Query Intent Detection with Deep Learning

Posted: 28 Jan 2021

Date Written: November 4, 2020

Abstract

Predicting intent of the user query is one of the fundamental tasks in building information retrieval systems. To solve this task, we present a deep learning method to classify user search queries into one of "get a doc", "topic search", "entity", and "locate source" classes. In this work, models are built and evaluated with actual user queries in Lexis Advance. At a high-level, our method consists of three components i.e. data augmentation, glove embeddings, and convolutional networks algorithm. This approach produced very good accuracy about 87% if we know the query belongs to one of the above four intents and very efficient at runtime by taking only 2 milliseconds per query and standard ec2 instance. Even though we evaluated our approach on four intents, this method applicable for general purpose classification of user queries and to use in production systems.

Suggested Citation

Punuru, Janardhana and Sharma, Sanjay, Query Intent Detection with Deep Learning (November 4, 2020). Proceedings of the 4th Annual RELX Search Summit, Available at SSRN: https://ssrn.com/abstract=3775039

Janardhana Punuru (Contact Author)

LexisNexis ( email )

P. O. Box 933
Dayton, OH 45401
United States

Sanjay Sharma

LexisNexis ( email )

P. O. Box 933
Dayton, OH 45401
United States

Do you have negative results from your research you’d like to share?

Paper statistics

Abstract Views
235
PlumX Metrics