Anglo-Bangla Language-Based AI Chatbot for Bangladeshi University Admission System

Many Bangladeshi students face significant challenges during university admission because of a lack of communication. However, some of the usual chatbots have been implemented in English/Bangla. Most students are comfortable communicating in a slang mixture of Bengali and English (Anglo-Bangla). This article presents a language detection model for Anglo-Bangla using machine learning algorithms, including Logistic Regression (LR), Decision Tree (DT), K-Nearest Neighbours (KNN), Support Vector Machine (SVM), and Multinomial Naive Bias (MNB). The model is designed to assist the university admission process in Bangladesh through an automated chatbot system. The chatbot system utilizes the language detection model to understand and respond to the queries of potential students in English, Bengali, and Anglo-Bangla. When the chatbot is unable to provide a proper answer in rare cases, the system logs the query along with user contact information. University officials will then update the appropriate response and the system will send the response to the user via email or SMS. The detector continuously detects the language and creates separate datasets to improve the chatbot’s performance. Our model was trained on a dataset of 0.15 million conversations collected from Socian Ltd., a leading NLP company. The dataset consisted of unlabeled Bangla and Anglo-Bangla languages and an English dataset collected from Kaggle, which was a mixed-language dataset. The model’s performance shows that the Multinomial Naive Bayes algorithm yields the highest accuracy in detecting the following languages: 99.64% for the Mixed, 99.13% for English, 100% for Bangla, and 99.51% for Anglo-Bangla. The proposed model can be used to improve the performance of other multilingual chatbot systems and assist with tasks such as customer service, medical assistance, and data collection in polyglot societies.