MENTAL MAP NAVIGATING HIDDEN EMOTIONS

Abstract: Assessment of mental health is challenging in the changing and digitally networked world today. Mental map navigating hidden emotions is a web-based platform that can integrate machine learning models across the three key modalities such as text, speech and behavioral inputs to detect the underlying emotional states and the possible risks. It uses natural language processing methods to process user with journal text converting raw text into labels using Sentence-BERT embeddings and logistic regression classification. The speech module uses audio signal processing drawing the MFCC features from user uploaded and recorded audio sample applying random forest for the emotions detected with associated labels. Behavior inputs having lifestyle activity are tracked through surveys with custom mapping logics and classifier for in depth risk. It supports muti-step process with the user information showing the result as diagnosis, confidence and suggestions.

Keywords: Random forest classifier, Sentence-BERT, MFCC, logistic regression.