MindWave app: Leveraging AI for Mental Health Support in English and Arabic

The global concern regarding mental health, especially amidst crises like stress, burnout or depression, necessitates accessible platforms catering to diverse populations, including Arabic speakers. This paper introduces “MindWave,” a pioneering web application for English and Arabic-speaking communities. It is an innovative web application designed to offer support and early detection for individuals experiencing or at risk of mental health. The app uses advanced Natural Language Processing (NLP) and Sentiment Analysis (SA) to analyze user-generated text to detect signs of mental health, providing personalized insights, resources, and support for users in need. Its innovation lies in seamlessly integrating text and voice to comprehensively assess emotional well-being. The app fosters a supportive community where users can connect, share experiences, and provide mutual support. Users can engage in discussions, join support groups, or share coping strategies in a safe and anonymous environment. To address the absence of Arabic sentiment datasets for mental health, this study explores the effectiveness of fine-tuning various pre-trained Machine Translation (MT) models from English to Arabic using our parallel dataset of 945,383 sentences. Furthermore, the study delves into training different variants of Bidirectional Encoder Representations from Transformers (BERT) models on two distinct datasets: the original English dataset and the translated Arabic dataset. The classifiers’ performance trained on the source English dataset is juxtaposed with the results achieved when trained on the translated Arabic dataset. Moreover, we conduct a comparative evaluation of diverse tokenization techniques for Arabic, including Stochastic tokenizer, Disjoint-letter tokenizer, and Morphological tokenizer.