Citation

Xu LP, Song WA. Zhongbei Daxue Xuebao (Ziran Kexue Ban)/Journal of North University of China (Natural Science Editio 2019; 40(4): 350-357.

Copyright

(Copyright © 2019)

DOI

10.3969/j.issn.1673-3193.2019.04.010

PMID

unavailable

Abstract

In order to improve the classification precision of suicidal idea detection model based on n-gram feature, a suicide dictionary with strong migration ability was established about suicide idea detection, and the part-of-speech feature was proposed. Taking n-gram features, linguistic features (include vocabulary features and part-of-speech features) as input feature was driven by random forest and support vector machine algorithm, the contrast experiment and control variables method was used to study on the impact of language features on model performance. The results show that the language features have greatly improved the performance of the model. By comparison, the performance of the model based on n-gram features and linguistic features is better than model based on n-gram features and model based on n-gram features and dictionary features. And the performance of the random forest algorithm is improved by approximately 20%. © 2019, The Press of NUC. All right reserved.

Language: zh

Keywords

Suicidal ideation; Support vector machine; Random forest; Linguistic characteristics