You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Formulating selective information needs results in queries that implicitlyspecify set operations, such as intersection, union, and difference. Forinstance, one might search for "shorebirds that are not sandpipers" or"science-fiction films shot in England". To study the ability of retrievalsystems to meet such information needs, we construct QUEST, a dataset of 3357natural language queries with implicit set operations, that map to a set ofentities corresponding to Wikipedia documents. The dataset challenges models tomatch multiple constraints mentioned in queries with corresponding evidence indocuments and correctly perform various set operations. The dataset isconstructed semi-automatically using Wikipedia category names. Queries areautomatically composed from individual categories, then paraphrased and furthervalidated for naturalness and fluency by crowdworkers. Crowdworkers also assessthe relevance of entities based on their documents and highlight attribution ofquery constraints to spans of document text. We analyze several modernretrieval systems, finding that they often struggle on such queries. Queriesinvolving negation and conjunction are particularly challenging and systems arefurther challenged with combinations of these operations.
AkihikoWatanabe
changed the title
あ
QUEST: A Retrieval Dataset of Entity-Seeking Queries with Implicit Set
Operations, Chaitanya Malaviya+, N/A, arXiv'23
May 22, 2023
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)
The text was updated successfully, but these errors were encountered: