Adaptive Candidate Retrieval with Dynamic Knowledge Graph Construction for Cold-Start Recommendation

ACL ARR 2026 January Submission10117 Authors

Keywords: Recommender Systems, Retrieval-augmented Generation

Abstract:

The cold-start problem remains a critical challenge in real-world recommender systems, as new items with limited interaction data or insufficient information are frequently introduced.
Despite recent advances leveraging external knowledge such as knowledge graphs (KGs) and large language models (LLMs), recommender systems still face challenges in practical environments:
(1) static KGs are costly to construct and maintain and quickly become outdated as catalogs evolve; and
(2) LLM-based methods are constrained by limited context windows, forcing reliance on pre-filtered candidate lists.
To address these limitations, we propose ColdRAG, a retrieval-augmented framework that dynamically constructs a knowledge graph from raw metadata, extracts entities and relations to construct an updatable structure, and introduces LLM-guided multi-hop reasoning at inference time to retrieve and rank candidates without relying on pre-filtered lists.
Experiments across multiple benchmarks show that ColdRAG consistently outperforms strong seven baselines.

Paper Type: Long

Research Area: Retrieval-Augmented Language Models

Research Area Keywords: Information Extraction,Information Retrieval and Text Mining,NLP Applications

Languages Studied: English

Submission Number: 10117