Dynamic Topic Indexing for Chatbots

By: Vanchhit Khare   |   Pages: 15 - 19  |   pdf icon   Open

Abstract

Big chatbots like ChatGPT and Gemini are very good at talking, but they often forget what you said before[1]. If you have a long chat or come back another day, it feels like the AI has no memory. This can be frustrating when you want the AI to remember what you already told it. In this paper, we share a new idea called smart indexing[2]. Think of it like giving the AI a notebook with an index, so both you and the AI can quickly find important things from past talks. Our system does three special things. First, it gives you a side index view, like the table of contents in a book. This makes it easy to jump back to a topic you talked about earlier. Second, it uses topic detection, so when the subject changes, the AI can group parts of the chat into chapters and subtopics. This keeps the conversation neat and organized. Third, it can remember across different days with multi-session memory, so your chats do not disappear after each session. Together, these features make AI feel more helpful and human. It does not just talk, it remembers, organizes, and grows with you over time. This can be useful in school, hospitals, offices, and even research projects.
DOI URL: