Research On The Detection Method Of Suicidal Ideation In Chinese Microblog Based On Language Features

Suicide is one of the three leading causes of death.Therefore,it is urgent to identify suicidal ideation.However,most of the traditional methods for detection of suicidal ideation are based on n-gram feature.In order to improve the accuracy of model detection,a suicide dictionary based on training data was added into the original input feature.However,the accuracy of the obtained model is still not ideal.Aiming at the unsatisfactory recognition accuracy of suicidal ideation,a mobile suicide dictionary was established,and the new model was added with the linguistic features composed of the suicide dictionary and part of speech features,so as to improve the recognition accuracy of the model.In the form of comparative experiment and control variables,n-gram feature and language feature were used as model inputs,and random forest,logistic regression,support vector machine and naive bayesian algorithm were used to construct the classification model.The influence of language feature on the performance of the original model was mainly studied.Through comparison,it is found that the language feature improves the performance of the original model significantly,which reaches about 20% under the random forest algorithm.Contributions:(1)this paper provides a field of highly mobile suicide dictionaries;(2)language features are proposed,and it is proved that language features can improve the performance of n-gram feature-based and n-gram feature-based dictionary models;(3)the model performance of different classification algorithms under n-gram feature,dictionary feature and language feature is tested,which provides some basis for the selection of features and algorithm.