Mining Temporal Networks

In World Wide Web (WWW) systems, networks (or graphs) serve as a fundamental tool for representing, analyzing, and understanding linked data, providing significant insights into the underlying systems. Naturally, most real-world systems have inherent temporal information, e.g., interactions in social networks occur at specific moments in time and last for a certain period. Temporal networks, i.e., network data modeling temporal information, enable novel and fundamental discoveries about the underlying systems they model, otherwise not captured by static networks that ignore such temporal information. In this tutorial, we present state-of-the-art models and algorithmic techniques for mining temporal networks that can provide precious insights into a plethora of web-related applications. We present how temporal networks can be used to extract novel information, especially in web-related network data, and highlight the challenges that arise when modeling temporal information compared to traditional static network-based approaches. We first overview different temporal network models. We then show how such powerful models can be leveraged to extract novel insights through suitable mining primitives. In particular, we present recent advances addressing most foundational problems for temporal network mining---ranging from the computation of temporal centrality measures, temporal motif counting, and temporal communities to bursty events and anomaly detection.


TOPIC AND RELEVANCE
The interconnected nature of the web and online systems often leads to intricate data relationships.Networks serve as a fundamental tool for representing, analyzing, and understanding these relationships [48].A network, comprises two essential components: (i) a set of vertices that can represent various online entities (like websites, servers, users, or content) and (ii) a set of edges that define the relationships or connections between these entities, where such connections range from hyperlinks between web pages to social network interactions, and more [15,45].
While network data related to WWW applications is often viewed as static, its nature is inherently temporal.In fact, web-related systems evolve continuously, i.e., websites get updates, users interact on online platforms or social networks, and the content we are exposed to changes drastically over time.Accounting for such temporal information can provide novel and unique insights, possibly explaining the nature of many phenomena otherwise not well understood [11,23,33,44,57].
A temporal network encodes the inherent temporal evolution of a web-related system by capturing the dynamics over the network, e.g., users joining or leaving a social network or online communications like email and chats.Hence, temporal graphs provide a new lens to study and understand complex phenomena that would not be feasible to study otherwise, such as temporal patterns for misinformation spreading over social networks, burstiness in communication networks [5], suspicious transactions in financial networks [79], robustness of IP-networks and many more.Such more powerful representation power comes at a higher price.In fact, modeling temporal networks is non trivial, many models have been proposed in the literature and each of them can capture different temporal properties of the data [24].Additionally, many problems that can be studied on temporal networks become significantly harder than their counterpart for static networks [9,38].
In this tutorial, we present the different temporal network models that exist and illustrate their representation capabilities.We also discuss representative state-of-the-art mining techniques that can be applied to such temporal-network models.
We will not cover learning algorithms (i.e., machine learning, deep learning) for temporal networks-a comprehensive exploration of such a vast field is outside the intended scope of this tutorial [32,77,80].Instead, we will place emphasis on sound algorithmic approaches and their applications in the domain of webrelated systems.We will also address open problems and challenges in the field of temporal network mining.

Tutors
Aristides Gionis is a professor in the department of Division of Theoretical Computer Science at KTH Royal Institute of Technology in Stockholm, Sweden.He has been a fellow in the ISI foundation, Turin, and a visiting professor in the University of Rome.His previous appointment was with Yahoo!Research, Barcelona, where he has been a senior research scientist and group leader.He obtained his PhD in 2003 from Stanford University, USA.He is currently serving as an action editor in the Data Management and Knowledge Discovery journal (DMKD), and an associate editor in the ACM Transactions on the Web (TWEB).He has contributed in several areas of data science, such as algorithmic data analysis, web mining, social-media analysis, data clustering, and privacy-preserving data mining.His current research is funded by the European Commission with an ERC Advanced grant (REBOUND) and with RIA project SoBigData++, and by the Wallenberg AI, Autonomous Sys- In particular, recently he focused on efficient algorithms with probabilistic guarantees for extracting patterns and computing centrality measures in temporal networks.

TUTORIAL DETAILS
• Style: The tutorial will be a lecture-style tutorial.There is no requirement for the audience members to use any software during the tutorial.• Tutorial materials: We provide an accompanying website, 1  containing supplementary materials (such as a complete list of references, a detailed outline of the content of this tutorial, a list of software packages for temporal networks, etc.).Furthermore, the tutorial slides will be available on the website to ensure accessibility beyond the conference.There are no copyright issues.• Interactive engagement: We plan to engage the audience by incorporating interactive online quizzes, creating a dynamic learning experience where participants can test their knowledge, receive real-time feedback, and actively participate in discussions.

Audience
Target audience.This tutorial is designed for graduate students, researchers, and practitioners who are interested in the analysis of temporal networked data, especially related to the web (such as social networks, web servers, etc.), and researchers interested in algorithmic aspects of temporal networks and their recent advances.Requirements and topics.A basic familiarity with network analysis can help the learning experience.Nevertheless, for most of the tutorial we provide all the required notions to understand the content being presented.In the first part of the tutorial, we overview models and fundamental measures and show their possible applications, thus this part does not require any technical background.In the other two parts, we cover problem formulations and algorithmic techniques related to such problems.These parts are targeted to researchers who (or wish to) work in this area, but also to developers and practitioners looking for specific solutions captured by the problems that will be presented.While some of the mining primitives may be described technically, overall, the presentation is expected to be at the level of details to be well understood by a vast audience with heterogeneous backgrounds in computer science and related fields.
Expected outcomes.The audience, after attending the tutorial, is expected: (i) to gain novel insights on the importance of analyzing web-related systems with respect to their temporal information; (ii) get a comprehensive view of the main temporal network data models, algorithmic problems, their solutions, and software packages implementing the solutions, as well as fundamental works existing in such area, and understand how such results can be used in their own field; iii) get insights into open problems and challenges that are unsolved, that can provide positive a impact on the community.

Schedule
Our tutorial will be 3 hours long, outlined as follows.
Tools and code libraries [5 min] Challenges, open problems, and trends [10 min]

PREVIOUS EDITIONS
An earlier edition of this tutorial has been presented two times: (1) at KDD 2019 [66] with around 50 participants; and (2) at the EDBT summer school 2019 with around 60 participants.The tutors of both previous editions were Aristides Gionis and Polina Rozenshtein, and the materials of the previous editions can be found on the corresponding website. 2  Our new tutorial distinguishes itself from the previous versions through its updated content, where we added most of the latest advancements in the field of temporal network mining, as well as a more dedicated focus on web-based applications.Additionally, we improved and adapted the presentation to enhance clarity, engagement, and overall learning experience for our participants.
tems and Software Program (WASP) in Sweden.Aristides Gionis has presented several tutorials on graph mining and web mining in conferences such as the Web Conference (2008, 2018, 2020, 2021, and 2022), KDD (2013, 2015, 2018, and 2019), ECML PKDD (2008, 2013, and 2015), IJCAI (2011 and 2022), as well as in many summer schools, including the EDBT Summer School 2019 and the Hi! Paris Data Science Summer School 2022.Lutz Oettershagen is a postdoctoral researcher in the Division of Theoretical Computer Science at KTH Royal Institute of Technology in Stockholm, Sweden.He has been a postdoctoral researcher at the University of Bonn in Germany, where he also obtained his PhD.He obtained his master's degree at the TU University of Dortmund.His main research interests are algorithmic data analysis and data mining on temporal networks, focusing on social networks.In recent works, he covers temporal centrality measures, community search, and tie strength inference in temporal networks.Ilie Sarpe is a postdoctoral researcher in the Division of Theoretical Computer Science at KTH Royal Institute of Technology in Stockholm, Sweden.He obtained his PhD in Information Engineering from the University of Padova, Italy where he also earned his MSc in 2023 and 2019 respectively.His research interests focus on the development of scalable algorithms with rigorous theoretical guarantees for data mining primitives, and randomized algorithms.