Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Topic-Aware Knowledge Graph with Large Language Models for Interoperability in Recommender Systems

Minhye Jeon, Seokho Ahn, Young-Duk Seo

TL;DR

The paper tackles interoperability and data sparsity in knowledge-graph-based recommender systems by leveraging large language models to extract topic-centric representations from item side and context information. It introduces a three-step pipeline: general topic extraction from side and context, specific topic extraction from context, and a refining stage to normalize synonyms, all operating over a standardized metagraph to ensure cross-domain applicability. Empirical results on Amazon Beauty and Clothing show notable gains when context-derived topics are integrated with side information, especially with large-scale context (descriptions and reviews). The approach reduces reliance on domain experts and improves interoperability across systems while enhancing recommendation quality.

Abstract

The use of knowledge graphs in recommender systems has become one of the common approaches to addressing data sparsity and cold start problems. Recent advances in large language models (LLMs) offer new possibilities for processing side and context information within knowledge graphs. However, consistent integration across various systems remains challenging due to the need for domain expert intervention and differences in system characteristics. To address these issues, we propose a consistent approach that extracts both general and specific topics from both side and context information using LLMs. First, general topics are iteratively extracted and updated from side information. Then, specific topics are extracted using context information. Finally, to address synonymous topics generated during the specific topic extraction process, a refining algorithm processes and resolves these issues effectively. This approach allows general topics to capture broad knowledge across diverse item characteristics, while specific topics emphasize detailed attributes, providing a more comprehensive understanding of the semantic features of items and the preferences of users. Experimental results demonstrate significant improvements in recommendation performance across diverse knowledge graphs.

Topic-Aware Knowledge Graph with Large Language Models for Interoperability in Recommender Systems

TL;DR

The paper tackles interoperability and data sparsity in knowledge-graph-based recommender systems by leveraging large language models to extract topic-centric representations from item side and context information. It introduces a three-step pipeline: general topic extraction from side and context, specific topic extraction from context, and a refining stage to normalize synonyms, all operating over a standardized metagraph to ensure cross-domain applicability. Empirical results on Amazon Beauty and Clothing show notable gains when context-derived topics are integrated with side information, especially with large-scale context (descriptions and reviews). The approach reduces reliance on domain experts and improves interoperability across systems while enhancing recommendation quality.

Abstract

The use of knowledge graphs in recommender systems has become one of the common approaches to addressing data sparsity and cold start problems. Recent advances in large language models (LLMs) offer new possibilities for processing side and context information within knowledge graphs. However, consistent integration across various systems remains challenging due to the need for domain expert intervention and differences in system characteristics. To address these issues, we propose a consistent approach that extracts both general and specific topics from both side and context information using LLMs. First, general topics are iteratively extracted and updated from side information. Then, specific topics are extracted using context information. Finally, to address synonymous topics generated during the specific topic extraction process, a refining algorithm processes and resolves these issues effectively. This approach allows general topics to capture broad knowledge across diverse item characteristics, while specific topics emphasize detailed attributes, providing a more comprehensive understanding of the semantic features of items and the preferences of users. Experimental results demonstrate significant improvements in recommendation performance across diverse knowledge graphs.
Paper Structure (16 sections, 11 equations, 3 figures, 2 tables, 1 algorithm)

This paper contains 16 sections, 11 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Knowledge graph-based Recommender Systems
  4. LLM-based Knowledge graph Construction in Recommender Systems
  5. Preliminaries
  6. Theoretical Backgrounds
  7. Task Formulation
  8. Proposed Approach
  9. Approach Overview
  10. General topic extraction
  11. Specific topic extraction
  12. Topic Refining
  13. Experiments
  14. Experimental Settings
  15. Experimental Results
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Visualization of general and specific topic extraction from structured side information and unstructured context information.
  • Figure 2: Overview of our proposed approach. Each color represents an entity in the graph: yellow for items, green for side information, blue for context information, and purple for topics extracted from the context information. This figure shows a summary of the task prompt; the full prompt is provided in Figure \ref{['fig:prompt']}.
  • Figure 3: Examples of item information and user reviews used in the proposed approach and corresponding prompts.