Beyond Questions: Leveraging ColBERT for Keyphrase Search

TL;DR

This paper enhances document ranking for keyphrase queries by adapting the ColBERT architecture, utilizing large language models to convert question-like queries into keyphrases, and training specialized encoders to improve retrieval performance.

Contribution

It introduces a novel keyphrase-based ColBERT model and explores cost-effective training methods for improved keyphrase search performance.

Findings

Late interaction architecture benefits keyphrase search

Language models can effectively convert questions to keyphrases

Training only the query encoder is feasible and efficient

Abstract

While question-like queries are gaining popularity and search engines' users increasingly adopt them, keyphrase search has traditionally been the cornerstone of web search. This query type is also prevalent in specialised search tasks such as academic or professional search, where experts rely on keyphrases to articulate their information needs. However, current dense retrieval models often fail with keyphrase-like queries, primarily because they are mostly trained on question-like ones. This paper introduces a novel model that employs the ColBERT architecture to enhance document ranking for keyphrase queries. For that, given the lack of large keyphrase-based retrieval datasets, we first explore how Large Language Models can convert question-like queries into keyphrase format. Then, using those keyphrases, we train a keyphrase-based ColBERT ranker (ColBERTKP_QD) to improve the performance when working with keyphrase queries. Furthermore, to reduce the training costs associated with training the full ColBERT model, we investigate the feasibility of training only a keyphrase query encoder while keeping the document encoder weights static (ColBERTKP_Q). We assess our proposals' ranking performance using both automatically generated and manually annotated keyphrases. Our results reveal the potential of the late interaction architecture when working under the keyphrase search scenario.