investigate how to employ node attributes to learn latent representations of nodes in lower dimensional embedding spaces and use the learned representations to improve the performance of data mining tasks over attributed graphs.<br></div>

abstract: Graph is a ubiquitous data structure, which appears in a broad range of real-world scenarios. Accordingly, there has been a surge of research to represent and learn from graphs in order to accomplish various machine learning and graph analysis tasks. However, most of these efforts only utilize the graph structure while nodes in real-world graphs usually come with a rich set of attributes. Typical examples of such nodes and their attributes are users and their profiles in social networks, scientific articles and their content in citation networks, protein molecules and their gene sets in biological networks as well as web pages and their content on the Web. Utilizing node features in such graphs---attributed graphs---can alleviate the graph sparsity problem and help explain various phenomena (e.g., the motives behind the formation of communities in social networks). Therefore, further study of attributed graphs is required to take full advantage of node attributes. In the wild, attributed graphs are usually unlabeled. Moreover, annotating data is an expensive and time-consuming process, which suffers from many limitations such as annotators’ subjectivity, reproducibility, and consistency. The challenges of data annotation and the growing increase of unlabeled attributed graphs in various real-world applications significantly demand unsupervised learning for attributed graphs. In this dissertation, I propose a set of novel models to learn from attributed graphs in an unsupervised manner. To better understand and represent nodes and communities in attributed graphs, I present different models in node and community levels. In node level, I utilize node features as well as the graph structure in attributed graphs to learn distributed representations of nodes, which can be useful in a variety of downstream machine learning applications. In community level, with a focus on social media, I take advantage of both node attributes and the graph structure to discover not only communities but also their sentiment-driven profiles and inter-community relations (i.e., alliance, antagonism, or no relation). The discovered community profiles and relations help to better understand the structure and dynamics of social media. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2019

Ce travail de thèse concerne la fouille de données sur des graphes attribués. Il s'agit de graphes dans lesquels des propriétés, encodées sous forme d'attributs, sont associées à chaque sommet. Notre objectif est la découverte, dans ce type de données, de sous-graphes organisés en plusieurs groupes de sommets fortement connectés et homogènes au regard des attributs. Plus précisément, nous définissons l'extraction sous contraintes d'ensembles de sous-graphes densément connectés et tels que les sommets partagent suffisamment d'attributs. Pour cela nous proposons deux familles de motifs originales ainsi que les algorithmes justes et complets permettant leur extraction efficace sous contraintes. La première famille, nommée Ensembles Maximaux de Cliques Homogènes, correspond à des motifs satisfaisant des contraintes concernant le nombre de sous-graphes denses, la taille de ces sous-graphes et le nombre d'attributs partagés. La seconde famille, nommée Collections Homogènes de k-cliques Percolées emploie quant à elle une notion de densité plus relaxée permettant d'adapter la méthode aux données avec des valeurs manquantes. Ces deux méthodes sont appliquées à l'analyse de deux types de réseaux, les réseaux de coopérations entre chercheurs et les réseaux d'interactions de protéines. Les motifs obtenus mettent en évidence des structures utiles dans un processus de prise de décision. Ainsi, dans un réseau de coopérations entre chercheurs, l'analyse de ces structures peut aider à la mise en place de collaborations scientifiques entre des groupes travaillant sur un même domaine. Dans le contexte d'un graphe de protéines, les structures exhibées permettent d'étudier les relations entre des modules de protéines intervenant dans des situations biologiques similaires. L'étude des performances en fonction de différentes caractéristiques de graphes attribués réels et synthétiques montre que les approches proposées sont utilisables sur de grands jeux de données. / The work presented in this thesis deals with data mining approaches for the analysis of attributed graphs. An attributed graph is a graph where properties, encoded by means of attributes, are associated to each vertex. In such data, our objective is the discovery of subgraphs formed by several dense groups of vertices that are homogeneous with respect to the attributes. More precisely, we define the constraint-based extraction of collections of subgraphs densely connected and such that the vertices share enough attributes. To this aim, we propose two new classes of patterns along with sound and complete algorithms to compute them efficiently using constraint-based approaches. The first family of patterns, named Maximal Homogeneous Clique Set (MHCS), contains patterns satisfying constraints on the number of dense subgraphs, on the size of these subgraphs, and on the number of shared attributes. The second class of patterns, named Collection of Homogeneous k-clique Percolated components (CoHoP), is based on a relaxed notion of density in order to handle missing values. Both approaches are used for the analysis of scientific collaboration networks and protein-protein interaction networks. The extracted patterns exhibit structures useful in a decision support process. Indeed, in a scientific collaboration network, the analysis of such structures might give hints to propose new collaborations between researchers working on the same subjects. In a protein-protein interaction network, the analysis of the extracted patterns can be used to study the relationships between modules of proteins involved in similar biological situations. The analysis of the performances, on real and synthetic data, with respect to different attributed graph characteristics, shows that the proposed approaches scale well for large datasets.

