Sentiment Analysis Based on Deep Learning and Its Application in Screening for Perinatal Depression

In the specific field of sentiment analysis tasks, traditional machine learning models lack effective attention to certain features. To address this issue, we extracted and classified user emotion features based on the LSTM network model and applied them to perinatal depression screening scenarios. We use emoticons as feature extraction and modeling for document-level sentiment analysis in specific regions and achieved good results. The results were basically consistent with the findings of the Edinburgh Postnatal Depression Scale. This method greatly shortens the screening time and reduces the doctor-patient communication costs. It has positive significance for specific areas of sentiment classification tasks and provides references for document-level sentiment analysis.