4 ms·
2 things 1. Why was this formed as classification instead of as regression? Seems much more like a regression problem (predict how many upvotes you will get gi
by Der_Einzige 6y ago
2 things
1. Why was this formed as classification instead of as regression? Seems much more like a regression problem (predict how many upvotes you will get given a reddit post)
2. Seems like it could have been effective with better pre-training. I'd love to see the author rerun the experiments with the likes of GPT-2 or other pre-trained models vectorizing the text first.
- scollet 6y agoWouldn't you have to bucket and rank content and subreddits, respectively, anyways? Otherwise you will probably have a none-size fits all algorithm. Seems like an incremental decison tree problem with weights around karma and time of day.