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.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is in collaborative filtering recommendation based on user - item interaction graphs, how to recommend items that users are truly interested in while reducing the recommendation of uninteresting items. Specifically, current graph - collaborative filtering methods face a "seesaw" dilemma: on the one hand, not all user - interacted items are equally attractive to users; on the other hand, ignoring those items that users have interacted with but are not interested in will lead to an incomplete representation of user interaction intentions, thereby reducing the recommendation ability of the model. To solve this dilemma, the paper proposes the Feedback - Reciprocal Graph Collaborative Filtering (FRGCF) method, aiming to improve the recommendation quality by differentiating users' interest feedback while maintaining the integrity of user interaction intentions. ### Main Contributions 1. **Point out the "seesaw" dilemma in current graph - collaborative filtering methods**: This dilemma has been overlooked in previous research, making it difficult to simultaneously consider complete user - intention modeling and recommend more items that truly attract users. 2. **Propose the FRGCF method**: This is a novel method that uses two sub - graphs with different intentions - the I&F graph (Interacted & Fascinated) and the I&U graph (Interacted & Unfascinated). FRGCF models the I&F graph and the I&U graph respectively through feedback - reciprocal contrastive learning and macro - feedback modeling, thereby learning and exchanging meaningful feedback information while avoiding the negative impact of I&U interactions. 3. **Experimental verification**: Extensive experiments show that FRGCF can outperform existing models on four benchmark datasets and one large - scale industrial dataset while reducing the recommendation of uninteresting items. Online A/B tests also verify the effectiveness of FRGCF in Taobao's recommendation system. ### Method Overview 1. **Feedback - Partitioned Graph Collaborative Filtering**: - Divide the entire interaction graph into the I&F graph and the I&U graph. - Perform graph convolution operations on the I&F graph and the I&U graph respectively to generate high - quality embedding representations. 2. **Feedback - Reciprocal Contrastive Learning**: - Connect the I&F graph and the I&U graph through the contrastive learning method, so that the I&F graph can obtain useful information from the I&U graph without being misled by its negative factors. 3. **Macro - Feedback Modeling**: - Model the I&F graph and the I&U graph from a macro perspective to enhance the comprehensiveness of the representation. - Infer the direction of node updates based on the differences between clustering and GNN layers, and aggregate macro - information accordingly. 4. **Optimization and Inference**: - Introduce distance regularization to optimize the decoupling between different graph - structure representations. - Train the model using the pairwise BPR loss function and optimize it in combination with multiple loss terms. - In the inference stage, use the model of the I&F graph to generate recommendation results. ### Experimental Results - **Offline Evaluation**: The Recall and NDCG metrics of FRGCF on all datasets are superior to existing methods, with average improvements of 6.28% and 6.37% respectively. In particular, for the large - scale industrial dataset, Recall and NDCG are improved by 6.55% and 7.82% respectively. - **Consider the Importance of the I&U graph**: The performance of the model trained only on the I&F graph drops significantly, indicating that simply removing all I&U interactions will lead to sub - optimal performance. Through these contributions and experimental results, the paper effectively solves the "seesaw" dilemma in graph - collaborative filtering and improves the performance of the recommendation system and user experience.

Feedback Reciprocal Graph Collaborative Filtering

Improved Metric-Based Recommender by Historical Interactions.

Improving Recommendations With Collaborative Factors

A Fast Interactive Item-Based Collaborative Filtering Algorithm

Enhanced Knowledge Transfer for Collaborative Filtering with Multi-Source Heterogeneous Feedbacks

RGCF: Refined Graph Convolution Collaborative Filtering with concise and expressive embedding

A Personalized Collaborative Filtering Recommendation System Based on Bi-Graph Embedding and Causal Reasoning

KGCFRec: Improving Collaborative Filtering Recommendation with Knowledge Graph

Dynamic Graph Collaborative Filtering

Online Collaborative Filtering With Implicit Feedback

On Manipulating Signals of User-Item Graph: A Jacobi Polynomial-based Graph Collaborative Filtering

Disentangled Graph Collaborative Filtering

Graph-ICF: Item-based collaborative filtering based on graph neural network

Hypergraph Contrastive Collaborative Filtering

Multi-Component Graph Convolutional Collaborative Filtering

Alleviating Behavior Data Imbalance for Multi-Behavior Graph Collaborative Filtering

Dual Channel Hypergraph Collaborative Filtering

Graph Collaborative Signals Denoising and Augmentation for Recommendation

A Review-based Feature-level Information Aggregation model for graph Collaborative Filtering

Cluster-based Graph Collaborative Filtering

Graph Heterogeneous Multi-Relational Recommendation