Apprentice bot model design and implementation for psychological clients’ therapy

Abstract Conversational agents play a crucial role in advising clients and providing treatment for mental health issues. In Ethiopia, a developing country with a high prevalence of mental health issues, an AI text-based chatbot has been developed to assist users in addressing these cases. The bot re...

Full description

Saved in:
Bibliographic Details
Main Authors: Kibrom Gidey, Hailay Beyene, Fiseha Haileslassie, Haben Berihu
Format: Article
Language:English
Published: Springer 2025-02-01
Series:Discover Applied Sciences
Subjects:
Online Access:https://doi.org/10.1007/s42452-025-06515-2
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Conversational agents play a crucial role in advising clients and providing treatment for mental health issues. In Ethiopia, a developing country with a high prevalence of mental health issues, an AI text-based chatbot has been developed to assist users in addressing these cases. The bot recognizes features such as general knowledge queries, potential mental disorders, auto-generated advice, and client stories. The syntactic and semantic structure of data and user chat's memory network were investigated. Four neural networks (LSTM, single GRU, transposed GRU and double GRU) were experimented with to find the best-fitting deep learning model. The approach uses Amharic word2vec word embedding techniques (skip gram and CBOW) for semantic extraction of Amharic words and a seq2seq model for response generation. An ensemble architecture of generative-based and retrieval-based approaches was employed to handle user dialogue. The scruffy technique was used to generate word embedding, which was then used to extract semantically comparable terms. The model also embedded custom rules for smoothness and error handling. The performance and outcomes of the chatbot were evaluated using cross-validation, data organization, perplexity, accuracy measurement, f-1 score (precision and recall), and human-evaluation methods. The model's accuracy was 79.62%, with a 79.62% likelihood of delivering relevant responses. The results show that the ensemble architecture based on seq2seq modeling with embedded custom rules and Amharic word vector implementation provides pertinent responses for user utterances.
ISSN:3004-9261