Cette thèse s'est déroulée dans le cadre du projet ANR FOSTER, "FOuille de données Spatio-Temporelles : application à la compréhension et à la surveillance de l'ERosion" (ANR-2010-COSI-012-02, 2011-2014). Dans ce contexte, nous nous sommes intéressés à la modélisation de données spatio-temporelles dans des graphes enrichis de sorte que des calculs de motifs sur de telles données permettent de formuler des hypothèses intéressantes sur les phénomènes à comprendre. Plus précisément, nous travaillons sur la fouille de motifs dans des graphes relationnels (chaque noeud est identifié de fa\c con unique), attribués (chaque noeud du graphe est décrit par des attributs qui sont ici numériques), et dynamiques (les valeurs des attributs et les relations entre les noeuds peuvent évoluer dans le temps). Nous proposons un nouveau domaine de motifs nommé motifs de co-évolution. Ce sont des triplets d'ensembles de noeuds, d'ensembles de pas de temps et d'ensembles d'attributs signés, c'est à dire des attributs associés à une tendance (croissance,décroissance). L'intérêt de ces motifs est de décrire un sous-ensemble des données qui possède un comportement spécifique et a priori intéressant pour conduire des analyses non triviales. Dans ce but, nous définissons deux types de contraintes, une contrainte sur la structure du graphe et une contrainte sur la co-évolution de la valeur des attributs portés par les noeuds. Pour confirmer la spécificité du motif par rapport au reste des données, nous définissons trois mesures de densité qui tendent à répondre à trois questions. À quel point le comportement des noeuds en dehors du motif est similaire à celui des noeuds du motif ? Quel est le comportement du motif dans le temps, est-ce qu'il apparaît soudainement ? Est-ce que les noeuds du motif ont un comportement similaire seulement sur les attributs du motif ou aussi en dehors ? Nous proposons l'utilisation d'une hiérarchie sur les attributs comme connaissance à priori de l'utilisateur afin d'obtenir des motifs plus généraux et adaptons l'ensemble des contraintes à l'utilisation de cette hiérarchie. Finalement, pour simplifier l'utilisation de l'algorithme par l'utilisateur en réduisant le nombre de seuils à fixer et pour extraire uniquement l'ensemble des motifs les plus intéressants, nous utilisons le concept de ``skyline'' réintroduit récemment dans le domaine de la fouille de données. Nous proposons ainsi trois algorithmes MINTAG, H-MINTAG et Sky-H-MINTAG qui sont complets pour extraire l'ensemble de tous les motifs qui respectent les différentes contraintes. L'étude des propriétés des contraintes (anti-monotonie, monotonie/anti-monotonie par parties) nous permet de les pousser efficacement dans les algorithmes proposés et d'obtenir ainsi des extractions sur des données réelles dans des temps raisonnables. / This thesis was conducted within the project ANR FOSTER, ``Spatio-Temporal Data Mining: application to the understanding and monitoring of erosion'' (ANR-2010-COSI-012-02, 2011-2014). In this context, we are interested in the modeling of spatio- temporal data in enriched graphs so that computation of patterns on such data can be used to formulate interesting hypotheses about phenomena to understand. Specifically, we are working on pattern mining in relational graphs (each vertex is uniquely identified), attributed (each vertex of the graph is described by numerical attributes) and dynamic (attribute values and relations between vertices may change over time). We propose a new pattern domain that has been called co-evolution patterns. These are trisets of vertices, times and signed attributes, i.e., attributes associated with a trend (increasing or decreasing). The interest of these patterns is to describe a subset of the data that has a specific behaviour and a priori interesting to conduct non-trivial analysis. For this purpose, we define two types of constraints, a constraint on the structure of the graph and a constraint on the co-evolution of the value worn by vertices attributes. To confirm the specificity of the pattern with regard to the rest of the data, we define three measures of density that tend to answer to three questions. How similar is the behaviour of the vertices outside the co-evolution pattern to the ones inside it? What is the behaviour of the pattern over time, does it appear suddenly? Does the vertices of the pattern behave similarly only on the attributes of the pattern or even outside? We propose the use of a hierarchy of attributes as an a priori knowledge of the user to obtain more general patterns and we adapt the set of constraints to the use of this hierarchy. Finally, to simplify the use of the algorithm by the user by reducing the number of thresholds to be set and to extract only all the most interesting patterns, we use the concept of ``skyline'' reintroduced recently in the domain of data mining. We propose three constraint-based algorithms, called MINTAG, H-MINTAG and Sky-H-MINTAG, that are complete to extract the set of all patterns that meet the different constraints. These algorithms are based on constraints, i.e., they use the anti-monotonicity and piecewise monotonicity/anti-monotonicity properties to prune the search space and make the computation feasible in practical contexts. To validate our method, we experiment on several sets of data (graphs) created from real-world data.

