Abstract:Modern recommendation systems often create information cocoons, limiting users' exposure to diverse content. To enhance user experience, a crucial challenge is developing systems that can balance content exploration and exploitation, allowing users to adjust their recommendation preferences. Intuitively, this balance can be achieved through a tree-structured representation, where depth search facilitates exploitation and breadth search enables exploration. However, current works face two challenges to achieve this target: (1) Euclidean methods fail to fully capture hierarchical structures and lack flexibility in balancing exploration-exploitation, while (2) hyperbolic approaches, despite better hierarchical modeling, suffer from insufficient semantic alignment due to their reliance on Euclidean text encoders. To address these challenges, we propose HARec, a hyperbolic representation learning framework that jointly aligns user-item collaborative information with textual descriptions in hyperbolic space. Our framework introduces two key technique novelty: (1) a hierarchical-aware graph-llm alignment mechanism that enables better hierarchical representation, and (2) a hyperbolic hierarchical tree structure that facilitates user-adjustable exploration-exploitation trade-offs. Extensive experiments demonstrate that HARec consistently outperforms both Euclidean and hyperbolic baselines, achieving up to 5.49% improvement in utility metrics and 11.39% increase in diversity metrics.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the information cocoons phenomenon prevalent in modern recommendation systems, which restricts users from being exposed to diverse content. Specifically, the paper focuses on developing systems that can balance content exploration and exploitation, allowing users to adjust their recommendation preferences. Exploration refers to recommending new, unencountered content to users, while exploitation is to recommend similar content based on users' existing preferences. This balance is crucial for enhancing the user experience because over - emphasizing exploration may lead to recommended content not matching users' interests, and over - emphasizing exploitation may cause users to be trapped in information cocoons and unable to be exposed to new things. The paper points out that existing recommendation systems face two main challenges in achieving this goal: 1. **Euclidean methods**: These methods are insufficient in capturing hierarchical structures and lack flexibility in balancing exploration and exploitation. 2. **Hyperbolic methods**: Although they perform better in hierarchical structure modeling, these methods have insufficient semantic alignment due to their reliance on Euclidean text encoders, thus affecting the robustness and accuracy of the recommendation system. To solve these problems, the paper proposes **HARec** (Hyperbolic Graph - LLM Alignment for Exploration and Exploitation in Recommender Systems), which is a hyperbolic representation learning framework that enhances the performance of the recommendation system by jointly aligning user - item collaborative information and text descriptions in hyperbolic space. HARec introduces two key techniques: 1. **Hierarchy - aware graph - LLM alignment mechanism**: By jointly aligning text descriptions and user - item collaborative information in hyperbolic space, the quality of hierarchical representations is improved and noise sensitivity is reduced. 2. **Hyperbolic hierarchical tree structure**: Construct a hierarchical tree structure to support user - adjustable exploration - exploitation trade - offs, thereby better meeting users' personalized needs. Through extensive experimental verification, HARec outperforms existing Euclidean and hyperbolic baseline models on multiple metrics, especially achieving significant improvements in practicality and diversity. Specifically, HARec has increased by 5.49% on the practicality metric and 11.39% on the diversity metric.

HARec: Hyperbolic Graph-LLM Alignment for Exploration and Exploitation in Recommender Systems

Hyperbolic Neural Collaborative Recommender

HyperSoRec: Exploiting Hyperbolic User and Item Representations with Multiple Aspects for Social-aware Recommendation

HSR: Hyperbolic Social Recommender

Automatic Hypergraph Generation for Enhancing Recommendation with Sparse Optimization

A Hyperbolic-to-Hyperbolic User Representation with Multi-aspect for Social Recommendation

HCGR: Hyperbolic Contrastive Graph Representation Learning for Session-based Recommendation

Enhancing Recommendation with Automated Tag Taxonomy Construction in Hyperbolic Space

HRCF: Enhancing Collaborative Filtering Via Hyperbolic Geometric Regularization

Logical Relation Modeling and Mining in Hyperbolic Space for Recommendation

A Recommendation Method Based on Multi-Source Heterogeneous Hypergraphs and Contrastive Learning

HSA: Hyperbolic Self-attention for Sequential Recommendation

Llm-guided multi-view hypergraph learning for human-centric explainable recommendation

Learning User Representations with Hypercuboids for Recommender Systems

Hyperbolic Hypergraphs for Sequential Recommendation

Hyperbolic Representation Learning: Revisiting and Advancing

Exploiting High-Order Local and Global User-Item Interactions for Effective Recommendation

DaRec: A Disentangled Alignment Framework for Large Language Model and Recommender System

Representation Learning with Large Language Models for Recommendation

Hierarchical User Intention-Preference for Sequential Recommendation with Relation-Aware Heterogeneous Information Network Embedding