QAGCF: Graph Collaborative Filtering for Q&A Recommendation

TL;DR

This paper tackles the challenge of Q&A recommendation by addressing two entanglement phenomena: collaborative information entanglement between questions and answers, and semantic information entanglement within and across Q&A pairs. It introduces QAGCF, a multi-view graph neural network that builds separate collaborative (U–Q, U–A) and semantic (Q–A, Q–Q) graphs and then merges them into a global graph. To handle heterophily and rich semantic structure, the method employs Jacobi polynomial-based graph filters to decompose graph signals into low-, mid-, and high-frequency components, along with a contrastive learning objective to produce robust embeddings; predictions combine user–answer and user–question interactions via a BPR loss framework. Experiments on ZhihuRec and a Commercial Q&A dataset show state-of-the-art performance, with ablations confirming the necessity of each component, including the dual-view graph construction, polynomial filters, and contrastive learning. The approach offers a principled, scalable way to improve Q&A recommendations and has practical impact for real-world knowledge platforms.

Abstract

Question and answer (Q&A) platforms usually recommend question-answer pairs to meet users' knowledge acquisition needs, unlike traditional recommendations that recommend only one item. This makes user behaviors more complex, and presents two challenges for Q&A recommendation, including: the collaborative information entanglement, which means user feedback is influenced by either the question or the answer; and the semantic information entanglement, where questions are correlated with their corresponding answers, and correlations also exist among different question-answer pairs. Traditional recommendation methods treat the question-answer pair as a whole or only consider the answer as a single item, which overlooks the two challenges and cannot effectively model user interests. To address these challenges, we introduce Question & Answer Graph Collaborative Filtering (QAGCF), a graph neural network model that creates separate graphs for collaborative and semantic views to disentangle the information in question-answer pairs. The collaborative view disentangles questions and answers to individually model collaborative information, while the semantic view captures the semantic information both within and between question-answer pairs. These views are further merged into a global graph to integrate the collaborative and semantic information. Polynomial-based graph filters are used to address the high heterophily issues of the global graph. Additionally, contrastive learning is utilized to obtain robust embeddings during training. Extensive experiments on industrial and public datasets demonstrate that QAGCF consistently outperforms baselines and achieves state-of-the-art results.

Figures (9)

• Figure 1: An illustration of Q&A Recommendation. When users access the Q&A platform, the recommender system provides them with a list of recommended Q&A pairs. Users can interact with Q&A pairs they are interested in.
• Figure 2: Analysis of the two challenges on ZhihuRec. (a): We filter out questions that have been exposed more than 10 times and have more than 10 answers. (b): We analyzed the similarity in four groups: wqa (within question-answer pairs), bqa (between questions and unrelated answers), weq (within clicked questions), beq (between clicked questions and unclicked ones)
• Figure 3: The overall framework of QAGCF consists of four modules: (1) Construction of Collaborative View and semantic View, (2) Band-stop and Band-pass Filters for processing information in different frequency bands of the Global Graph, (3) Contrastive Learning for noise-based embedding augmentation, (4) Prediction module based on the learned embeddings.
• Figure 4: Correlation between eigenvalues $\mathbf{\Lambda}$ of $\hat{\mathbf{A}}_{\mathrm{train}}$ and diagonal of $\mathbf{U^\top \mathbf{A}_{\mathrm{test}}U}$ on two Q&A recommendation datasets.
• Figure 5: Validation of QAGCF's Disentanglement of Collaborative Information. Logarithmic operation is applied for better visualization.