Steps to reproduce

First, download the financebench dataset and extract text chunks from the pdf files.

python download_financebench.py
python create_text_collection.py

Then, run the Text RAG using python textrag.py. It will generate a json file with the retrived chunks and the generated answer for each question in financebench.

From the output of the Text RAG, you can extract triples using the text_chunks_to_graph.py, which will scan the retrieved chunks and extract triples from them. It will also generact facts, which are text representation of triples.

Finally, the textrag_fromfacts.py script will query the RAG system using extracted facts instead of raw chunks.