Chatbot for Information Service of New Student Admission Using Multinomial Naïve Bayes Classification and TF-IDF Weighting

Abstract

New student admission is a process where prospective students need the information to decide which higher education institution they will enroll. Live chat on the institution website is one of the reliable information sources to find information about the institution. Most higher education institutions have their website but not every single of them has implemented a live chat feature on the website. Live chat requires humans to answer website visitors' questions. However, there are limitations in humans to always be able to respond and provide answers accurately. A chatbot is a machine learning implementation that can be applied to overcome these limitations. Natural Language Processing (NLP) concept can help chatbots translate human language and help the computer understand what humans mean in their language. The machine learning model that is used for classification is Multinomial Naïve Bayes with the help of term weighting using TF-IDF. With the classification model, prospective students' questions can be classified based on their intents, but the model needs labeled questions as the training data. Chatbot with quality training data set and 24/7 service availability makes prospective students' questions can be answered quickly anytime and anywhere. In this research, 1.330 questions were gathered as training data and grouped into 17 intents. More than 90% of questions predicted correctly using K-Fold Cross Validation, but only 65% when the chatbot is tested by website visitors due to less clean and less complete training data set that obviously can be improved in the future.