An Early Prediction and Label Smoothing Alignment Strategy for User Intent Classification of Medical Queries

Yuyu Luo, Zhenjie Huang, Leung Pun Wong, Choujun Zhan, Fu Lee Wang, Tianyong Hao

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

3 Citations (Scopus)

Abstract

Deep learning models such as RoBERTa and Bi-LSTM are widely utilized in user intention classification tasks. However, in the medical field, there are difficulties in recognizing user intents due to the complexity of medical query representations and medical-specific terms. In this paper, an alignment strategy based on early prediction and label smoothing named EP-LSA is proposed to classify user intents of medical text queries. The EP-LSA strategy uses a Chinese pre-training model RoBERTa to encode sentence features with rich semantic information, predicts the early features of Bi-LSTM in RCNN and aligns them with output features. The early knowledge from early prediction is processed utilizing cross-entropy loss incorporating label smoothing, which enhances random information to the early knowledge and helps the strategy to extract more fine-grained features related to intention labels. Experiment evaluation was performed based on two publicly available datasets KUAKE and CMID. The results demonstrated that the proposed EP-LSA strategy outperformed other baseline methods and demonstrated the effectiveness of the strategy.

Original languageEnglish
Title of host publicationNeural Computing for Advanced Applications - 3rd International Conference, NCAA 2022, Proceedings
EditorsHaijun Zhang, Yuehui Chen, Xianghua Chu, Zhao Zhang, Tianyong Hao, Zhou Wu, Yimin Yang
Pages115-128
Number of pages14
DOIs
Publication statusPublished - 2022
Event3rd International Conference on Neural Computing for Advanced Applications, NCAA 2022 - Jinan, China
Duration: 8 Jul 202210 Jul 2022

Publication series

NameCommunications in Computer and Information Science
Volume1637 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference3rd International Conference on Neural Computing for Advanced Applications, NCAA 2022
Country/TerritoryChina
CityJinan
Period8/07/2210/07/22

Keywords

  • Alignment
  • Early prediction
  • Intent classification
  • Label smoothing

Fingerprint

Dive into the research topics of 'An Early Prediction and Label Smoothing Alignment Strategy for User Intent Classification of Medical Queries'. Together they form a unique fingerprint.

Cite this