Inferential data

There are tools available today to identify the data on 'the page' and aggregate, gather and mash it all together for other purposes. Some companies are building themselves a lot of buzz about their ability to give you automated tools to identify, extract, re-organize and re-present data in various ways. Its cool - it does solve problems (aggregating financial data from hundreds of banks etc). It starts with data (in a tough place to reach) and ends with data (in an easier place to reach).

365 Media does this and more. We work with what I call Inferential Data or Hidden Data. Data that is not data until the (clever) system INFERS that a data event is happening and then creates the data out of the non-data source. In other words it starts with non-data (in a tough place to reach) and ends with data (in an easier place to reach). Example; a press release that says that 365 Media has a new CEO because John Tilly has left the company to play golf professionally (it's well overdue) and is being replaced by Mr M. Mouse. This is a string of unstructured text. A system that takes that unstructured text, identifies the subjects and then changes or creates data to reflect the event that is embedded in the text is a case of Inferential Data being discovered and manipulated.

That's what 365 Media's system does, in real-time.

Eye of the Beholder

One of the things we do at 365 Media is real-time sentiment analysis of commentary-based content. We're due to launch a very powerful social media monitoring system in the next few months, so we're facing these challenges right now. Maybe that's a secret.

Here's the challenge - in an expressed sentiment, can you truly isolate an objective sentiment? Let's say in a review a restaurant critic says that the portions are small - is that good or bad? I have an "issue" with the size of some portions in restaurants, we really don't need a 1,200 calorie starter. I hear about a restaurant with small portions, I think it must be high quality. However, I remember a work colleague of mine complaining that the portions were small at a restaurant near our office (they weren't, neither was he).

To rate the sentiment of a comment, we have to have a perspective. The solution in the case of any gray area for a software program is lots of perspectives and pick the average. And so we come back to the chess player in the box.