Intent recognition and classification from sports and legal conversations

Abstract

In today's digital world, the demand for automated chatbots capable of understanding human intent and providing relevant information and advice has grown significantly. Such chatbots can effectively reduce human workload and improve user experience. To train a chatbot, a substantial amount of data is required, along with intent words to guide the model in accurately predicting outputs. In this project, we utilized two conversational datasets: one focused on the Sports domain and the other on the Legal domain. The Sports domain dataset comprises 6 subdomains and consists of 610 utterances, capturing Question Answering Conversations between users and the bot. Our experiments involved implementing various machine learning and deep learning methods. Notably, Support Vector Machine (SVM) emerged as the most successful model, achieving an impressive accuracy rate of over 73%. On the other hand, the Legal dataset encompasses 235 conversations, totalling 3178 utterances. Here, we applied similar machine learning and deep learning techniques. Random Forest and Support Vector Machine again demonstrated their effectiveness, delivering the best results with an accuracy rate just over 78%. Throughout this intent classification project, we showcase the importance of data quality and the impact of different machine learning algorithms in training chatbot models. The findings from our experiments provide valuable insights into the applicability of these models in specific domains, paving the way for more efficient and accurate automated chatbot systems.