The styles across different platforms do indeed show noticeable differences. I think this part can be refined further, which will make it more vertical and precise. This way, it will look more streamlined and accurate.
Scoring headlines against a trained model beats the usual word-count heuristics. Would be useful to see which words moved the score most - even a simple highlight of the weakest tokens would make rewrites much faster.
The styles across different platforms do indeed show noticeable differences. I think this part can be refined further, which will make it more vertical and precise. This way, it will look more streamlined and accurate.
Scoring headlines against a trained model beats the usual word-count heuristics. Would be useful to see which words moved the score most - even a simple highlight of the weakest tokens would make rewrites much faster.