Explainable Recommender Systems with Knowledge Graphs and Language Models

Introduction

In this tutorial, we delve into recent advances in explainable recommendation using Knowledge Graphs (KGs). The session begins by introducing the fundamental principles behind the increasing adoption of KGs in modern recommender systems. Then, the tutorial explores recent techniques that leverage KGs as an input for language models tailored to explainable recommendation, describing also data types, methods, and evaluation protocols and metrics. Conceptual elements are complemented with hands-on sessions, providing practical implementations using open- source tools and public datasets. Concluding with a comprehensive case study in the education domain as a recap, the tutorial analyses emerging issues and outlines prospective trajectories in this field.

Target Audience

This beginner/intermediate-level tutorial is accessible to researchers, technologists and practitioners. For people not familiar with recommender systems, this tutorial covers necessary background material. Moreover, no prior knowledge on explanations, knowledge graphs, and recommender systems with knowledge graphs is assumed. Basic knowledge of Python programming and of quite common libraries, such as Pandas and NumPy, is preferred. One aspect relevant from the outline is that the explainability perspective of our tutorial is an interdisciplinary topic, touching on several dimensions beyond algorithms and being of interest for people with different backgrounds.

Outline

This tutorial will take place on March 24th, 2024 in Glasgow, Scotland, as part of the The 46th European Conference on Information Retrieval (ECIR 2024)

Timing	Content
5 mins	Welcome and Presenters' Introduction
20 mins	Introduction to explainable recommendation We will first provide a historical overview of the explainable recommendation research, tracing its origins and evolution We will explore foundational concepts behind explainable recommendations. We will discuss explanations based on the timing of their generation (intrinsic vs post-hoc), whether they are tied to specific models or can be applied universally, and the various sources that can be used to construct them.
25 mins	Hands on interaction data with KGs We first present three data sets representing the movie (MovieLens-1M (ML1M)), music (LastFM-1B (LASTFM)) and e-commerce (Amazon-Cellphones) domains. They are public and vary in domain, extensiveness, and sparsity. We load and present existing knowledge graphs for the three considered data sets and show how such information should be pre-processed to enable explainability in recommendation models. We will present objectives influenced by explanations (utility, coverage, diversity, novelty, visibility, exposure) and provide related work. Explanations also have an impact on several perspectives such as the economy, law, society, trust, technology, and psychology.
20 mins	Explainable RS with path reasoning methods We will provide an overview of the recommendation pipeline, highlighting stages necessary for explainability's integration. We introduce the primary methodological families in knowledge-grounded recommendations, such as embedding-based and path-based approaches, along with representative methods from each category. We pivot to a granular exploration of path-based reasoning, spotlighting methods like PGPR and CAFE.
25 mins	Path Reasoning: PGPR & CAFE We dive into the practical aspects of implementing PGPR and CAFE
20 mins	Language Models for Explainable RS We provide a overview on the paradigms of language modeling (i.e., causal, masked, and text-to-text) and their applications, focusing on recommendation. We explore innovative integrations of KGs and language models with PLM and PEARLM, showcasing their unique ability in path reconstruction for explainable recommendations.
25 mins	Language Models: PLM & PEARLM We initiate with an insight into formulating training data by sampling paths from knowledge graphs. We will leverage pretrained models to generate top-k recommendations alongside their explanations. We tackle the issue of hallucination where language models generate explanations not aligned with the KG.
25 mins	Entire Pipeline in Action for Educational Recommendation We will recap the entire recommender system workflow, addressing challenges like complex KG creation, model training and evaluation in the context of education. We will provide an immersive hands-on experience tailored to overcome the sector-specific hurdles and ensure the delivery of pertinent educational resources.
20 mins	Challenges, Open Issues and Conclusions

Material

If the tutorial slides are useful for your research, we would appreciate an acknowledgment by citing our summary in the ECIR 2024 proceedings:

Balloccu, G., Boratto, L., Fenu, G., Malloci, F. M., & Marras, M. (2024). Explainable Recommender Systems with Knowledge Graphs and Language Models. In European Conference on Information Retrieval (pp. 352-357). Cham: Springer Nature Switzerland.

Balloccu, G., Boratto, L., Fenu, G., & Marras, M. (2022). Post Processing Recommender Systems with Knowledge Graphs for Recency, Popularity, and Diversity of Explanations. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 646–656).

Balloccu, G., Boratto, L., Fenu, G., & Marras, M. (2023). Reinforcement recommendation reasoning through knowledge graphs for explanation path quality. Knowledge-Based Systems, 260, 110098.

Balloccu, G., Boratto, L., Cancedda, C., Fenu, G., & Marras, M. (2023, March). Knowledge is Power, Understanding is Impact: Utility and Beyond Goals, Explanation Quality, and Fairness in Path Reasoning Recommendation. In European Conference on Information Retrieval (pp. 3-19). Cham: Springer Nature Switzerland.

Balloccu, G., Boratto, L., Cancedda, C., Fenu, G., & Marras, M. (2023). Faithful Path Language Modelling for Explainable Recommendation over Knowledge Graph. arXiv preprint arXiv:2310.16452.

-->

Presenters

Giacomo Balloccu

Giacomo Balloccu
University of Cagliari (Italy)

Ludovico Boratto

Ludovico Boratto
University of Cagliari (Italy)

Gianni Fenu

Gianni Fenu
University of Cagliari (Italy)

Francesca Maridina Malloci

Francesca Maridina Malloci
University of Cagliari (Italy)

Mirko Marras

Mirko Marras
University of Cagliari (Italy)

Registration

Registration to the tutorial will be managed by the ECIR 2024 main conference organization. Registration is open.

Contacts

Please, reaching out to us at giacomo.balloccu@unica.it, ludovico.boratto@acm.org, fenu@unica.it,francescam.malloci@unica.it ,and mirko.marras@acm.org for any request you might have.

Tutorial Explainable Recommender Systems with Knowledge Graphs and Language Models