Hacker News Reader

Comment by sunkencity | Hacker News Reader

The feed items that truly contain new information is not one trains for traditionally with bayes or so, because it would imply something that has not been seen before, rather than just being similar to something already seen. Searching for meaningfully different stuff rather than similar stuff is a whole different problem.

I'd like to have a treemap view of google reader rather than a list, and where I could visually, in real time filter the feeds by adding negative and positive keywords/tags, and click around in various levels of a treemap.

Click on "technology" => make a new treemap of significant tags in that area, click on the next tag to dive into that space etc, and eventually see all the feed items somehow. Perhaps present them in all levels by some sidebar or hoverthing. Color code items for popularity / activity / freshness.

I'm thinking something like dabble.db but with a treemap interface and cappuchino js interface.

View on HN

Here's another way to approach the user's end of it: have them log in to their Google, Facebook, Twitter, et al, and shove all of it into a single stream. Let them filter by source, topic, relevance, time, and anything else you can implement. Record their preferences by watching what items they click through or spend time on or interact with, and your relevance sort gets better. Add other users' data, and you have a fantastic recommendation engine without ever making anyone rate anything.

Another problem to solve with current news reading is duplication. Even if you do want to read about Tiger Woods, there's a lot of duplication in the echo chamber of the internet.

Yep, that's the position google has to analyze data, it's incredibly powerful to have almost the whole activitystream on the internet.

An interesting sidenote, apparently google could activate face recognition for google goggles but have chosen not to do that at this moment for privacy issues.

As much as I hate to say it, It seems like this is something that could be easily accomplished by piggybacking off of Twitter.

Just drop the link, some tags, and misc attributes along with a short description.. Is it much more complicated than that?

Remember the end goal here is to deliver content to the user that they want to spend time consuming. Why use Twitter for anything but another way to find content?

I'm not sure I understand your comparison. What is the difference between finding content I want to spend time consuming, and finding content another way (via some content filtering process), which has every intention of limiting the content to what you "want" to consume.

I agree. The problem is definitely a matter of related vs. unrelated content. My suggestion (or babble) was more so just pointing out how easily we could implement a "source content filtering" logic into an already existing interface.. If we felt that was the best way to go about it.

I just mean that I don't see a benefit of using Twitter as a delivery system.

Of course. No doubt.

If you re-frame the problem not as searching for "different gold" but "similar chaff" (and hiding it) then simple trainable filters look more useful.

I can see an interesting parallel to searching for prime numbers, which are basically our most pure "new, unknown primitive facts." In this model, a Bayesian filter trained to detect "sameness" would almost be a generalization of a sieve of Erastothenes.

That sounds like a great interface. But it doesn't stop you from having to do all of the filtering/categorizing. Maybe if after having done all of that work on your end, there would be a nice way to share that with followers.

Yep it could learn from the usage.

Nah, the only categorization that really needs to be done is by topic. It's easy to tell if something is, say, a tech article, or tech/programming, or tech/linux, based on source and keywords. If your algorithm did that once per item that's in anyone's feed, everything else is refinement. Ideally, a human won't need to touch it at all. Even if the machine gets it wrong, it could be corrected with a "this is in the wrong place" button. In any case, an article shouldn't rise too high in any category that it's not relevant for.

I'd like to see the categories automatically created via SVD. I wonder how people would react to the knowledge that their favorite posts are written by 30-something males on Saturdays, and include the word "has" twice as often as the word "my"?

I wouldn't mind an interface with some knobs and dials, given that one can escape the linearity of the feed, and replace that with a recombinant stratified interface which can present the information space from several different vantage points. It'd be a line of flight interface:

“This is how it should be done: Lodge yourself on a stratum, experiment with the opportunities it offers, find an advantageous place on it, find potential movements of deterritorialization, possible lines of flight, experience them, produce flow conjunctions here and there, try out continuums of intensities segment by segment, have a small plot of new land at all times. It is through a meticulous relation with the strata that one succeeds in freeing lines of flight...” – Deleuze and Guattari

http://www.linesofflight.net/linesofflight.htm

An interface that weird had better be devastatingly well-implemented, or people will just be confused.

Reply on news.ycombinator.com