Inspired by ChatGPT-Paper-Reader, this project aims to explore how can LLMs help us read research papers. For now, I have rewritten the core parts of ChatGPT-Paper-Reader with LangChain library to make the project more extensible.
Ideas and contributions are welcomed.
- Identifying section titles of the paper by regex matching and using a special prompt
- Context length control: it splits the paper based on paper sections and
TextSplitterfromLangChain - Summarization with focusing on a set of user-defined key points
- Split the paper with more accurate content recognition (excludes references, extract abstract, etc.)
- Recursive summarization to make the output to the desired length.
- summary < 500 words → read the abstract!
- summary > 3000 words → read the introduction & conclusion!
- summary ~ 1000 words → LLM might help. 🤔
- Answer specific questions when reading