microsoft
I want to use graph rag to extract some additional entities that I want to define manually
Discussed in https://github.com/microsoft/graphrag/discussions/536
Originally posted by streetstylecoder July 13, 2024 , for example here when I am working on a businesss proposals , where there are hundreds of files , but I want to query "what was the average budget of proposals" or "what were the total number of proposals in software category" which was always the problem with the naive rag because it can only return a specific number of business proposals , now that graph rag is here , I want to make use of it so that while embedding the data we can use the total summarised information of the set of files
Can someone please help me out with the same ?
Have you tried the automatic templating function? Try entering "business proposals" in the --domain section, and it should generate series of extraction prompts suitable to your needs.
https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/
If you want to have a bit more control, you could define the extracted entities directly by fiddling with the prompt. In the data root directory, there's a prompts directory, and under that you will find a file named entity_extraction, you can manually enter some additional entities in the entity_type, but you might want to do some PE to get the most out of it.
