RSS Readers Part 2

How would Bayesian Tagging work and what would it look like? I put together some conceptual images of what is floating around in my head. I plan on sticking this all into a feature request for RSSOwl.



The stars for each tag represent how much you trust that the rss reader can correctly predict if a news article is part of that tag. When creating a new tag it should sit at 0 stars for awhile. When processing a new items and generating statistical scores for each of the tags the trust level defines the minimum probability that the item must make to be officially tagged. If the rss reader thinks it should be in that tag but the percentage is below the trust level it instead would add it to the "suggested tags list" for that article. This brings us to my next image.



In the image above you can see the list of suggested tags for this article. If the user agrees with one of the suggested tags she can then click that tag to officially tag it. When that happens the rss reader should process the item text for new words to add to the tag and generate new statistical probabilities. Then it should move the tag from the suggested tag list to the tag list. Since a user may want to add many suggested tags to an article it is important not to bring up a dialog box when clicking a suggest tag. That way the user can quickly click the correct suggest tags for each item and move on.

One thing to note in the second image is the fact that at first the tags are hidden under the item description. This is because there may be many suggested and real tags for a item. If each item listed its tags then the item list would become cluttered and hard to read.

My dream aggregator

There are certain features missing from all the RSS aggregator out there. If someone were to implement them it would save the world tons of time. In fact I would be willing to donate money to a pool to get these features implemented in a single client.

• Bayesian Tagging - Tagging is a great feature. Unfortunately in its current state it is really only useful for archiving purposes. Lets say I get on and want to see todays news on Nintendo. Well I can't I can only see yesterdays news on Nintendo because I have to tell my aggregator which articles to give the Nintendo tag. Why not keep track of what I am tagging Nintendo that way when new articles come in the aggregator can properly tag them. If I only want to read about certain things then I don't have to comb through hundreds of feeds my aggregator has done it already.


• Redundancy Checking - The more feeds that I add to my aggregator the more articles I get about the same event. It gets really old reading about the same thing over and over just with slightly different wording. There must be some way to compare two articles and determine a percent chance that they are about the same thing. Once the aggregator figures out which articles are about the same thing it would display only one of them and under the description it could say "24 other feeds also wrote about this" which would be a link to see the other articles. But there should also be a way to mass mark all that are about the same thing as read.


• Daily Dumps - Sometimes I dont want to sit on my computer to read the news. There should be a way to dump your feeds into a pdf file for printing purposes. You could specify certain requirements such as a time range or only a certain tag to include. Also it would format them in such a way to be easy to read and deal with the redundancy checking. Almost like your own newspaper. That way I could print this thing off take it to school and read through what happened while sitting in class.

To summarize my complaints about current aggregators there needs to be more work done by the program and less by me. They need to be smarter and they should learn from the information I feed it.