Translation of Natural Language ƒQuery Into Keyword ƒery Using a RNN Encoder-Decoder

摘要

Œe number of natural language queries submiŠed to search enginesrnis increasing as search environments get diversi€ed. However,rnlegacy search engines are still optimized for short keyword queries.rnŒus, the use of natural language queries at legacy search enginesrndegrades the retrieval performance of the engines. Œis paper proposesrna novel method to translate a natural language query into arnkeyword query relevant to the natural language query for retrievingrnbeŠer search results without change of the engines. Œe proposedrnmethod formulates the translation as a generation task. Œat is,rnthe method generates a keyword query from a natural languagernquery by preserving the semantics of the natural language query. Arnrecurrent neural network encoder-decoder architecture is adoptedrnas a generator of keyword queries from natural language queries.rnIn addition, an aŠention mechanism is also used to cope with longrnnatural language queries.