Data Mining: Text Mining, Visualization and Social Media

UnAutomated News

An interesting update from TechMeme regarding their new use of a human in the loop. The post talks a lot about the results being better – but I don’t see any clear description of what their criteria are. The thing I get out of these sites in addition to timely notification of news stories, is an indication of how much attention a news story is getting. For me, it is very interesting if suddenly everyone is linking to an old news story because it is suddenly relevant in the context of a recent issue.

Ultimately, this direction will push TechMeme and related sites in the direction of tools to assist editors in selecting news items. Their goal is to break news stories, which is quite different from tracking what is important to bloggers and what is getting attention. Personally, I think that the latter is way more interesting as the agenda is emergent and synthetic. In the editorial model, if the editor thinks a certain story is ‘important’ it will find its way to the top of the stack. Will TechMeme be able to continue to differentiate?

TechMeme - Beyond the List

While there may be plenty to say about the tyranny of preferential attachment, TechMeme has provided a compelling application that presents data in something other than the simple list. Recently, this video appeared on TechMeme's news blog. It shows a time lapse of the state of TechMeme - stories bubble up to the top of the site, and the conversations around those stories build up around them, consuming more and more real estate.

ReadBurner Burner

ReadBurner provides a social rank for items shared via link blogs. Now is the right time to mention ReadBurner as it is itself at the top of the list of shared items due to this post on the Google Operating System.

[thanks, Justin]

Technorati Update

I've been strangely fascinated by Technorati's latest incarnation - essentially a meme tracker front page to the search engines full feature set. Part of the reason is that I really like the idea of attention ranking and think that there is lots of opportunity (not to mention plenty of challenges) in this area both in terms of user experience and algorithms and mining. It is interesting to note Peter Hirshberg's pot shot at TechMeme:

I suppose we could build a system that looks at all the posts on a topic from a tightly proscribed white list (there are many services that do this) but that loses much of the emergent serendipity of the blogosphere and excludes everyone except for a proscribed static elite. That’s neither fair nor interesting nor scalable. We're interested in what the whole blogosphere has to say on a topic. But to find the interesting relevant stuff we want to give a little more weight to bloggers that are revealed to be authoritative in a subject (not just because they say so in a tag, but because we observe many other topical bloggers linking to them in a democratic vote of editorial goodness).

I always saw TailRank as filling this space (perhaps not as elegantly as they could). The weird thing is that while Technorati is claiming to use no restrictions in terms of where they find links, the measure of attention that their highly rated stories achieve is amazingly low. Right now, the top news story has 1 link (Attention [1]). This indicates a more complex algorithm (or very light data rates), and the problem with complex algorithms is that users don't always follow why they are seeing something popping up at the top of the list. My feeling is that the solution to this problem is to discard the list for something with more dimensions to it.

Update: Well, that just shows my ignorance (or does it show an interface problem?) The attention score is not the number of inlinks found but, um, something else. There is an explanation here:

Technorati measures Attention by calculating a weighted rank based on time, number of links, rate of new links, Technorati Authority, and the Technorati Authority of linking blogs. Attention changes over time, so something that was getting a lot of attention this morning may be much less interesting later in the day. A Technorati attention-based view allows you to concentrate on those items that are gaining the most attention now, even when they were created hours ago.

Though I don't find it very clear. Attention might be a rank, in which case '1' is good. However, the story that I'm looking at right now has Attention [1] and 1 'attention link'. Confusing.

ClickMeme

Wouldn't it be interesting if TechMeme (et al), TailRank, Megite, etc. added click-through information to their displays?

The News Width

One of the basic properties of old world programmed media was the fixed time slot. This was a particularly defining feature of the news. There had to be exactly 30 minutes (or whatever) of news to fill the slot. If that day was slow, the news was padded with light stuff, if that day was hot, then the light stuff was dropped as were other moderately important features.

You would think that that approach to media would be one of the first to drop online. We may, perhaps, be forgiving of the online presence of mainstream media - there is a fixed real estate on their front page. However, having become aware of the issue, I'm less forgiving of Web 2.0 aggregators including: memeorandum/techmeme, TailRank and even our own BlogPulse. Take techmeme for example. Stories are ranked according, in part (I'm guessing) to how many citations they get from which bloggers. However, on one day, a story with citations from A and B may appear on the front page whereas on another it may not - depending not on how important that story is absolutely, but how important it is relative to all other stories.

I say that we should be looking for interfaces to information that reflect how important that information is and which don't persist artifacts of the very media that we are (apparently) trying to escape.

TailRank Info Porn

Kevin over at TailRank has put up a graph showing the cyclical nature of story volume from TailRank's POV. The graph shows the number of stories per day that TailRank ingests - that is to say, the number of threads that it considers when creating its real-time view of the blogosphere. Typically, TailRank's home page shows 30 such stories. TailRank appears to be looking at something like 6k stories on a typical week day. If we guess that there are 5 links per story (the head is going to have many more, but the long tail will lower the average number) then TailRank would be considering something around 30k posts per day. This is, however, the number of posts involved in a thread. TailRank will actually be crawling many more (though I'm guessing that it works with a moderated blog list - that is to say it doesn't crawl any old blog out there).

How Does TechMeme/Memeorandum Work?

With the proliferation of memetrackers, I'm starting to wonder more about the algorithms that they use to determine the relationships between blog posts. Initially, I had assumed that they were purely link based: If A links to B, then A is about B. I think this is how TailRank works - so the differentiator that TailRank is after is mostly to do with coverage.

However, I don't think that this is how the Memeorandum suite of trackers works. Or, at least, they have this plus some other approach. The reason I believe this is that  a post I wrote is currently listed as being in the discussion set associated with a post from Venture Beat about Powerset. However, my post didn't link to the Venture Beat post. So why is it in the discussion? In fact, my post isn't even about the Venture Beat post. The only real explanation is that Venture Beat linked to the Powerset website and so did I.

Unfortunately, I can't link to the article on TechMeme - permalinks there seem to provide a view of the discussion that is different from that given on the front page.

While it is interesting (and vital) that services like Memeorandum explore new approaches, cases like this bring in to focus issues of accuracy - how accurate is the description of the 'discussion' around a topic (which, here, means a post or news article)?