[Paper] ProdRev - A framework for empowering customers through Generative models

Our paper was accepted at DASA 2022
A fine-tuned GPT3 (curie-engine) model is used to understand thousands of customer reviews. It processes them and provides the user with Pros, Cons and a Final Verdict. For the experiment, a proprietary dataset was created.

“Big Data is good for machines, but humans have not evolved to consume it.”

Human brains use short-term memory for reviewing information and making a decision. E-commerce platforms provide extensive information regarding a product. Such reviews offer essential details regarding the product and its performance in real life. However, the user can easily get overwhelmed with too much information. Due to the large sets of heterogeneous data, information overload can lead to decision paralysis. We have proposed an architecture based on generative techniques to mitigate this scenario and empower the customers:

  1. We fine-tune a GPT3 model using OpenAI.
  2. The GPT3 architecture helps the user to get a clear understanding of the reviews by summarizing and clustering the reviews into pros & cons.
  3. The generative transformer model also provides a final verdict about the concerned product.
  4. Our proposed framework aims at minimizing misleading machine verdicts. By transparently providing all the information to the user and letting them make the final decision

Paid human annotators were used for the completion part. Each set of reviews was manually annotated.

Link to the paper

ProdRev - A generative framework for empowering customers