An adaptive mechanism of improved heuristic algorithm and multiscale feature integration with residual GRU for emotion with mental health recognition.

Globally, mental illnesses affect the individual peace of mind in multiple demographics. So, more precise identification of mental disease are termed as important for suggesting better treatment for the individual in the initial stage. Late diagnosis may result in harmful behavioral changes, suicidal thoughts, and death. To end this, an automated system of emotion with mental health recognition is introduced by an adaptive deep learning model. Firstly, the input texts are gathered from the online public data sources. Further, the collected data are undergoing the text pre-processing stage, where the unwanted, irrelevant data are removed. Subsequently, the pre-processed texts are fed as input to the feature extraction procedure. In this phase, the features are captured by the Bidirectional Long Short-Term Memory with Hierarchical Attention (BiLSTM-HA), Term Frequency-Inverse Document Frequency (TF-IDF), and Glove embedding. At the final stage of recognizing the emotions, these three features are subjected to the novel method, named Multiscale Fused Feature-based Adaptive Residual Gated Recurrent Unit (MFF-ARGRU). To attain the optimum performance, the hyper-parameters are optimally selected using the Improved Random Variable-based Sculptor Optimization Algorithm (IRV-SOA). Therefore, the performance of the system is validated using different measures and compared with baseline methodologies. Hence, in contrast, the proposed recognition model reveals that it achieves the high desired value of significantly analyzing the mental state of humans.