Media RAG: A Chatbot to Explore Multimedia Collections (Master Computer Science Project, Ongoing)
Author
Description
Traditionally, multimedia retrieval systems like vitrivr index large datasets to extract features for future retrieval, typically in an offline phase. During the search process, users issue queries to learn more about the collection's content. Crafting these queries often requires prior knowledge of the collection, which users translate into structured queries for the system to process.
The vision of this project is to provide a more natural and exploratory approach to accessing data in large multimedia collections. Rather than formulating precise queries that yield results at the end of a traditional search, users can engage in a conversation with an LLM agent. This agent can offer insights into the structure of the collection and produce individual items.
Start / End Dates
2024/10/01 - 2025/02/28