Google

Wednesday, 9 March 2011

Conceptual Search: Putting "Mean" Into Meaning

Wordle: Conceptual Search
This evening I caught the SCL's Annual Lecture. This year it was expertly delivered by Dr Mike Lynch OBE on the "Advent of Meaning Based Computing".

I want to say that "meaning based computing" is the enterprise version of the semantic web, if only to keep 'price comparison' web sites in the technological cross-hairs. But I can't really, as it wasn't even mentioned.

According to Mike, the more fundamental Information Technology challenge is that the world is producing far more human-friendly, unstructured data than structure data (compound annual growth rate of 62% vs 21%). Cloud computing might help scale the technology aspect, but that doesn't help you find the right information. Keyword search, PageRank, meta data and so on all help with the data. But if computers are to search more comprehensively, they must 'understand' the meaning, concepts or ideas you're looking for, and how these relate to one another.

Conceptual search is partly constrained by who is searching, and whether they know what they're looking for. Boolean and linguistic search models in particular require 'training' and lots of maintenance by very smart people who know a lot about what's being searched for. But Mike explained that probabilistic systems are independent of language and investigator bias. Instead, they process the whole data set and look for scenarios where words are 'more likely than chance' to appear together. In this way, a more objective set of search results are returned.

So what?

Well, Mike says the main applications for conceptual search currently seem to be amongst spooks, regulators and major corporations. In other words, it's a mean weapon in major games of cat-and-mouse - one that US financial markets investigators seem to have spent five years perfecting. In fact, the most worrying statement of the evening for some must be that "most of Wall Street's messaging is in the cloud".

Specific uses lie in improving document management, retention and electronic discovery, particularly as an aid to early case assessment and crisis management. A killer app, no doubt, and one where Mike reckons law firms can indeed make a killing - possibly providing outsourced data systems for their clients. But pre-programmed, conceptual search systems are also good enough to enforce corporate policies real-time, preventing 'smoking gun' emails from ever being sent, or corporate bribes being accepted.

If only they could help find decent car insurance, utility and mortgage deals, we'd be shot of price comparison sites as well ;-)

No comments:

Related Posts with Thumbnails