Items Proxy Bridging: Enabling Frictionless Critiquing in Knowledge Graph Recommendations

TL;DR

This paper addresses the challenge of frictionless, critique-driven refinement in knowledge graph recommender systems built on collaborative filtering, where continual critiques risk catastrophic forgetting and require redesigned user–keyphrase correlations. It introduces IPGC, a generic Items Proxy Generic Critiquing framework that uses an items proxy to translate user–keyphrase critiques into user–item interactions, enabling seamless integration with existing CF-based KG models. A Bayesian posterior update of the user embedding is performed with a critiquing likelihood approximated through the items proxy, complemented by a gradient-based anti-forgetting regularizer to preserve core knowledge across rounds. Empirical results on MovieLens-1M and Last-FM across KGAT, KGIN, and DiffKG demonstrate strong, generalizable improvements in critiquing effectiveness and robustness to multi-step interactions, validating IPGC as a practical plug-in for frictionless iterative recommendations.

Abstract

Modern recommender systems place great inclination towards facilitating user experience, as more applications enabling users to critique and then refine recommendations immediately. Considering the real-time requirements, critique-able recommender systems typically straight modify the model parameters and update the recommend list through analyzing the user critiquing keyphrases in the inference phase. Current critiquing methods require first constructing a specially designated model which establish direct correlations between users and keyphrases during the training phase allowing for innovative recommendations upon the critiquing,restricting the applicable scenarios. Additionally, all these approaches ignore the catastrophic forgetting problem, where the cumulative changes in parameters during continuous multi-step critiquing may lead to a collapse in model performance. Thus, We conceptualize a proxy bridging users and keyphrases, proposing a streamlined yet potent Items Proxy Generic Critiquing Framework (IPGC) framework, which can serve as a universal plugin for most knowledge graph recommender models based on collaborative filtering (CF) strategies. IPGC provides a new paradigm for frictionless integration of critique mechanisms to enable iterative recommendation refinement in mainstream recommendation scenarios. IPGC describes the items proxy mechanism for transforming the critiquing optimization objective of user-keyphrase pairs into user-item pairs, adapting it for general CF recommender models without the necessity of specifically designed user-keyphrase correlation module. Furthermore, an anti-forgetting regularizer is introduced in order to efficiently mitigate the catastrophic forgetting problem of the model as a prior for critiquing optimization.

Figures (10)

• Figure 1: (a) Illustration of continuous critiquing interactions (b) Alternative critiquing keyphrases available from the video app.
• Figure 2: Illustration of the proposed IPGC, which presents the whole procedure of how the framework functions in the KG recommender system to allow critiquing for users.
• Figure 3: Comparison of IPGC and baseline methods on KGAT for NDCG@K, Recall@K and HR@K, with dataset MovieLens and Last-FM.
• Figure 4: Comparison of IPGC and baseline methods on KGIN for NDCG@K, Recall@K and HR@K, with dataset MovieLens and Last-FM.
• Figure 5: Comparison of IPGC and baseline methods on DiffKG for NDCG@K, Recall@K and HR@K, with dataset MovieLens and Last-FM.