Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EasyRec: Simple yet Effective Language Models for Recommendation

Xubin Ren, Chao Huang

TL;DR

EasyRec addresses zero-shot generalization gaps in ID-based collaborative filtering by integrating text-based semantic profiles with collaborative signals through a contrastive, alignment-centric framework. It constructs user/item textual profiles via LLMs, embeds them with a bidirectional transformer, and optimizes with a contrastive loss plus MLM regularization, achieving strong zero-shot and text-enhanced CF performance. The approach scales across parameter sizes (100–400M) with fast inference (~0.01s), and benefits further from profile diversification that improves generalization and diversity, revealing a scaling law across model and data size. By acting as a plug-and-play component, EasyRec enhances existing CF systems and demonstrates rapid adaptation to evolving user preferences without retraining, offering practical impact for scalable, generalizable recommender deployments.

Abstract

Deep neural networks have emerged as a powerful technique for learning representations from user-item interaction data in collaborative filtering (CF) for recommender systems. However, many existing methods heavily rely on unique user and item IDs, which restricts their performance in zero-shot learning scenarios. Inspired by the success of language models (LMs) and their robust generalization capabilities, we pose the question: How can we leverage language models to enhance recommender systems? We propose EasyRec, an effective approach that integrates text-based semantic understanding with collaborative signals. EasyRec employs a text-behavior alignment framework that combines contrastive learning with collaborative language model tuning. This ensures strong alignment between text-enhanced semantic representations and collaborative behavior information. Extensive evaluations across diverse datasets show EasyRec significantly outperforms state-of-the-art models, particularly in text-based zero-shot recommendation. EasyRec functions as a plug-and-play component that integrates seamlessly into collaborative filtering frameworks. This empowers existing systems with improved performance and adaptability to user preferences. Implementation codes are publicly available at: https://github.com/HKUDS/EasyRec.

EasyRec: Simple yet Effective Language Models for Recommendation

TL;DR

EasyRec addresses zero-shot generalization gaps in ID-based collaborative filtering by integrating text-based semantic profiles with collaborative signals through a contrastive, alignment-centric framework. It constructs user/item textual profiles via LLMs, embeds them with a bidirectional transformer, and optimizes with a contrastive loss plus MLM regularization, achieving strong zero-shot and text-enhanced CF performance. The approach scales across parameter sizes (100–400M) with fast inference (~0.01s), and benefits further from profile diversification that improves generalization and diversity, revealing a scaling law across model and data size. By acting as a plug-and-play component, EasyRec enhances existing CF systems and demonstrates rapid adaptation to evolving user preferences without retraining, offering practical impact for scalable, generalizable recommender deployments.

Abstract

Deep neural networks have emerged as a powerful technique for learning representations from user-item interaction data in collaborative filtering (CF) for recommender systems. However, many existing methods heavily rely on unique user and item IDs, which restricts their performance in zero-shot learning scenarios. Inspired by the success of language models (LMs) and their robust generalization capabilities, we pose the question: How can we leverage language models to enhance recommender systems? We propose EasyRec, an effective approach that integrates text-based semantic understanding with collaborative signals. EasyRec employs a text-behavior alignment framework that combines contrastive learning with collaborative language model tuning. This ensures strong alignment between text-enhanced semantic representations and collaborative behavior information. Extensive evaluations across diverse datasets show EasyRec significantly outperforms state-of-the-art models, particularly in text-based zero-shot recommendation. EasyRec functions as a plug-and-play component that integrates seamlessly into collaborative filtering frameworks. This empowers existing systems with improved performance and adaptability to user preferences. Implementation codes are publicly available at: https://github.com/HKUDS/EasyRec.
Paper Structure (40 sections, 12 equations, 7 figures, 11 tables)

This paper contains 40 sections, 12 equations, 7 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Methodology
  4. Collaborative User and Item Profiling
  5. Item Profiling
  6. User Profiling
  7. Advantages of Collaborative Profiling
  8. Profile Embedder with Collaborative LM
  9. Bidirectional Transformer Embedder
  10. Contrastive Collaborative LM Learning
  11. Augmentation with Profile Diversification
  12. Evaluation
  13. Experimental Settings
  14. Datasets
  15. Evaluation Protocols
  16. ...and 25 more sections

Figures (7)

  • Figure 1: EasyRec outperforms state-of-the-art language models in text-based zero-shot recommendation.
  • Figure 2: The overall framework of our proposed collaborative information-guided language model EasyRec.
  • Figure 3: Contrastive tuning of the collaborative LM enables it to learn rich representations. It aligns the text-based semantic space with global collaborative signals.
  • Figure 4: Performance w.r.t. data size. "Augmentation Count" indicates the number $t$ of diversified profiles.
  • Figure 5: Case study on handling user preference shift.
  • ...and 2 more figures