My Photo

« Data As Art | Main | ICWSM Data Workshop Papers »

March 18, 2009


Sean Fitzgerald

There have been, even recently. Take a look at the CALO project.

Aaron deMello

Why hasn't it happened? Because its hard!


It has happened....but it's a more complex solution than simply UI. It's a sweet little combo of UI and much more...


I'm actually working on something like this, basing it on twitter for now but it will eventually grow and span into other protocols.

Barak Hachamov

I've wanted to see a post like this for a long time. props for taking initiative on the topic.
I agree with the need and challenge, but THE KILLER APP DOES EXIST. simply not the way u've described it. It's something I like to call 'DIGITAL INTUITION'. A solution that combines AI and UI to rank (not filter!) content from your streams (between and within), leading users to disregard 'big lumps' VS. 'atoms'.

I'm inviting you to take a look at my company my6sense (alpha).


um, there already is a summary level in between. when i go to gmail i get a list of all my unread emails: who they are from, the subject, and the first 100 or so characters. with this information i can decide which i want to read immediately, which i want to save for later, and which i want to delete outright.

Daniel Tunkelang

I'm not sure that clever ranking of the atoms is the answer--I've criticized this approach at some length in a tech talk I gave at Google:

I see two better alternatives, which may actually be complementary approaches. The first is to use attention bond mechanisms to let the "market" sort out what's important. They're originally intended to fight spam, but I think they have more general applicability. The second is to offer more useful summaries than the minimal quantitative ones, using text mining and other information extraction methods to summarize and organize into those big lumps.

Eric Yeh

There's at least two problems at play here: one is hoping to get a gist of the pile of data you do have, and the other is to understand why the elements/atoms selected for view are important, i.e. "buy-in" on the part of the user.

The first is usually done either through selecting exemplars from the set of input instances (i.e. text summarization, sentence simplification, etc), but the filtered results often don't have the ability to impart the same kind of understanding that reading through it yourself can give. Of course, the ideal (and expensive) solution would be to have a human expert read through everything, and produce a concise report that conveys the important trends and elements of interest from your data pile.

The second issue is usually either addressed by some measure, and is usually conveyed via some statistical visualization. However, at certain scales, unless some extra thought was put into finding the right dimensions to visualize, they can be just as confusing as the original pile of data itself. What could work better would be some form of concise explanation of why an atom or selected element of the data was important. There's certainly been a lot of work done in visualizing lines of argumentation, along with generating believable explanations (from proof trees, classifier feature weights, etc), which could be applicable, but AFAIK there still needs to be quite a bit of work to get automated versions presentable.


This hasn't happened because to solve the problem to the user's satisfaction you'd need General AI, and not just that, but it must also have a lot of up-to-date knowledge about the world. We aren't quite there yet. In fact, I'd argue that by and large we aren't even on the way.


I'm afraid you are a little bit behind. The answer lies in AI, not UI.

The comments to this entry are closed.

Twitter Updates

    follow me on Twitter

    March 2016

    Sun Mon Tue Wed Thu Fri Sat
        1 2 3 4 5
    6 7 8 9 10 11 12
    13 14 15 16 17 18 19
    20 21 22 23 24 25 26
    27 28 29 30 31    


    Blog powered by Typepad