Personalized LLM based Information Retrieval/Recommendation on Textual and Relational Knowledge Bases

Answering real-world complex queries, such as complex product search, often requires accurate retrieval from semi-structured knowledge bases that involve blend of unstructured (e.g., textual descriptions of products) and structured (e.g., entity relations of products) information. However, Prior works focused on either purely textual queries on unstructured knowledge or structured SQL or knowledge graph queries, which are limited in the span of knowledge and inadequate to study the complexities of retrieval or information recommendation. Recently, large language models (LLMs) have demonstrated significant potential on information retrieval tasks. Nevertheless, the existing works mainly focus mainly on general knowledge, e.g., from Wikipedia. However, the knowledge may commonly come from private sources, requiring retrieval systems to operate on private knowledge bases. Therefore, there is a gap of how current LLM retrieval systems handle the complex textual and relational requirements in queries that can involve private knowledge. 

[1] Wu et al., STARK: Benchmarking LLM Retrieval on Textual and Relational Knowledge Bases, 38th Conference on Neural Information Processing Systems (NeurIPS 2024) Track on Datasets and Benchmarks.