Publications

You can also find my articles on my Google Scholar profile.

The Limits of Graph Samplers for Training Inductive Recommender Systems

Published in VLDB: Very Large Data Bases, 2025

We investigate graph-based recommender systems using sampling techniques to reduce training time while maintaining performance. Experiments show that training with 50% of the data can cut training time by up to 86% with minimal loss, though further reduction harms performance.

Recommended citation: T.E. Jendal, M. Lissandrini, P. Dolog, K. Hose, "The Limits of Graph Samplers for Training Inductive Recommender Systems," in VLDB, 2025
Download Paper

Handling New Users and Items: A Comparative Study of Inductive Recommenders

Published in Data Mining and Knowledge Discovery, 2025

We propose a framework to evaluate inductive recommender systems on temporal splits across three real-world datasets. Our re-evaluation shows that aggregation-based methods consistently outperform non-aggregating ones, though performance varies greatly across settings.

Recommended citation: T.E. Jendal, M. Lissandrini, P. Dolog, K. Hose, "Handling new users and items: a comparative study of inductive recommenders," in Data Min. Knowl. Disc., 2025
Download Paper

Hypergraphs with Attention on Reviews for Explainable Recommendation

Published in ECIR: European Conference on Information Retrieval, 2024

We propose a novel review-specific Hypergraph (HG) model to capture high-order connections in review data and introduce a model-agnostic explainability module to clarify recommendations. Our experiments confirm the HG model's effectiveness in providing explanations on real-world datasets.

Recommended citation: T.E. Jendal, TH. Le, H.W. Lauw, M. Lissandrini, P. Dolog, K. Hose, "Hypergraphs with Attention on Reviews for Explainable Recommendation," in ECIR, 2024
Download Paper

Knowledge graph embeddings: open challenges and opportunities

Published in Transactions on Graph Data and Knowledge, 2023

This study surveys advances and challenges in Knowledge Graph embeddings, highlighting the focus on structural information and recent progress in incorporating semantics, multi-modal, temporal, and multilingual features, while also addressing limitations such as neglecting real-world applications, handling dynamic changes, and data biases.

Recommended citation: R. Biswas, L.A. Kaffee, M. Cochez, S. Dumbrava, T.E. Jendal, et al.. "Knowledge graph embeddings: open challenges and opportunities," in TGDK, 2023
Download Paper

GInRec: A Gated Architecture for Inductive Recommendation using Knowledge Graphs

Published in Proceedings of the Fifth Knowledge-aware and Conversational Recommender Systems Workshop (KaRS), 2023

We propose GInRec, a state-of-the-art method using a graph neural network with relation-specific gates and a knowledge graph, to provide better recommendations for new users and items, outperforming existing inductive methods. Our work reevaluates state-of-the-art methods, identifies improved evaluation protocols, and highlights flaws in previous conclusions.

Recommended citation: T. E. Jendal, M. Lissandrini, P. Dolog, and K. Hose, “GInRec: A gated architecture for inductive recommendation using knowledge graphs,” in KaRS, 2023.
Download Paper

MindReader: Recommendation over Knowledge Graph Entities with Explicit User Ratings

Published in CIKM: Proceedings of the 29th ACM International Conference on Information & Knowledge Management, 2020

We introduce the MindReader dataset, providing explicit user ratings for both items and KG entities, collected from over 1,174 users in the movie domain. Our study demonstrates that incorporating non-item ratings enhances recommendation quality in various models, highlighting the potential of using explicit ratings on KG entities for improving both warm and cold-start recommendations.

Recommended citation: A. H. Brams*, A. L. Jakobsen*, T. E. Jendal*, M. Lissandrini, P. Dolog, and K. Hose, “Mindreader: Recommendation over knowledge graph entities with explicit user ratings,” in CIKM, 2020
* Equal contribution
Download Paper | Download Dataset

Theis E. Jendal