FACE: A General Framework for Mapping Collaborative Filtering Embeddings into LLM Tokens

2025-10-17

AI Summary

Collaborative filtering (CF) embeddings lack semantic interpretability, hindering effective utilization by large language models (LLMs). To address this, we propose FACE—a novel framework that achieves semantic alignment of CF embeddings into the LLM’s token space without fine-tuning the LLM. FACE first decouples user- and item-specific structural information via a projection module; then employs a quantized autoencoder to map the embeddings onto discrete, semantically meaningful LLM tokens; finally enforces semantic consistency through contrastive learning. The resulting descriptors are highly interpretable, human-readable, and support model-agnostic, plug-and-play deployment. Extensive experiments on three real-world datasets demonstrate significant improvements in recommendation accuracy. Human evaluation further confirms that FACE-generated descriptors exhibit strong readability and semantic plausibility, validating both functional efficacy and linguistic coherence.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Humans and AI: Intelligent User Interfaces

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Recently, large language models (LLMs) have been explored for integration with collaborative filtering (CF)-based recommendation systems, which are crucial for personalizing user experiences. However, a key challenge is that LLMs struggle to interpret the latent, non-semantic embeddings produced by CF approaches, limiting recommendation effectiveness and further applications. To address this, we propose FACE, a general interpretable framework that maps CF embeddings into pre-trained LLM tokens. Specifically, we introduce a disentangled projection module to decompose CF embeddings into concept-specific vectors, followed by a quantized autoencoder to convert continuous embeddings into LLM tokens (descriptors). Then, we design a contrastive alignment objective to ensure that the tokens align with corresponding textual signals. Hence, the model-agnostic FACE framework achieves semantic alignment without fine-tuning LLMs and enhances recommendation performance by leveraging their pre-trained capabilities. Empirical results on three real-world recommendation datasets demonstrate performance improvements in benchmark models, with interpretability studies confirming the interpretability of the descriptors. Code is available in https://github.com/YixinRoll/FACE.

Problem

• Maps collaborative filtering embeddings into LLM tokens
• Enables semantic interpretation of non-semantic recommendation embeddings
• Improves recommendation performance without fine-tuning language models

Innovation

• Mapping CF embeddings into LLM tokens
• Using disentangled projection and quantized autoencoder
• Achieving semantic alignment without fine-tuning LLMs