What is the best approach to Automate Extracting and Answering Questions from Documents in UiPath using AI

Help Studio
,
1

Hello Everyone,

I have an Excel file containing a list of questions that require precise answers in a specific format (please see the example below).

A spreadsheet displaying columns for various data points regarding climate-related disclosures, such as question descriptions, data types, and related references, with annotations in red text indicating sections for format and answer insertion. (Captioned by AI)
A spreadsheet displaying columns for various data points regarding climate-related disclosures, such as question descriptions, data types, and related references, with annotations in red text indicating sections for format and answer insertion. (Captioned by AI)2760×490 61.9 KB

I also have various documents (PDFs, images, Word files, etc.) stored in a folder that contain the information needed to answer these questions.

I’m exploring the best solutions to automate this process using UiPath, and I would appreciate your suggestions and insights on the following:

  1. Extracting Questions: What’s the best way to automatically extract specific questions from the text? For example, a question like “Percentage of remuneration recognized that is linked to climate-related considerations.” and get a precise question from it (or maybe have similars questions using NLP to help finding data in the next step)
  2. Finding and Matching Answers: Once the question is identified, how can we effectively search through the documents for the most accurate answer? Which technologies or activities would you recommend for this, considering the need for precise matching and accuracy?

I’ve considered using a chatbot solution (e.g., Poe, which I find more suited than ChatGPT for this task), but I’m open to other suggestions that might be more efficient or cost-effective.

Thank you in advance for your help and recommendations!

Best regards,

2

I’m not an expert so may be wrong, but I think specifically for stuff like this a Retrieval Augmented Generation (RAG) model should work, where you’re making the model retrieve answers based on your dataset of pdf’s, images, etc. I think Azure AI Search can do something like it.

As I said, I’m not an expert on this so I’m unaware if you need to constantly retrain your model for new documents or if it’s dynamic, but the links below will hopefully help you on your search.