Q: Inquiry About Extracting Product Descriptions from PDFs Using Socrates
I would like to know if it's possible to extract all sentences from a PDF containing product descriptions using Socrates. The extracted sentences should remain unaltered and include reference numbers (if available).
I attempted to do this and initially received results that I was satisfied with. However, I later discovered that some sentences contained hallucinated information, including fabricated sentences and references that were not present in the original PDF.
I used the "Deep Dive Tools" and "Prompt Loop - Section" features for this task, but I couldn’t find specific instructions regarding this issue. Do you have any recommendations or guidelines on how to handle this problem effectively?
Jon_Socrates
Nov 23, 2024A: Hello --
I saw your support email on this topic so will answer here.
Yes, the best way to do this is:
1) run a Prompt Loop separated by paragraph
2) have the prompt be something like: "Extract all product descriptions in this paragraph. If there are no product descriptions, say 'No Product Description'".
^this should minimize hallucinations.