Doctoral Consortium presentation @ UMAP'22: 30th ACM Conference on User Modeling, Adaptation and Personalization (July 4-7, 2022 | Barcelona, Spain)

1. Beyond-Accuracy Perspectives on
Graph Neural Network-Based Models
for Behavioural User Profiling
Erasmo Purificato

2. Beyond-Accuracy Perspectives on Graph Neural
Network-Based Models for Behavioural User Profiling
◎ User profiling infers an indivisual’s interests, personality traits or behaviours
from generated data to create an efficient representation, i.e. a user model.
◎ Early profiling approaches considered only the analysis of static characteristics
(explicit user profiling), with data often coming from surveys.
◎ Modern systems focus on profiling users’ data based on individuals’ actions and
interactions (implicit user profiling).

3. Beyond-Accuracy Perspectives on Graph Neural
Network-Based Models for Behavioural User Profiling
◎ Graphs are a natural way to model behaviours (node/edge ≡ user/interaction).
◎ Graph Neural Networks (GNNs) are the perfect class of neural methods to deal
with data represented by graph data structures.
◎ Recent studies have demonstrated the effectiveness of GNNs in modelling
graph data on several domains, such as recommender systems, natural
language processing and user profiling.

4. Beyond-Accuracy Perspectives on Graph Neural
Network-Based Models for Behavioural User Profiling
◎ Existing approaches evaluate GNN-based user profiling models based on the
effectiveness of a classification task at predicting a user’s characteristics.
◎ I aim to look beyond the usual accuracy-based approaches by simultaneously
considering the perspectives of fairness, explainability and privacy.
◎ Only a few studies produced on these topics for GNNs and none of them
combine them together.

5. Main goal
Leverage GNN models to produce fair and
privacy-preserving user representations,
having the ability to provide tailored
explanations to the end-users through an
adaptive and personalised user interface

6. GNN-based behavioural user profiling framework

7. [RQ1] Privacy
How can we guarantee personal data protection on a graph data structure while
avoiding affecting user models construction and retaining the performance of the
recommender system built upon them?

8. [RQ2] Fairness
How do we build fair user representations from GNN-based user profiling models to
keep the input of the downstream recommender debiased?

9. [RQ3] Explainability
How can we personalise user interfaces to adapt the explanations to the needs,
demands and requirements of different end-user profiles, considering their distinct
knowledge, background and expertise?

10. Contributions to date
[RQ1]
E. Purificato, S. Wehnert, and E. W. De Luca.
Dynamic Privacy-Preserving
Recommendations on Academic Graph Data.
In Computers 10, 9 (2021), 107.
[RQ3]
E. Purificato, C. Musto, P. Lops, and E. W. De Luca.
First Workshop on Adaptive and Personalized
Explainable User Interfaces (APEx-UI 2022). In 27th
International Conference on Intelligent User Interfaces
(IUI ’22 Companion). ACM, New York, NY, USA, 1–3.
[RQ3]
E. Purificato, B. Aiyer, P. Karanam, M. Pattadkal, and E. W. De Luca.
Evaluating Explainable Interfaces for a Knowledge Graph-Based
Recommender System. In Proceedings of the 8th Joint Workshop
on Interfaces and Human Decision Making for Recommender
Systems, co-located with RecSys’21. 73–88.
[RQ2 – RQ3]
E. Purificato, F. Lorenzo, F. Fallucchi, and E. W. De Luca.
The Use of Responsible Artificial Intelligence in the
Context of Loan Approval Processes. In International
Journal of Human-Computer Interaction (2022), 1-20.

11. Thanks!
I am Erasmo Purificato
You can find me at:
https://erasmopurif.com
@erasmopurif11
erasmo.purificato@ovgu.de