Designing and Developing Bilingual Chatbot for Assisting Ethio-Telecom Customers on Customer Services

Abstract

A chatbot is an application that is used to make conversation with a computer by using natural language. This study aims to develop a bilingual chatbot that supports both Amharic and Afaan Oromo languages for assisting Ethio telecom customers on customer services. Ethio-telecom is an Ethiopian telecommunication company serving as the major internet and telephone service provider. In this study, the dataset was prepared by using both Amharic and Afaan Oromo languages in the form of JSON by filtering the frequently asked questions from the organization website. For the model design purpose, the tflearn DNN and Bi-LSTM model has applied. For the proposed system development purpose, we compared those models by using accuracy metrics to select the best. Based on this, the accuracy of the DNN model was 83.52% and Bi-LSTM model was 85.23% before applying the regularization technique. To overcome the model overfitting challenge L2 regularization technique has applied. After regularization, the model has evaluated by using Accuracy, Precision, Recall, and F1-score and got 82.6, 85.7, 82.6, and 87.7 respectively. In addition to metrics evaluation, the model was evaluated by using human evaluation. This evaluation was used to identify two criteria; that is system performance and user acceptance. The model performance was evaluated based on the number of queries they provide for the proposed system and the number of responses that returned from the system. The second evaluation was users' acceptability of the proposed system. This evaluation was evaluated based on five parameters those are attractiveness, response time, user-friendly, efficiency, and system feasibility.