0
$\begingroup$

I want to convert the Extractive QA task as a Reinforcement Learning Problem Statement. So I want to integrate NLP problem into Reinforcement Learning and see if my results were achieving better when compared to only NLP model and see why it is performing better or why it is performing worse

I have a set of legal contracts and I want to use an extractive QA task to extract relevant information such as payment terms and insurance. I have tokenized the data and have a Hugging Face transformer model for this. My ultimate goal is to use a DQN where the transformer model serves as the base model.

To achieve this, I need to convert the tokenized data into an RL environment and train the combined model on a group of contract environments. However, I am currently facing difficulties in creating the environment for DQN.

I would greatly appreciate it if you could offer any advice or guidance on how I can proceed with this task. I am eager to learn and am open to any suggestions you may have.

Thank you very much for your time and consideration.

$\endgroup$
3
  • $\begingroup$ Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. $\endgroup$
    – Community Bot
    Mar 13, 2023 at 15:48
  • $\begingroup$ The problem is, like how RL agent is been used to play video games using images, in a similar manner I want to test, if RL agent is able to perform better than Extractive QA transformers, when I am using DQN and Transformer as base model for DQN. I want to see how RL DQN+Transformer is performing when compared to only Transformer. So here I want to convert legal documents into environment and action space to get start and end logits or Answer token to extract from document based on question I ask to the model. So I have fields like payment terms, insurance, Indemnity, etc in my legal contract $\endgroup$ Mar 14, 2023 at 10:25
  • $\begingroup$ For Video games using RL agent to play the game, trained on images of game play, similarly I want to use RL agents to perform Question Answering NLP Task for my legal documents, where I want to see is it performing better or worse and why is it performing better or worse.. $\endgroup$ Mar 14, 2023 at 10:54

0

You must log in to answer this question.