Most of the time, online messaging group users must scroll and read a large number of irrelevant posts in order to gain a clear understanding of what is being discussed in the group to which they belong. Messaging groups can get congested with unnecessary messages, causing members to miss out on important issues and information. There is a need to assist users of multi-user chat systems in understanding what the group discussion is all about at any particular time without having to read all of the posted messages. This paper describes an approach to discovering topics in online chat groups. In order to extract and categorize subjects from unseen texts in online group discussions. We developed a new multi-user chat system (ML-CHAT-APP) that automatically identifies and categorizes topics within posts/messages as they appear. We implemented a combination of a Latent Dirichlet Allocation (LDA)-based model with Multinomial Logistic Regression. The resulting model was integrated into the ML-CHAT-APP built with Python and Tkinker framework for Graphical User Interface. The results show that the application was helpful in identifying topics in text conversations and adding identified topics as labels to message posts in real-time. Keywords: NLP, Topic Modeling, Latent Dirichlet Allocation; Logistic Regression