Mental Health Status Recognition Based on Bidirectional Long Short Term Memory with Hierarchical Attention Model

In recent years, mental health disorders have become a serious global issue, impacting individuals across the world. It is essential to accurately recognize the mental health issues for effective treatment still conventional diagnostic methods, such as clinician estimations and surveys, have limitations such as time-consuming, and inaccessible for effective interpretation. Therefore, the rapid improvement of deep learning facilitates to provide scalable and data-driven approach to mental health assessment. This approach suggests a novel approach using combination of effective hybrid deep learning models. Initially the data is collected from the dataset psychinfo which is further processed using data cleaning and normalization. Then, the processed data is fed into a transformer model namely T5 which generates and extracts optimal attributes and in the end recognizes the key parameters of mental health illness recognition using a robust hybrid model namely Bidirectional Long Short Term Memory (Bi-LSTM) - Hierarchical Attention Model (HAN). Hence, the proposed method is evaluated by comparing with existing methods such as LSTM and has acquired higher results with an accuracy of 98.63%.