Unsupervised query expansion for theme queries by exploration and fusion

Theme queries are a subtype of informational queries that cover broader topics than those expressed explicitly in the query terms. For this type of query, it is natural to resort to query reformation techniques such as query expansion. While traditional query expansion may be used to improve the query, new query terms are ranked only once and no verification of quality of the new ranking is made. In comparison, we propose to address such queries with iterative unsupervised expansion: adaptive query exploration and rank fusion. In the lack of human labels, we combine multiple signals to help select the best expansion candidates and further use rank fusion to generate the final ranking and the new query with discretion. The two modules iteratively improve each other: the rank fusion results provide an estimated ranking improvement score to guide the query exploration module and the query exploration modules selects the most promising query variants to add to the fusion. Experiments on the NYT dataset and the Washington Post dataset confirm that our proposed method successfully improves upon baseline methods.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2019

Author dc:creator

• Li, Sha

Contributors dc:contributor

• Han, Jiawei

• Information retrieval
• query expansion
• contextual bandit algorithm

Statement dc:rights

• Copyright 2019 Sha Li

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/105828

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/105828