Introduction
The proposed tutorial aims to introduce the UMAP community to modern user profiling approaches leveraging graph neural networks (GNNs). We will begin by discussing the conceptual foundations of user profiling and GNNs and providing a literature review of the two topics. We will then present a systematic overview of the state-of-the-art GNN architectures designed for user profiling, including the types of data that are typically used for this purpose. We will also discuss ethical considerations and beyond-accuracy perspectives (i.e. fairness and explainability), which can arise within the potential applications of adopting GNNs for user profiling. In the practical session of the tutorial, attendees will have the opportunity to understand concretely how recent GNN models for user profiling are built and trained with open-source tools and publicly available datasets. The audience will also be engaged in investigating the impact of the presented models on case studies involving bias detection and mitigation, as well as user profiles explanations. The tutorial will end with an analysis of existing and emerging open challenges in the field and their future research directions.
Target Audience
The tutorial is intended for a beginner or intermediate audience and is open to researchers, industry technologists and practitioners. It covers all the necessary background on user profiling and graph neural networks in order to make it also accessible to people without prior knowledge about these topics, which is not assumed. Basic knowledge of Python programming is preferred, as well as familiarity with common data science libraries, such as Pandas and NumPy. It is worth noticing that, despite the technical aspects, the key concepts illustrated during the tutorial and their applications touch a range of interdisciplinary fields, making also the proposed outline of interest to an interdisciplinary audience.
Outline
This tutorial will take place on June 26, 2023 in Limassol, Cyprus, as part of the 31st ACM Conference on User Modeling, Adaptation and Personalization (UMAP 2023).
| Timing | Content |
|---|---|
| 5 mins | Opening and instructors' presentation |
| 80 mins | User profiling session |
|
Introduction to user profiling (15 mins)
Historical overview of the research on user profiling is given to set the basis for understanding the recent advances. In particular, starting from the definition of the key terms in the field (e.g. the difference between explicit and implicit user profiling), we will illustrate several user profiling contributions in different domains.
|
|
|
Introduction to GNNs (15 mins)
As for the user profiling part, in this portion of the tutorial, we will cover the most important notions about graph neural networks, such as basic terminology and most popular architectures (e.g. GCN and GAT), with the aim to create a common background to make the audience able to follow the core tutorial sections.
|
|
|
GNN-based models for user profiling (25 mins)
We will present in detail the current state-of-the-art GNN-based model for user profiling, such as CatGCN, RHGN and others, describing their architectures, their training procedures, and discussing their strengths and weaknesses.
|
|
|
Hands on state-of-the-art GNN-based models for user profiling (25 mins)
To show in practice how the described GNN models are designed and implemented, we will execute and explain some of the models, such as CatGCN and RHGN, in their original configuration, also illustrating the used datasets for every contribution (e.g. Alibaba and JD).
|
|
| 5 mins |
Q&A
|
| 30 mins | Coffee Break |
| 75 mins | Beyond-accuracy perspectives session |
|
Overview on beyond-accuracy perspectives for user profiling with GNNs: Algorithmic Fairness (15 mins)
The fairness aspects encountered in user profiling research will be discussed. Several points of view will be taken into account:
|
|
|
Use cases on beyond-accuracy perspectives (Fairness) (35 mins)
We will first illustrate the implementation of four standard fairness metrics to the illustrated models.
Then, we describe FairGNN, a state-of-the-art GNN-based framework for debiasing, and run it on the datasets used in the original publication (i.e. NBA and Pokec).
We will also apply the three approaches described in the last section of the theoretical session by making use of a standardised framework developed for analysing fairness in GNN-based models for user profiling. In particular, the framework can get a graph structure in NetworkX or Neo4J format and generate the specific input for each of the GNNs.
|
|
|
Overview on beyond-accuracy perspectives for user profiling with GNNs: Explainability (10 mins)
The explainability aspects encountered in user profiling research will be discussed.
|
|
|
Open challenges and concluding remarks (15 mins)
We will discuss the existing open challenges in user profiling research, as well as wrap up the tutorial content.
|
|
| 10 mins | Q&A |
| 5 mins | Closing |
Material
- Slides
- Github
- Resource Folder (no datasets)
- Datasets
- Notebook 1: CatGCN
- Notebook 2: RHGN
- Notebook 3: Fairness
- Notebook 4: FairGNN
If you find the content and the slides of the tutorial useful for your research, we would appreciate an acknowledgment by citing our summary in the UMAP '23 proceedings:
E. Purificato, L. Boratto, and E. W. De Luca. 2023. Tutorial on User Profiling with Graph Neural Networks and Related Beyond-Accuracy Perspectives. In Proceedings of the 31st ACM Conference on User Modeling, Adaptation and Personalization (UMAP '23). Association for Computing Machinery, New York, NY, USA, 309-312.
If you also appreciate the hands-on sessions and the notebooks, we would be glad to have your acknowledgement in case they are helpful for your research by citing our CIKM'22 and SIGIR'23 papers:
E. Purificato, L. Boratto, and E. W. De Luca. 2022. Do Graph Neural Networks Build Fair User Models? Assessing Disparate Impact and Mistreatment in Behavioural User Profiling. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management (CIKM '22). Association for Computing Machinery, New York, NY, USA, 4399-4403.
M. Abdelrazek, E. Purificato, L. Boratto, and E. W. De Luca. 2023. FairUP: a Framework for Fairness Analysis of Graph Neural Network-Based User Profiling Models. To appear in Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '23). Association for Computing Machinery, New York, NY, USA.
Presenters
Erasmo Purificato
Otto von Guericke University Magdeburg (Germany)
Leibniz Institute for Educational Media | Georg Eckert Institute (Germany)
Ludovico Boratto
University of Cagliari (Italy)
Ernesto William De Luca
Otto von Guericke University Magdeburg (Germany)
Leibniz Institute for Educational Media | Georg Eckert Institute (Germany)
Registration
Registration to the tutorial will be managed by the UMAP 2023 main conference organization.
Contacts
For any request you might have, please contact us at erasmo.purificato [at] ovgu.de, ludovico.boratto [at] acm.org and ernesto.deluca [at] ovgu.de.