Show HN: RSS reader with bayesian filter

[nck4222]

There's lots of improvements I'd like to make but thought I should get some feedback first to see if anyone's interested in it.

The main page just shows the most recent feed items (currently from only the 3 feeds I've added myself), but the idea is that it'll show the most popular items aggregated from all users. It also provides TLDR summaries for each item, based on word frequency.

http://bayfig.com

EDIT: Typo

[e1ven]

I like the idea, nck4222!

How do I mark things as spam or ham?

A sidebar on the right that explains what the site is might help me know what I'm doing, or how to use it.

Also-

It might be nice to add a few suggested feeds, under the add feed button.

If I import a bad feed, (example: import "http://daringfireball.net/index.xml without stripping of the http) the app seems to hang.

Also- Importing an OMPL might speed adoption.

Summary- It looks neat, I like the idea, but I can't figure out how to Bayseanify it. It's just displaying all my feeds.

[nck4222 → e1ven]

Thanks for your feedback, glad you like the idea.

To get just the bayesian items, click the "Your Items" tab, and then select "Top Items" from the first drop down box. Although to begin with none of your items will pass the filter because it has no data to run on.

When you click one of your items to open it, the page does a partial post back and marks that item as read. It then uses that data (read items vs. unread items) to filter all future items that are read through the feed.

But I think your right, some explanation on the site would help.

For the bad feed what was the exact text you entered? http://daringfireball.net/index.xml seems to work fine.

[koenigdavidmj]

Cool! I just got done with a capstone project class at university, and one of the groups did a project that was something like this. The main difference was that could specify your own categories, and the Bayesian filtering was done per-user and per-category.

[nck4222 → koenigdavidmj]

Well, the bayesian filter is done by per user (if you sign up). The front page would be the aggregation of all the users.

The category stuff sounds cool. I'd like to add text classification to this at some point.

[e1ven → koenigdavidmj]

Is that up someplace? It seems like it would be cool to look at.

[koenigdavidmj → e1ven]

Let me see if I can get the sources and throw them on Github or something. I don't want to slam their server with half of HN without telling them first.