Nowadays, people are significantly affected by mental illnesses such as depression and anxiety disorders. These conditions can often be treated successfully if appropriate therapy is provided promptly. However, due to a shortage of mental health professionals, limited resources, economic constraints, and fear of social stigma, many people do not receive sufficient therapy. As a result, providing effective mental health interventions remains challenging. This study addresses this gap by designing and developing an Amharic language mental health chatbot to provide supportive counseling for depression and anxiety. The primary research question is as follows: How can a deep learning-based, retrieval-focused Amharic chatbot be effectively implemented to support mental health care? To achieve this, the study employed a design science research methodology, integrating natural language preprocessing with a Bidirectional Long Short-Term Memory (BiLSTM) network and Word2Vec embeddings to capture semantic relationships. Data were collected from both documented and non-documented sources, preprocessed, and structured into intents, patterns, and responses. Experimental results demonstrate that the BiLSTM model achieved 91.25% accuracy in classifying user inputs. In addition, a preliminary User Acceptance Test involving mental health experts and volunteers yielded an average satisfaction score of 86.6%, confirming that the system is user-friendly, provides clear responses, and is practically applicable in real-world settings. Unlike prior rule-based or English-language chatbots, this work makes a novel contribution by applying advanced deep learning techniques to a low-resource, morphologically complex language in a culturally sensitive domain. The findings highlight the potential of natural language processing and deep learning to deliver scalable, accessible, and stigma-free mental health support in Ethiopia and similar low-resource settings.
Wubneh et al. (Wed,) studied this question.