How to extract pros and cons from news article


Does anyone know if GPT-3 can find pros and cons from a news article and how to go about it.

Hi there, could you clarify what you mean by that? Are you referring to extracting arguments for or against a certain topic? I’d love to see some examples of what you mean. I’m not sure the average article really has “pros and cons” (e.g. what are the pros and cons of an article about a hurricane or a building collapse or a study on climate change).

1 Like

Hi Sorry,

My idea is to build a stock market site where users go to research the qualitative side of the stock. The news articles are mainly about companies in the news and the pros and cons are to be extracted. The pros and cons would than be extracted sentence by sentence where a crowd would vote on them to see if they’re helpful. Perhaps pros and cons wasn’t the best way to model the prompt.

You could try re-purposing the tweet classifier as a general sentiment classifier, and then classify stock text content as positive or negative.

1 Like

This is pretty easy. Take the following article as an example: Johnson & Johnson asks high court to void $2B talc verdict

Use REGEX to clean up the text of the article and the prompt is simple

There are several caveats here:

  1. Financial advice is a prohibited use case for GPT-3
  2. It is unethical (and illegal in some places) to summarize or aggregate news articles without attribution: The Rise of the News Aggregator: Legal Implications and Best Practices by Kimberley A. Isbell :: SSRN

Anyways, the above result was with 100% default settings on DAVINCI-INSTRUCT-BETA


Thank you everyone, it turns out that the idea is a no go due to restrictions of terms of service an attribution to the author.

I tried something similar but to calculate a better solution you need more than positive and negative. You need the matrix. Otherwise it kinda loses context.