Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph Reasoning for Explainable Cold Start Recommendation

Jibril Frej, Marta Knezevic, Tanja Kaser

TL;DR

This work tackles cold-start recommendation by leveraging Graph Reasoning on Knowledge Graphs to produce explainable item suggestions via explicit user-to-item paths. It introduces GRECS, which adapts GR methods (PGPR and UPGPR) to cold-start scenarios by integrating cold entities into the KG and deriving embeddings from neighboring nodes, enabling path-based recommendations without retraining. Across five public datasets, GRECS achieves strong performance for strict cold-start users and cold-start items, reduces popularity bias, and benefits from using a small set of known relations. The study highlights the importance of embedding initialization and reward design in GR-based cold-start recommendations and points to extensions with other GR approaches and LLM-based methods for future work.

Abstract

The cold start problem, where new users or items have no interaction history, remains a critical challenge in recommender systems (RS). A common solution involves using Knowledge Graphs (KG) to train entity embeddings or Graph Neural Networks (GNNs). Since KGs incorporate auxiliary data and not just user/item interactions, these methods can make relevant recommendations for cold users or items. Graph Reasoning (GR) methods, however, find paths from users to items to recommend using relations in the KG and, in the context of RS, have been used for interpretability. In this study, we propose GRECS: a framework for adapting GR to cold start recommendations. By utilizing explicit paths starting for users rather than relying only on entity embeddings, GRECS can find items corresponding to users' preferences by navigating the graph, even when limited information about users is available. Our experiments show that GRECS mitigates the cold start problem and outperforms competitive baselines across 5 standard datasets while being explainable. This study highlights the potential of GR for developing explainable recommender systems better suited for managing cold users and items.

Graph Reasoning for Explainable Cold Start Recommendation

TL;DR

This work tackles cold-start recommendation by leveraging Graph Reasoning on Knowledge Graphs to produce explainable item suggestions via explicit user-to-item paths. It introduces GRECS, which adapts GR methods (PGPR and UPGPR) to cold-start scenarios by integrating cold entities into the KG and deriving embeddings from neighboring nodes, enabling path-based recommendations without retraining. Across five public datasets, GRECS achieves strong performance for strict cold-start users and cold-start items, reduces popularity bias, and benefits from using a small set of known relations. The study highlights the importance of embedding initialization and reward design in GR-based cold-start recommendations and points to extensions with other GR approaches and LLM-based methods for future work.

Abstract

The cold start problem, where new users or items have no interaction history, remains a critical challenge in recommender systems (RS). A common solution involves using Knowledge Graphs (KG) to train entity embeddings or Graph Neural Networks (GNNs). Since KGs incorporate auxiliary data and not just user/item interactions, these methods can make relevant recommendations for cold users or items. Graph Reasoning (GR) methods, however, find paths from users to items to recommend using relations in the KG and, in the context of RS, have been used for interpretability. In this study, we propose GRECS: a framework for adapting GR to cold start recommendations. By utilizing explicit paths starting for users rather than relying only on entity embeddings, GRECS can find items corresponding to users' preferences by navigating the graph, even when limited information about users is available. Our experiments show that GRECS mitigates the cold start problem and outperforms competitive baselines across 5 standard datasets while being explainable. This study highlights the potential of GR for developing explainable recommender systems better suited for managing cold users and items.
Paper Structure (16 sections, 5 equations, 3 figures, 12 tables)

This paper contains 16 sections, 5 equations, 3 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Methodology
  4. Problem Formulation
  5. Graph Reasoning Models
  6. Graph Reasoning for Explainable Cold Start Recommendation
  7. Datasets
  8. Results
  9. Exp. 1: GRECS outperforms baselines on strict cold start users
  10. Exp. 2: GRECS achieves high coverage of cold start items
  11. Exp. 3: GRECS mitigates the popularity bias
  12. Exp. 4: GRECS requires few relations to provide accurate recommendations
  13. Discussion & Conclusion
  14. Datasets Statistics
  15. Results on all users
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Illustration of the KGRE-REC problem on a KG with a warm user $u_1$ (in red), who has already purchased a product and a cold user $u_2$ (in blue) who has not purchased any product, but whose preferences are assumed to be known (e.g. through a questionnaire). The goal is to find paths from the users to the items to be recommended. The red and blue arrows indicate the paths chosen for the recommendations. While the agent uses past interactions for the warm start user (purchased), only non-interaction relations are leveraged to provide a recommendation for the cold start user. All recommendations are explainable as they use entities and relations from the KG.
  • Figure 2: Evolution of the HR against the number of training interactions. The shaded area represents the standard deviation across 3 runs.
  • Figure 3: Evolution of the HR against the number of relations for strict cold start users. The shaded area represents the standard deviation across 3 runs.