Explainable Mental Health Chat Support using MKO-DEBiGRU with Knowledge Graph and SHAP

¹ Department of Computer Science, Nitte Meenakshi Institute of Technology, Nitte (Deemed to be University), Bengaluru, India
² Department of Computer Science and Engineering (AI&ML), Geetanjali College of Engineering and Technology, Hyderabad, India
³ Department of Electronics and Communication Engineering, KLE Technological University, Hubli India
⁴ Department of Artificial Intelligence and Machine Learning, Ballari Institute of Technology & Management, Ballari, India

^* Corresponding author: sujata.patil@kletech.ac.in

Abstract

Recently, mental health chat support has become increasingly important for providing timely assistance and guidance to users seeking psychological assistance. This task becomes more challenging because of unstructured and noisy text inputs, where slang, abbreviations, and inconsistent phrasing limit the reliability of the responses. However, traditional models struggle to maintain accuracy and offer explainable reasoning in these scenarios. Hence, this study proposes a Medical Knowledge Graph integrated orthogonal independent Deep Bidirectional Gated Recurrent Unit (MKO-DEBiGRU) method for mental health text classification. To enhance interpretability, a Medical Knowledge Graph is used to map the predictions to domain knowledge, and SHapley additive explanations (SHAP)-based Explainable AI (XAI) is integrated to provide feature-level insights into model decisions. The proposed model, MKO-DEBiGRU, is evaluated using the Anxiety, ADHD, Bipolar, and Depression datasets, where preprocessing includes removing stop words, punctuation, hashtags, and special characters to improve text quality. The experimental results show that the proposed MKO-DEBiGRU method achieves 99.98% accuracy, significantly outperforming existing BiLSTM models in terms of classification performance.