Hey HN, I needed to assess the state of sentiment models and couldn’t find a good way to compare them. I built this interactive site that lets you compare 12 models side by side, from Python libraries like NLTK Vader, to top performing models on HuggingFace, to commercial sentiment APIs and GPT4o. This is a research project, there is no paywall - you can enter your own text (1) and get the results back immediately. The results are fascinating and we made it easy to explore not just the leaderboard, but where models get it wrong. For example, most models (including AWS Comprehend) can’t get this positive sentiment: "Food doesn’t get better than this. I was sad when I finished, actually sad. To die for." (2) And yes, GPT4o is currently the best performing. It's crazy how many laboriously researched models are superseded by general purpose foundation models. Let me know what you think? 1: https://addmaple.com/sentiment/own-text 2. https://addmaple.com/sentiment/public-reviews/manteca/C9bvMu... |
No comments yet