Last night we ran a little experiment during BBC Question Time aimed at measuring the volume and sentiment of tweets during the programme.

You can view the various stats in our Twitter feed. We hope you found it useful and would love to know your thoughts and feedback.

Several people asked us how it all worked. The aim of the experiement was mainly to test various tools and technologies (that we will be releasing in the near future) around a confined timeframe/event and population (those viewing and commenting on the event).

The goal is to shake and open up the way analysis is done, to measure the pulse of stuff now, not tomorrow, and most importantly to eventually empower anyone to contribute to an analytical process.

Full details will be revealed then, so we can’t share too much while we work to prepare the tools for launch as we want to make sure you find it valuable, but in the mean time, here are a few notes about yesterday:

• We measured the volume of tweets around Question Time - the challenge here is to go beyond the hashtag (which quite a few people actually don’t use) by dynamically looking for relevant references within tweets (for example the name of a panellist). At the close of the programme there were 53,500 tweets counted, coming in at a rate of 12.49 tweets per second.
• Dynamically determining a population: during yesterday’s analysis we looked at three populations: 1) a pre-determined set (eg MPs, PPCs, journalists, bloggers and news sources) 2) the evolution of sentiment within a determined and relevant sample (eg how sentiment evolved amongst 100 people before, during and after the programme) 3) an evolving sample: those watching and commenting on the programme.
• Sentiment and trends analysis: the starting point here is a network analysis - determining relevance, reach and influence within the various populations, determining sentiment and linguistic patterns within this network, automating the analysis, search and measurement of these patterns and emerging trends within the analysed content, and finally measuring and sharing the results with the world in a meaninful and timely way.
• Oh yeah, trying to do lots of the above in real time.

All this requires still a lot of tweaking, experimenting and “training and learning” for the machines, but we’re excited by the potential, and most importantly we’ve only just began to scratch the surface of what will be possible.

Stay tuned

Posted at Fri, Oct 23rd 2009, 10:21
permalink