23 June 2009

Curated v. crawled data

Technology Review discusses at length the launch of Wolfram Alpha, a computational engine that answers users' questions from web sources, rather than crawling indexed pages and generating a list of references, as current search engines do. The results of Wolfram Alpha queries are described as 'detailed, intelligent, and graphically stunning' (although a current shortcoming is that these results are not linked to the original material that generated them, something that will come). I was impressed by its ability to create a nutrition panel for a recipe (shown above) just because it taps into research I have done into people's use of nutrition information.

In order to generate these kinds of responses, data must be hand-curated, with the result that, so far, there are many subject areas to which the engine can't respond (Wolfram Alpha is currently biased towards 'geeky domains'). At some point, some of this curation will surely have to be handed over to a community (more like Wikipedia) rather than limited to central control. Additionally search queries have to be carefully phrased as the interface has limited natural language processing. On the other hand Wolfram Alpha shows the way search engines will go (according to the article, currently not as fast as Sergei Brin had hoped; and a nice aside, Brin interned, as an graduate in 1993, at Wolfram Research). Elements of computational search have been introduced to popular search engines: Yahoo's SearchMonkey and Google's richsnippets and Google Squared, which extract some data from indexed articles.

There's a completely different approach to answering questions/making decisions at the newly launched Hunch. Unfortunately the experience and results get a pretty rough ride at Techcrunch, although my trial of 'What shall I eat for lunch' (note time of posting) was quite engaging. It came up with an answer I didn't agree with (suggesting I eat out, but I don't have time) and I can see how, with a cohort of members prepared to edit decision trees, it could become quite fun (although maybe not for important decisions!). It seems to me that some of the many decision points in the path to a final decision have different weightings and I'm not sure the system can really take account of that.

And Techcrunch continues to monitor Bing's campaign to take Google's market share. I note that I have stopped my daily image check of Bing's home page. Either I'm too busy or it simply isn't enough to pull me away from my Google habit.

No comments:

Post a Comment