3 ms·

I have been working on one, it's not really ready for prime time yet. The idea is that it's a news aggregator that learns what you like, so it forms a probabil

I have been working on one, it's not really ready for prime time yet. The idea is that it's a news aggregator that learns what you like, so it forms a probabilistic model of what it thinks your interests are and applies that model to the network of submitted content to decide what you see. Instead of subreddits the system forms subcommunities by showing users with similar interests similar lists of content. The main thing I'm having a lot of trouble with is figuring out how to do efficient keyword extraction on pictures and video. Otherwise it only supports links that have a lot of text. The other problem is it doesn't have enough of a community feel so that's something else I'm working on. One of the ideas I am about to experiment with is removing user accounts and using browser fingerprints instead. I think if users dont have to sign up it will encourage users to more actively participate in the community.

meesterdude 11y ago
As a fellow builder, also with something not ready for prime time, I say: ship that shit. If you're going to strike, now's the time. Any reason it can't be live, end of the month? > Otherwise it only supports links that have a lot of text Sounds good to me! > using browser fingerprints instead By all means experiment; but this has a few problems with it. Some people like (or need) multiple accounts, which would mean separate browsers at the least. Equally, it's trivial to spin up a VM with a browser. So, the users who need multiple usernames can't (or lack the technical understanding), and the ones you don't want having multiple can easily create hundreds.
- faizshah 11y ago
  Thanks for the support, I can probably make it live by the end of the month I just thought that a lack of image and video support would limit its chances at survival. The browser fingerprinting idea is a bit more involved. The idea is that everyone is anonymous and the types of people who you seem to like, based on factors like leaving your cursor over their comment for a while, are slowly deanonymized to you. The way browser fingerprinting would work in this scenario is that instead of storing usernames attached to models, we store a few different indicators (like browser fingerprints) and create probabilistic models relating these indicators to the aforementioned models. That way even if we have a completely new user we can create a new initial model for them based on what they might like. I have only just started working on this idea though, so I'm not sure how well this would work in practice and I'm not sure what other indicators I might use. With this in mind I don't think there would be a reason to spin up VMs and create new 'accounts' because it would just limit your personal experience.

meesterdude 11y ago

As a fellow builder, also with something not ready for prime time, I say: ship that shit. If you're going to strike, now's the time. Any reason it can't be live, end of the month? > Otherwise it only supports links that have a lot of text Sounds good to me! > using browser fingerprints instead By all means experiment; but this has a few problems with it. Some people like (or need) multiple accounts, which would mean separate browsers at the least. Equally, it's trivial to spin up a VM with a browser. So, the users who need multiple usernames can't (or lack the technical understanding), and the ones you don't want having multiple can easily create hundreds.

faizshah 11y ago
Thanks for the support, I can probably make it live by the end of the month I just thought that a lack of image and video support would limit its chances at survival. The browser fingerprinting idea is a bit more involved. The idea is that everyone is anonymous and the types of people who you seem to like, based on factors like leaving your cursor over their comment for a while, are slowly deanonymized to you. The way browser fingerprinting would work in this scenario is that instead of storing usernames attached to models, we store a few different indicators (like browser fingerprints) and create probabilistic models relating these indicators to the aforementioned models. That way even if we have a completely new user we can create a new initial model for them based on what they might like. I have only just started working on this idea though, so I'm not sure how well this would work in practice and I'm not sure what other indicators I might use. With this in mind I don't think there would be a reason to spin up VMs and create new 'accounts' because it would just limit your personal experience.