Using GPT-3 to identify relationships in a corpus

Question

I have a corpus of 15K news articles. I would like to train a GPT model (3 or 4) to ingest these texts and then output how the locations, events, actions, participants, and things described in the texts are related to one another. So if the corpus says John Smith took part in a protest, I'd like to tell me this and what other people took part, how the protest was related to specific locations, etc. Is this possible?

If so can someone please point me in the right direction for learning how to do it? When I do searches all I'm finding is links about using GPT models to give extractive or abstractive summaries of individual texts. I suppose that's related but not quite the same.

Answer 1

So if the corpus says John Smith took part in a protest, I'd like to tell me this and what other people took part, how the protest was related to specific locations, etc. Is this possible?

The task is referred to as named-entity recognition and relation extraction+classification. See {1} for an evaluation of ChatGPT for named-entity recognition and relation extraction+classification. It is possible to do these tasks with ChatGPT.

---

References:

• {1} Lai, Viet Dac, Nghia Trung Ngo, Amir Pouran Ben Veyseh, Hieu Man, Franck Dernoncourt, Trung Bui, and Thien Huu Nguyen. "ChatGPT Beyond English: Towards a Comprehensive Evaluation of Large Language Models in Multilingual Learning." arXiv preprint arXiv:2304.05613 (2023).