Retrieving More Concrete Product Reviews by Query Rewriting with Retrieved Review Concretization

This chapter delves into the challenge of retrieving concrete product reviews from abstract ones, a task that is often hindered by the lack of effective query formulation. The authors propose a method called Query Rewriting with Retrieved Review Concretization (QR-ReReC), which uses a large language model (LLM) to rewrite abstract reviews into more concrete queries. The process involves three stages: Seed Review Retrieval, Query Rewriting, and Concrete Review Retrieval. In the first stage, an initial query is used to retrieve seed reviews from a corpus. These seed reviews, along with the original abstract review, are then fed into the LLM to generate pseudo-concrete reviews. These pseudo-concrete reviews are used to derive a rewritten query, which is then used to retrieve more concrete reviews from the corpus. The effectiveness of QR-ReReC is evaluated through experiments using a dataset of 50 abstract reviews. The results show that QR-ReReC outperforms baseline methods, including those using pseudo relevance feedback and LLM-based query rewriting without retrieved review concretization. The chapter also discusses the experimental setup, including the use of sparse and dense retrieval models, and the evaluation metrics used. The findings highlight the potential of QR-ReReC to improve the retrieval of concrete product reviews, offering valuable insights for professionals in the field of natural language processing and e-commerce.