Abstract:
A chatbot is a computer program used to interact between humans and computer systems
using natural language processing via speech or text. In this research, we developed
Amharic chatbots that support lawyers and other users of chatbots in assisting with civil
codes i.e., about people's rights and inheritance rights. The majority of the population either
do not afford to hire a lawyer or do not know the civil code. As a result, this study aims to
develop an Amharic chatbot that can support individuals, lawyers, and judges in making
legal decisions. We used an experimental research methodology to develop a model for
civil code laws in this research. In designing a model, we used RNN (LSTM and BiLSTM),
and transformer encoder-decoder chatbot in training and testing a model. We used BLEU
score metrics and user acceptance testing to measure the model's performance. The BLEU
score measurement technique measures the similarity between the target civil code answer
and the model-predicted civil code answers on the user's query. After experimenting with
each model with different hyperparameters, we got a BLEU score of 9.88% in the
transformer model. Compared with the LSTM and BiLSTM models, the transformer model
achieves good performance on training time, memory usage, and prediction accuracy.
Generally, the design model responds to the user's query based on the learned weight in
training. Improving the performance of a model for the transformer model, the dataset size,
and different user utterances involvement remains a challenge.
Keywords: -Amharic language, LSTM, BiLSTM, chatbot, conversational agent