4 Sep 2009

Listening in (corporate eavesdroppers).

We spend quite a lot of time at Harvest using and assessing monitoring tools such as Jodange, Radian6, SentimentMetrics, and other like them.  Measuring ‘sentiment’ is a hot area for good reasons, as the ability to ‘listen’ to your consumers’ conversations can’t be ignored (is in fact more like a responsibility), and new applications to help us do this seem to be coming out with dizzying frequency.

This morning I have been tyre-kicking www.twitratr.com which looks at twitter data and categorises tweets using negative, neutral or positive keywords.  Simple stuff - no-one could accuse Twitratr of over-complexity - providing a nice snapshot view for big brands (benchmarking territory this).  However, dig into the detail, and holes start to appear…  here’s a search for UK supermarket Sainsbury:

Sainsbury Twitratr.com failings

The top Positive mention has misinterpreted the word brave, and should really be Negative. In the Neutral column, the word horrid in the 4th tweet down clearly marks this as also Negative.  Finally, looking at the Negative colum, the top tweet is nothing whatsoever to do with the supermarket…

Does this mean automated monitoring of this kind is inherently flawed?  I don’t think so, just that you need a large enough data-set before forming theories or drawing conclusions.  And you need to work to improve the rules / algorithms - changing the categorisation of horrid from neutral to negative is an easy fix…

It’s not just microblogging data streams that are being diced and sliced either.  I discovered yesterday that online bank First Direct uses voice recognition to turn ALL the calls through their call centre into a data-stream that is analFirst Direct logoysed by keyword.  Their system then picks up on certain words to trigger responses (customer service measures of one kind or another).  You can see how this data would also give First Direct early warning of problems with, for example, their website or mobile services.  So if you’ve ever wondered what that recorded voice telling you that 'Your call might be recorded for training purposes' was really on about, you now know!

A few years ago voice recognition was widely talked about as life-changing technology, but for most of us this never happened.  Even poster-child for the sector, wireless voice-to-text service Spinvox (Corporate tagline - ‘What we say and how we say it is a wonderful and powerful thing’.  Quite.) is now embroiled in controversy after it emerged that some proportion of its ‘technogy’ is in fact a call centre in Pakistan.

blog comments powered by Disqus