<p>The most important networks from different domains—such as Computing, Organization, Economic, Social, Academic, and Biology—are networks that change over time. For example, in an organization there are email and collaboration networks (e.g., different people or teams working on a document). Apart from the connectivity of the networks changing over time, they can contain attributes such as the topic of an email or message, contents of a document, or the interests of a person in an academic citation or a social network. Analyzing these dynamic networks can be critical in decision-making processes. For instance, in an organization, getting insight into how people from different teams collaborate, provides important information that can be used to optimize workflows.</p> <p><br></p> <p>Network generative models provide a way to study and analyze networks. For example, benchmarking model performance and generalization in tasks like node classification, can be done by evaluating models on synthetic networks generated with varying structure and attribute correlation. In this work, we begin by presenting our systemic study of the impact that graph structure and attribute auto-correlation on the task of node classification using collective inference. This is the first time such an extensive study has been done. We take advantage of a recently developed method that samples attributed networks—although static—with varying network structure jointly with correlated attributes. We find that the graph connectivity that contributes to the network auto-correlation (i.e., the local relationships of nodes) and density have the highest impact on the performance of collective inference methods.</p> <p><br></p> <p>Most of the literature to date has focused on static representations of networks, partially due to the difficulty of finding readily-available datasets of dynamic networks. Dynamic network generative models can bridge this gap by generating synthetic graphs similar to observed real-world networks. Given that motifs have been established as building blocks for the structure of real-world networks, modeling them can help to generate the graph structure seen and capture correlations in node connections and activity. Therefore, we continue with a study of motif evolution in <em>dynamic</em> temporal graphs. Our key insight is that motifs rarely change configurations in fast-changing dynamic networks (e.g. wedges intotriangles, and vice-versa), but rather keep reappearing at different times while keeping the same configuration. This finding motivates the generative process of our proposed models, using temporal motifs as building blocks, that generates dynamic graphs with links that appear and disappear over time.</p> <p><br></p> <p>Our first proposed model generates dynamic networks based on motif-activity and the roles that nodes play in a motif. For example, a wedge is sampled based on the likelihood of one node having the role of hub with the two other nodes being the spokes. Our model learns all parameters from observed data, with the goal of producing synthetic graphs with similar graph structure and node behavior. We find that using motifs and node roles helps our model generate the more complex structures and the temporal node behavior seen in real-world dynamic networks.</p> <p><br></p> <p>After observing that using motif node-roles helps to capture the changing local structure and behavior of nodes, we extend our work to also consider the attributes generated by nodes’ activities. We propose a second generative model for attributed dynamic networks that (i) captures network structure dynamics through temporal motifs, and (ii) extends the structural roles of nodes in motifs to roles that generate content embeddings. Our new proposed model is the first to generate synthetic dynamic networks and sample content embeddings based on motif node roles. To the best of our knowledge, it is the only attributed dynamic network model that can generate <em>new</em> content embeddings—not observed in the input graph, but still similar to that of the input graph. Our results show that modeling the network attributes with higher-order structures (e.g., motifs) improves the quality of the networks generated.</p> <p><br></p> <p>The generative models proposed address the difficulty of finding readily-available datasets of dynamic networks—attributed or not. This work will also allow others to: (i) generate networks that they can share without divulging individual’s private data, (ii) benchmark model performance, and (iii) explore model generalization on a broader range of conditions, among other uses. Finally, the evaluation measures proposed will elucidate models, allowing fellow researchers to push forward in these domains.</p>