A hierarchical recurrent encoder-decoder for generative context-aware query suggestion
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Standard
A hierarchical recurrent encoder-decoder for generative context-aware query suggestion. / Sordoni, Alessandro; Bengio, Yoshua; Vahabi, Hossein; Lioma, Christina; Simonsen, Jakob Grue; Nie, Jian-Yun.
CIKM '15 Proceedings of the 24th ACM International on Conference on Information and Knowledge Management. Association for Computing Machinery, 2015. p. 553-562.Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - A hierarchical recurrent encoder-decoder for generative context-aware query suggestion
AU - Sordoni, Alessandro
AU - Bengio, Yoshua
AU - Vahabi, Hossein
AU - Lioma, Christina
AU - Simonsen, Jakob Grue
AU - Nie, Jian-Yun
PY - 2015
Y1 - 2015
N2 - Users may strive to formulate an adequate textual query fortheir information need. Search engines assist the users bypresenting query suggestions. To preserve the original searchintent, suggestions should be context-aware and account forthe previous queries issued by the user. Achieving contextawareness is challenging due to data sparsity. We presenta probabilistic suggestion model that is able to account forsequences of previous queries of arbitrary lengths. Our novelhierarchical recurrent encoder-decoder architecture allowsthe model to be sensitive to the order of queries in the contextwhile avoiding data sparsity. Additionally, our modelcan suggest for rare, or long-tail, queries. The produced suggestionsare synthetic and are sampled one word at a time,using computationally cheap decoding techniques. This is incontrast to current synthetic suggestion models relying uponmachine learning pipelines and hand-engineered feature sets.Results show that it outperforms existing context-aware approachesin a next query prediction setting. In addition toquery suggestion, our model is general enough to be used ina variety of other applications.
AB - Users may strive to formulate an adequate textual query fortheir information need. Search engines assist the users bypresenting query suggestions. To preserve the original searchintent, suggestions should be context-aware and account forthe previous queries issued by the user. Achieving contextawareness is challenging due to data sparsity. We presenta probabilistic suggestion model that is able to account forsequences of previous queries of arbitrary lengths. Our novelhierarchical recurrent encoder-decoder architecture allowsthe model to be sensitive to the order of queries in the contextwhile avoiding data sparsity. Additionally, our modelcan suggest for rare, or long-tail, queries. The produced suggestionsare synthetic and are sampled one word at a time,using computationally cheap decoding techniques. This is incontrast to current synthetic suggestion models relying uponmachine learning pipelines and hand-engineered feature sets.Results show that it outperforms existing context-aware approachesin a next query prediction setting. In addition toquery suggestion, our model is general enough to be used ina variety of other applications.
U2 - 10.1145/2806416.2806493
DO - 10.1145/2806416.2806493
M3 - Article in proceedings
SP - 553
EP - 562
BT - CIKM '15 Proceedings of the 24th ACM International on Conference on Information and Knowledge Management
PB - Association for Computing Machinery
T2 - CIKM 2015
Y2 - 19 October 2015 through 23 October 2015
ER -
ID: 159746271