Explaining prediction explanations automatically

One useful method that helps in deriving insights from prediction explanations is none other than using large language models (LLMs) such as ChatGPT. ChatGPT is a transformer model that is trained to provide results that match logical reasoning related to the instructions provided. The theory here is that if you can format your prediction explanations data in a way that it can be fed into a transformer model, and instruct the LLM to derive insights from it, you will be able to obtain insights from multiple different perspectives.

In the previous tutorial, we attempted to explain the explanations of four different samples consisting of two correctly predicted positive sentiment examples and two correctly predicted negative sentiment examples. Now, let’s use an LLM model to gain insights. Here, we will separately generate insights for the two correctly identified positive sentiment and two correctly identified negative sentiment...