Monday, August 26, 2013
Making sense of complex social media data
Before I left on vacation, I took part in a meeting to review an analysis comparing Twitter data with that from tracking studies. In some cases there was a relatively strong correlation between the two with social leading changes in the attitudinal data. In others, the relationship appeared to be reversed. Now Nielsen and Twitter report similar relationships between Twitter data and TV ratings.
In brief, the Nielsen study examined Twitter commentary and minute-by-minute Nielsen ratings for 221 episodes of prime-time shows on major networks. The key findings were as follows:
- A change in Twitter commentary was shown to cause a “significant increase” in ratings 29 percent of the time.
- A change in TV show ratings had an effect on the volume of related Twitter commentary 48 percent of the time.
By inference, this suggests that in a fairly large proportion of cases no significant relationship was identified (we cannot assume the two sets identified above are mutually exclusive, so the proportion showing no relationship is probably greater than 25 percent). Irrespective, the findings are interesting and point to the degree to which Twitter and other social media are now an integral part of the media scene, not some weird digital adjunct to it.
An article by Cotton Delo in AdAge highlights an important issue. Identifying a general relationship between Twitter and TV tune-in is one thing, understanding whether a specific program will benefit from or boost Twitter commentary is still lacking. Delo cites the example of “Sharknado” which evoked a lot of Twitter commentary but relatively little tune-in (just reading the outline of the plot might give you a clue as to why that was; not everything needs to be explained by research data).
When it comes to looking at the relationship between Twitter data and brand and advertising tracking metrics, there are many issues to be resolved. Not the least of these is simply organizing and cleaning the Twitter data to ensure we are looking at commentary that really does relate to the brand (see Anne Czernek’s POV for more on this topic). Then we need to make sense of the commentary using a framework that allocates those comments to different brand equity “buckets” related to meaning, difference and salience. Given the scale of the Twitter data involved, the classification is done using supervised machine learning. Only once that is done can relationships be reliably identified between the two data sets (see Bill Pink’s POV on Big Data for more on this topic).
Finally, even in the cases where Twitter data appears to lead brand metrics, we can’t assume the relationship is a direct one. One big possibility is that the Twitter commentary is simply responding to traditional media spend. For instance, a TV ad that conveys a compelling and memorable impression of the brand may both shift opinions of the brand and get people to talk about it. Superficial relationships between data sets may only hint at even more complex relationships in play.
So what is your experience of using Twitter and other social media data for analysis? Please share your thoughts.
This entry was posted on Monday, August 26, 2013
and is filed under Digital, Research.
You can leave a response.