There is a lot of chatter these days about analytics related to social phenomena: social media analytics, social network analytics, etc. My impression (after talking about it with some folks who specialize in such things) was that there was more talk than action on the subject. But I keep hearing about organizations that are actually doing something, including making connections among the various forms of social analytics. So it’s definitely worth a post, and we’ll be starting an IIA research project on this subject as well.
Let me first say that social media analytics are not for the faint-hearted! The idea behind using automated tools to monitor social media is to reduce the human labor required to read and classify blogs, tweets, comments, and so forth. Yet a fair amount of smart humans will be necessary to help set up your analysis in the first place, and they’ll also be useful as a check on automated analysis.
What’s the current art of the possible in social media analytics? The most common application is sentiment analysis, in which a company can find out whether customers are saying positive or negative (or neutral) things about your company or your brand. That may sound easy, but it’s not. It requires that you can classify ambiguous adjectives—“sick,” for example—as to whether they are positive or negative. And if someone says a hotel pool is “cool,” are they complimenting it or complaining about the temperature? You’ll also need a taxonomy of terms that are used in your industry. All of these classifications require some degree of textual analysis—and people who know how to set it up—to establish the context of the words.
Even after an automated analysis has been set up, you may want to use humans to check up on its accuracy. One analysis of automated classification suggests that automated analysis alone is often inaccurate in classifying terms. In order to address this issue, SAS’ social media analytics offering, for example, comes as a service, rather than just software; one analysis suggests that it includes consulting services to improve the accuracy of classification.
In addition to positive or negative sentiment, social media analytics can also create lists or charts of keywords associated with your company or brand, and rank them by frequency. Nestle has examined the attributes of its brand. JetBlue ranks common complaint terms in order of frequency and even converts online text into a “net promoter score.”
The next level of sophistication is to track relationships among terms. Companies may want to know whether one term associated with their company or brand is often found in the same message as another. A candy firm, for example, analyzed social media content to determine whether “chocolate” is positively associated with “health.” Whether chocolate is really good for you or not, the company found a positive association.
The next level of analysis is to determine whether the people who are saying good or bad things about you really matter—i.e., their level of online influence. Maybe their blogs say bad things about your company, but does anyone read or link to it? If they’re very critical and very influential, the only answer is to put out a contract on them—just kidding! Maybe you should try to fix the thing they’re criticizing you for.
Relationships between people don’t only determine influence, but also buying behavior. A few companies are talking about combining social media analysis with social network analysis. If you find some customers who are saying positive things about you online, and you know some members of their social networks, you could target promotions or offers to them. Perhaps this is happening already, but I don’t know of any concrete activity.
Social media analytics have a high potential payoff, but they require substantial work. In order to make them pay off for you, they require not only software and hardware, but smart people who know about social media, textual analytics, and matters quantitative. If you’re serious about social media analytics, look for those people now.
