Using LLMs for Analyzing Pancreatic Cancer Literature - A RAG Approach

Authors

Mateja Vočanšek, Faculty of Information Studies in Novo mesto, SloveniaFollow
Milica Peršić Nanut, Jožef Stefan Institute, SloveniaFollow
Biljana Mileva Boshkoska, Jožef Stefan Institute, SloveniaFollow

Abstract

There is a serious overflow in biomedical publications, making it increasingly difficult for researchers to keep up with the literature. An effective, scalable system is needed that can utilize large language models and retrieval augmented generation, and enable researchers to directly ask complex questions and receive accurate, detailed answers based on current scientific evidence. Initially, we built a basic RAG (Retrieval Augmented Generation) system, focused on textual data. We then improved the system by refining key parameters including chunk size, chunk overlap, and the number of top retrieved contexts. We added an upgrade to display output context and references. Expert evaluation shows RAG outperforms ChatGPT and DeepSeek by delivering more accurate references and fewer hallucinations.

Recommended Citation

Vočanšek, Mateja; Nanut, Milica Peršić; and Boshkoska, Biljana Mileva, "Using LLMs for Analyzing Pancreatic Cancer Literature - A RAG Approach" (2025). Proceedings of the 2025 Pre-ICIS SIGDSA Symposium. 41.
https://aisel.aisnet.org/sigdsa2025/41