Pemanfaatan Chatbot Retrieval-Based dan Analisis Sentimen untuk Meningkatkan Layanan Informasi Interaktif di Radio Untar

Abstract

This article discusses the implementation of a retrieval-based chatbot integrated with sentiment analysis to improve the efficiency of information services at Radio Untar. The chatbot developed uses the TF-IDF and cosine similarity methods to match user questions with FAQ data, and is able to handle requests for songs, articles, and podcasts. Sentiment analysis was performed on user interaction logs to assess satisfaction and effectiveness of answers. Based on the results of testing 150 interactions, the chatbot showed an increase in MRR scores from 0.468 to 0.91 and a satisfaction level from 50% to 92% after the fine-tuning process. These findings indicate that a lightweight chatbot retrieval-based system can be used effectively in a campus environment to improve digital interactions.