Document Expansion by Query Prediction

TL;DR

This paper introduces a method that predicts potential user queries for documents and expands the documents with these queries to improve retrieval effectiveness, achieving state-of-the-art results efficiently.

Contribution

It presents a novel query prediction-based document expansion technique using sequence-to-sequence models for improved retrieval performance.

Findings

Achieves state-of-the-art results in two retrieval tasks.

Retrieval alone approaches the effectiveness of neural re-rankers in latency-critical settings.

Method is faster and maintains high effectiveness without re-ranking.

Abstract

One technique to improve the retrieval effectiveness of a search engine is to expand documents with terms that are related or representative of the documents' content.From the perspective of a question answering system, this might comprise questions the document can potentially answer. Following this observation, we propose a simple method that predicts which queries will be issued for a given document and then expands it with those predictions with a vanilla sequence-to-sequence model, trained using datasets consisting of pairs of query and relevant documents. By combining our method with a highly-effective re-ranking component, we achieve the state of the art in two retrieval tasks. In a latency-critical regime, retrieval results alone (without re-ranking) approach the effectiveness of more computationally expensive neural re-rankers but are much faster.