Currently you can only add keywords to the intelligence trainer for a feed by highlighting the words in the title of an existing story in the feed. I would like to be able to manually enter a list of keywords for a feed up front - without having to wait around for those keywords to actually be used in a story title.
Great idea. You won’t even begin to believe how many ideas I have planned in this vein. In fact, this was the original impetus for NewsBlur. I mentioned this in another thread, but Social is NewsBlur v2, and Intelligence is v3. Intelligence will include lots of doodads like this, including showing you a bunch of extracted keywords, ngrams/bigrams/trigrams, and tf/idf phrases that you might be thinking before you think it. The reality is that this takes time, so we’re looking well into 2012 for this stuff.
What about Bayesian filtering on all the words in the post? Vote a post up or down on whether you find it interesting and all the words and other features of the post will be passed to Bayesian filtering, in the same way that Thunderbird classifies emails as spam or not.
Personally I prefer the explicit model that NewsBlur has now. I don’t trust blackboxes to predict exactly what I want. This could be an interesting alternative option. I just wouldn’t want it to be the *only* option.
Well the black boxes need to be designed well. I’d much prefer to use something like this that misses once in a while than have to waste a lot of my time manually adding keywords to a filter (which will also miss fairly often).
Ahh, you asked about the collaborative filtering approach in another thread. Anyway, just to mention it here as well, I am willing to try this approach later next year, but it’s far lower on the priority list for right now.
Just to be clear, Bayesian filtering is based only on my own preferences and the words in the post, while collaborative filtering is based only on the similarity of my preferences to the preferences of others. The two could probably be combined, too.
I’d like to revive this idea, I would love to use manual keyword entry to filter in OR out articles. While the trainer is useful for feeds whose authors are diligent about tagging, there’s many feeds I use that don’t have useful tagging.
I know this is an old thread. Is here any reader out there that uses bayesian filtering beyond cream and fever? I really like cream but I want my reader to sync. Really wish an online alternative such as newblur offered bayesian filtering. My experience with cream makes me think it really does work but you need to give it time.