Personalized Mimic Chatbot Based on WhatsApp Conversation History Using a Hybrid Intent Classification and Local LLM Pipeline

Abstract

Currently, chatbots have been widely used in various types of applications, most of them still generate responses that are general in nature and have not been able to reflect the specific communication style of different users. Therefore, this study aims to create a personal chatbot that is able to learn communication patterns and conversations from exported WhatsApp chats, allowing the system to imitate user habits in using the WhatsApp application, such as message length, certain vocabulary, and also distinctive response patterns. The system is designed using a hybrid pipeline with two main components, namely intent classification and response generation. In order to analyze and evaluate the system properly, several approaches were implemented gradually to determine the best-performing model. First, V1 Base uses Vanilla LSTM with a limited dataset, followed by V2 Base using LSTM with a larger dataset, and V3 as the final model combining SVM and LLM. Based on the evaluation from the hybrid pipeline, which is model V3 achieved an accuracy of 82.53%. That indicates the model is producing more natural responses and better simulating the user’s unique communication style. At the end, the results of this study show that combining a machine learning algorithm like an intent classifier with local language models can improve personalization in chatbot systems.