4 ms·

Nice Idea, I am curious to know how contextsense give ratings.. could you please describe in more detail .. Thank you

paraschopra 17y ago
It doesn't give out ratings. It will rather generate tags and categories which represent a submission. Tags such as: erlang, database, module, in-memory, etc. You can also generate other features such as word count, presence of images, (domain name, poster's name, etc.) Now what you do is to go through all (or as many as you could) stories that have been on HN front page, calculate features for those stories. Those features are used to learn a prediction model which you can apply to new stories to predict ratings.
- whatusername 17y ago
  I like the general concept.. The only issue I really see is with promoting groupthink. Doesn't that approach create a positive feedback loop. (TO be specific - if those predictions were exposed in a meaningful way, people would be more likely to view those stories, thus they would be upvoted, thus similar stories would be predicted to be successful) The second issue is that stuff like this: (One of mine but I think it was a fascinating article) would probably fail any relevancy test: http://news.ycombinator.com/item?id=1046378 http://news.ycombinator.com/item?id=1046378 Concepts: Charm, Bed and Breakfast, Localities, Lodging, Ohio
  - paraschopra 17y ago
    I think HN community is composed of mostly independent thinkers, so what it will more likely do is to expose interesting stories to them, it is their choice in the end to actually upvote or not. (That is precisely the problem we are trying to solve: interesting stories not getting noticed). Yes, the algorithm won't be perfect but it could factor in historical data such as previously how many upvotes submitor's other stories got or how many votes did the stories from that website got. On both those counts, your post will score high.
    - RiderOfGiraffes 17y ago
      I like the idea - I've been toying on and off with similar thoughts, it appears you're more advanced in your thinking than I. However, you say: > I think HN community is composed of mostly independent thinkers ... That's certainly been true, and is probably still mostly true, but the evidence suggests it's becoming less true as HN becomes more popular and well-known. I think that's inevitable, and is recognised by PG, which is why he's testing and experimenting with the way votes and karma work. I think you need to plan for the quality and subjects to drift significantly in the future, starting now. If you can solve (or anticipate) that, you may have a winner.

It doesn't give out ratings. It will rather generate tags and categories which represent a submission. Tags such as: erlang, database, module, in-memory, etc. You can also generate other features such as word count, presence of images, (domain name, poster's name, etc.) Now what you do is to go through all (or as many as you could) stories that have been on HN front page, calculate features for those stories. Those features are used to learn a prediction model which you can apply to new stories to predict ratings.

whatusername 17y ago
I like the general concept.. The only issue I really see is with promoting groupthink. Doesn't that approach create a positive feedback loop. (TO be specific - if those predictions were exposed in a meaningful way, people would be more likely to view those stories, thus they would be upvoted, thus similar stories would be predicted to be successful) The second issue is that stuff like this: (One of mine but I think it was a fascinating article) would probably fail any relevancy test: http://news.ycombinator.com/item?id=1046378 http://news.ycombinator.com/item?id=1046378 Concepts: Charm, Bed and Breakfast, Localities, Lodging, Ohio
- paraschopra 17y ago
  I think HN community is composed of mostly independent thinkers, so what it will more likely do is to expose interesting stories to them, it is their choice in the end to actually upvote or not. (That is precisely the problem we are trying to solve: interesting stories not getting noticed). Yes, the algorithm won't be perfect but it could factor in historical data such as previously how many upvotes submitor's other stories got or how many votes did the stories from that website got. On both those counts, your post will score high.
  - RiderOfGiraffes 17y ago
    I like the idea - I've been toying on and off with similar thoughts, it appears you're more advanced in your thinking than I. However, you say: > I think HN community is composed of mostly independent thinkers ... That's certainly been true, and is probably still mostly true, but the evidence suggests it's becoming less true as HN becomes more popular and well-known. I think that's inevitable, and is recognised by PG, which is why he's testing and experimenting with the way votes and karma work. I think you need to plan for the quality and subjects to drift significantly in the future, starting now. If you can solve (or anticipate) that, you may have a winner.