Bursting Filter Bubble: Enhancing Serendipity Recommendations with Aligned Large Language Models

2025-02-19

AI Summary

Recommendation systems often suffer from feedback loops that foster filter bubbles, leading to content homogenization and diminished user satisfaction. To address this, we propose SERAL—a framework for low-latency, high-fidelity serendipitous recommendation. First, we design a multi-granularity user behavior compression mechanism to construct lightweight cognitive user profiles. Second, we introduce SerenGPT, a preference-aligned knowledge distillation paradigm that enables large language models (LLMs) to accurately model human serendipity perception. Third, we develop a nearline adaptive inference architecture to meet industrial-scale latency constraints. Extensive experiments on Taobao’s “You May Like” production system demonstrate significant improvements: +5.7% serendipitous item exposure rate (PVR), +29.56% click-through rate (CTR), and +27.6% gross merchandise volume (GMV). This work represents the first integration of LLM-based serendipity perception modeling with industrial nearline deployment, establishing a novel paradigm for trustworthy recommendation.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Search and Optimization: Learning to Search
• 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
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recommender systems (RSs) often suffer from the feedback loop phenomenon, e.g., RSs are trained on data biased by their recommendations. This leads to the filter bubble effect that reinforces homogeneous content and reduces user satisfaction. To this end, serendipity recommendations, which offer unexpected yet relevant items, are proposed. Recently, large language models (LLMs) have shown potential in serendipity prediction due to their extensive world knowledge and reasoning capabilities. However, they still face challenges in aligning serendipity judgments with human assessments, handling long user behavior sequences, and meeting the latency requirements of industrial RSs. To address these issues, we propose SERAL (Serendipity Recommendations with Aligned Large Language Models), a framework comprising three stages: (1) Cognition Profile Generation to compress user behavior into multi-level profiles; (2) SerenGPT Alignment to align serendipity judgments with human preferences using enriched training data; and (3) Nearline Adaptation to integrate SerenGPT into industrial RSs pipelines efficiently. Online experiments demonstrate that SERAL improves exposure ratio (PVR), clicks, and transactions of serendipitous items by 5.7%, 29.56%, and 27.6%, enhancing user experience without much impact on overall revenue. Now, it has been fully deployed in the"Guess What You Like"of the Taobao App homepage.

Problem

• Filter bubble effect in recommender systems
• Aligning serendipity judgments with human preferences
• Efficient integration into industrial RS pipelines

Innovation

• Compresses user behavior into profiles
• Aligns LLMs with human preferences
• Integrates SerenGPT into RS pipelines