I spent significant time on binary text classification, specifically with HN titles. You can actually get up to 65-70% post popularity accuracy just by looking

I spent significant time on binary text classification, specifically with HN titles. You can actually get up to 65-70% post popularity accuracy just by looking at the post title. I am currently creating a filter that filters HN news and similar sources using a similar classfier. It learns on the fly and the accuracy of guessing my 'taste' is about 75%-80%. Better accuracy for this is explainable by the fact that my interests are more focused and the classifier has easier time predicting posts I would be interested in.