Using Sentiment Analysis to Classify Texts

When I started out, we essentially used dictionary-based approaches where you would essentially just count up words in a sentence. But obviously when we look at these larger language models and fine-tuning of these large language models, we can see that some of the classification performances on these texts are actually remarkably outperforming these more crude approaches. And that's particularly the case because they can pick up more complex interrelationships I would say between words with in sentences. So for example, irony might not be the best example but we might be able to pick up on more nuanced linguistic markers like that with models like GPT-3 as you may be aware.