Feedback Reciprocal Graph Collaborative Filtering

TL;DR

This work tackles the seesaw dilemma in graph-based collaborative filtering by separating user-item interactions into Interacted & Fascinated (I&F) and Interacted & Unfascinated (I&U) graphs. It introduces FRGCF, a framework with dual GCF branches, feedback-reciprocal contrastive learning to connect the branches, and macro-level feedback modeling to mitigate data incompleteness. Across four benchmarks and a billion-scale Taobao dataset, FRGCF consistently outperforms state-of-the-art methods and demonstrates the ability to recommend more fascinating items while reducing unfascinated ones, with strong online A/B results in terms of PCTR, UCTR, GMV, and StayTime. The approach shows practical impact for industrial recommender systems by improving user engagement and satisfaction with minimal latency overhead.

Abstract

Collaborative filtering on user-item interaction graphs has achieved success in the industrial recommendation. However, recommending users' truly fascinated items poses a seesaw dilemma for collaborative filtering models learned from the interaction graph. On the one hand, not all items that users interact with are equally appealing. Some items are genuinely fascinating to users, while others are unfascinated. Training graph collaborative filtering models in the absence of distinction between them can lead to the recommendation of unfascinating items to users. On the other hand, disregarding the interacted but unfascinating items during graph collaborative filtering will result in an incomplete representation of users' interaction intent, leading to a decline in the model's recommendation capabilities. To address this seesaw problem, we propose Feedback Reciprocal Graph Collaborative Filtering (FRGCF), which emphasizes the recommendation of fascinating items while attenuating the recommendation of unfascinating items. Specifically, FRGCF first partitions the entire interaction graph into the Interacted & Fascinated (I&F) graph and the Interacted & Unfascinated (I&U) graph based on the user feedback. Then, FRGCF introduces separate collaborative filtering on the I&F graph and the I&U graph with feedback-reciprocal contrastive learning and macro-level feedback modeling. This enables the I&F graph recommender to learn multi-grained interaction characteristics from the I&U graph without being misdirected by it. Extensive experiments on four benchmark datasets and a billion-scale industrial dataset demonstrate that FRGCF improves the performance by recommending more fascinating items and fewer unfascinating items. Besides, online A/B tests on Taobao's recommender system verify the superiority of FRGCF.

Figures (3)

• Figure 1: (a) The Interacted & Fascinated (I&F) and Interacted & Unfascinated (I&U) examples. (b) The recommended unfascinated items of GCF models between trained on the entire graph and the I&F graph. (c) The recommendation performance difference of GCF models between trained on the entire graph and the I&F graph.
• Figure 2: The overall framework of our FRGCF, which contains three main components: (i) Feedback-partitioned graph collaborative filtering to train GCF models for I&F graph and I&U graph, respectively, and (ii) Feedback-reciprocal contrastive learning to connect the I&F GCF model and the I&U GCF model. (iii) Macro-level feedback modeling to enhance the representation of I&F and I&U. (iv) Finally, the inference predictions are computed by the I&F graph convolution model.
• Figure 3: Ablation study results on three FRGCF variants.