Leveraging Language Semantics for Collaborative Filtering with TextGCN and TextGCN-MLP: Zero-Shot vs In-Domain Performance

2025-10-14

AI Summary

This paper addresses the challenge of simultaneously achieving strong zero-shot and in-domain recommendation performance in collaborative filtering. We propose two architectures: TextGCN and TextGCN-MLP. TextGCN is the first parameter-free graph convolutional model directly applied to LLM-generated item title embeddings, propagating semantic information over the user–item interaction graph to significantly improve zero-shot recommendation. TextGCN-MLP extends this with a trainable MLP and contrastive learning module, enabling semantic-guided domain specialization and achieving state-of-the-art performance in supervised settings. Both models jointly encode linguistic semantics and collaborative interaction structure, revealing an inherent trade-off between semantic representation quality for zero-shot generalization and domain-specific adaptation. Our open-source implementation empirically validates the effectiveness and strong generalization capability of semantic-driven recommendation.

Technology Category

• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In recent years, various approaches have been proposed to leverage large language models (LLMs) for incorporating textual information about items into recommender systems. Existing methods primarily focus on either fine-tuning LLMs to generate recommendations or integrating LLM-based embeddings into downstream models. In this work, we follow the latter direction and propose extbf{TextGCN}, which applies parameter-free graph convolution layers directly over LLM-based item-title embeddings, instead of learning ID-based embeddings as in traditional methods. By combining language semantics with graph message passing, this architecture achieves state-of-the-art zero-shot performance, significantly outperforming prior approaches. Furthermore, we introduce extbf{TextGCN-MLP}, which extends TextGCN with a trainable multilayer perceptron trained using a contrastive loss, achieving state-of-the-art in-domain performance on recommendation benchmarks. However, the zero-shot performance of TextGCN-MLP remains lower than that of TextGCN, highlighting the trade-off between in-domain specialization and zero-shot generalization. We release our code on github at href{https://github.com/ChernovAndrey/TFCE}{github.com/ChernovAndrey/TFCE}.

Problem

• Incorporating language semantics into collaborative filtering systems
• Improving zero-shot recommendation performance using graph convolutions
• Balancing in-domain specialization with zero-shot generalization trade-offs

Innovation

• TextGCN applies graph convolution on LLM embeddings
• TextGCN-MLP adds contrastive MLP for in-domain performance
• Method combines language semantics with graph message passing