Incorporate LLMs with Influential Recommender System

Mingze Wang,Shuxian Bi,Wenjie Wang,Chongming Gao,Yangyang Li,Fuli Feng

DOI: https://doi.org/10.48550/arXiv.2409.04827

2024-09-07

Abstract:Recommender systems have achieved increasing accuracy over the years. However, this precision often leads users to narrow their interests, resulting in issues such as limited diversity and the creation of echo chambers. Current research addresses these challenges through proactive recommender systems by recommending a sequence of items (called influence path) to guide user interest in the target item. However, existing methods struggle to construct a coherent influence path that builds up with items the user is likely to enjoy. In this paper, we leverage the Large Language Model's (LLMs) exceptional ability for path planning and instruction following, introducing a novel approach named LLM-based Influence Path Planning (LLM-IPP). Our approach maintains coherence between consecutive recommendations and enhances user acceptability of the recommended items. To evaluate LLM-IPP, we implement various user simulators and metrics to measure user acceptability and path coherence. Experimental results demonstrate that LLM-IPP significantly outperforms traditional proactive recommender systems. This study pioneers the integration of LLMs into proactive recommender systems, offering a reliable and user-engaging methodology for future recommendation technologies.

Information Retrieval

What problem does this paper attempt to address?

This paper attempts to solve the problems of narrow user interests and the echo - chamber effect in recommendation systems due to over - precise recommendations. Specifically, traditional recommendation systems recommend content that matches users' interests by analyzing their browsing histories, but this approach often causes users to be trapped in a cycle of continuously repeating the recommendation of known - interest content, limiting the scope of users' interest exploration. To overcome this challenge, researchers have proposed Proactive Recommender Systems, aiming to guide users' interests to shift towards target items by recommending a series of items (referred to as influence paths). However, existing methods have difficulties in constructing coherent influence paths and cannot ensure that each item in the path can attract users. For this reason, this paper proposes an Influence Path Planning method based on Large Language Models (LLMs), namely LLM - IPP. This method utilizes the powerful instruction - execution and path - planning capabilities of LLMs, can generate coherent influence paths containing target items, and at the same time maintain the relevance of adjacent items in the path, thereby increasing users' acceptance and participation. Verified by experiments, LLM - IPP is significantly superior to traditional methods in terms of path coherence and user acceptance, providing a reliable and user - friendly solution for future recommendation technologies.

Incorporate LLMs with Influential Recommender System

LLMs for User Interest Exploration in Large-scale Recommendation Systems

All Roads Lead to Rome: Unveiling the Trajectory of Recommender Systems Across the LLM Era

Recommender Systems in the Era of Large Language Models (LLMs)

Large Language Models for Recommendation: Progresses and Future Directions

LLaRA: Aligning Large Language Models with Sequential Recommenders.

Aligning Large Language Models for Controllable Recommendations

Enhancing High-order Interaction Awareness in LLM-based Recommender Model

Mitigating Propensity Bias of Large Language Models for Recommender Systems

HLLM: Enhancing Sequential Recommendations via Hierarchical Large Language Models for Item and User Modeling

Large Language Model Can Interpret Latent Space of Sequential Recommender

Personalized Recommendation Systems Powered By Large Language Models: Integrating Semantic Understanding and User Preferences

Empowering Few-Shot Recommender Systems With Large Language Models-Enhanced Representations

Adapting Large Language Models by Integrating Collaborative Semantics for Recommendation

LLM-enhanced Reranking in Recommender Systems

LLMRG: Improving Recommendations through Large Language Model Reasoning Graphs

Large Language Models Enhanced Collaborative Filtering

Empowering Few-Shot Recommender Systems with Large Language Models -- Enhanced Representations

Towards Next-Generation LLM-based Recommender Systems: A Survey and Beyond

MMREC: LLM Based Multi-Modal Recommender System

Large Language Models for Recommendation: Past, Present, and Future