Multi-value Classification Of Patient Reviews Based On Long Short Term Memory Model

[Purpose] This study addresses the problem of patient review classification requiring a lot of manpower,time cost,subjective impact of classification,and the possibility that each patient review may have multiple classification categories.We use a scientific classification standard to manually label the collected patient reviews.And design a method based on the combination of multiple binary classifiers based on Long Short Term Memory models to classify patient reviews,aiming to(1)construct a manually labeled scientific patient review multi-value classification corpus,(2)and Efficient multi-value classification for patient reviews is achieved using a patient review multi-value classification classifier based on long short-term memory models.[Methods] Based on the 17184 patient reviews obtained in a hospital follow-up system and patient satisfaction survey,the study conducted multiple rounds of manual labeling and consistency testing to obtain a multi-valued corpus of patient reviews after.Using the corpus,we trained a patient review multi-value classification classifier based on Long Short Term Memory(LSTM)models,and implemented patient review sentiment classification,patient negative review multi-value classification,patient positive review multi-value classification.Finally we used classification model of Support Vector Machine,Random Forest and Gradient Boosting Decision Tree to compare the classification performance of these three models and LSTM.[Results] According to the consistency test,the patient review multi-value classification corpus constructed in this study has good consistency and high practical value.In the experiments of patient review multi-value classification,the LSTM model showed better classification performance than the other three classifiers,and in the final simulation experiment,for a independent test set,98.37% of the corpus predicted at least one class,93.91% of the corpus predicted all categories,73.49% of the corpus was completely predicted correctly,and only 1.42% of the corpus was not predicted at all,It also demonstrates excellent performance of the multi-value patient review classifications.[Conclusions] This study introduced a scientific multi-value classification criteria for patient reviews,adjusted the classification criteria according to the actual data.Then we used the classification criteria to manually label the collected patient reviews,forming a high-quality patient review multi-value Classification corpus.At the same time,a method combining multiple binary classifiers based on LSTM models is designed,which achieves efficient multi-value classification of patient reviews and lays a solid foundation for better understanding of patient reviews.