In the real world, when we think about sentiment we think about a person and their opinion. Joe hates X, while Jane loves it, and Frank is on the fence. When we want to reward our best customers (our brand advocates) or convert our detractors we again think about Joe, Jane and Frank. Yet in the virtual world of social media, the software programs that try to identify sentiment frequently only think about the content. The content is negative or the content is positive as though the content had a mind of its own. Actually if we really think about it, its kind of absurd. Now I appreciate that incorporating social analytics into the equation is non-trivial, however it’s hard to imagine us making any headway at all vis-a-vis social media engagement until we put social analytics (the people) up front and center. Some of the interesting challenges are:
Social ID Disambiguation
Now I mention this first because it is clearly one of the suckiest problems we have to solve in building social profiles, and while I am not an expert in this space I am glad to say that we have loads of people out there who are. Historically organizations from the intelligence sector to the gaming industry have been using a variety of techniques, such as IBM’s Identity Insights, to find patterns in data and use those patterns to disambiguate people, such as connecting aliases and the likes. Sound familiar?
Translating Content Sentiment into Personal Opinion
Just because Joe happens to say something negative about X doesn’t make him a detractor. However if it comes from someone who is normally positive about your brand (maybe an existing customer), then even a single comment could be important. Similarly a positive comment from a brand detractor is potentially an opportunity to convert someone from negative to positive (or at least neutral). So how do we take the aggregate of sentiment signals a person leaves out in the socialsphere, and in combination with other features, such as affiliations, translate this into an opinion classification?
Not all opinions are created equal
As one of my childhood teachers used to say “Ms. Wallace, Empty vessels make the most noise” and this also goes for social media. How do we measure the relative weights of a single positive comment from Jane against a load of negative comments from Frank? Perhaps Frank is an eternal pessimest, loves to blab, is on 20 different social channels, and generates a truckload of noise? Ok, now I know you are going to say “Its the influence, dummy!” (ala James Carville) and indeed this is one aspect, however its only part of the solution. For example; if Frank happens to be closely (and positively) affiliated with Oracle and he says something negative about DB2, you will clearly discount the statement to some degree but may also want to respond in a specific way. You minimally would want to know that he is an Oracle supporter before you respond. Whereas if Jane is positively affiliated with DB2 and says something negative about DB2, in that case you may have a customer support issue. However, if Jane looks as though she is a customer and has just recently registered for a competitors event (ala Lanyrd), then you might have a sales issue. The social attributes, network of relationships, historical opinions, topics of interest, all combine to give you a Social Profile for an individual that has to be considered when thinking about sentiment and how to respond.
Filling in the gaps in the conversation
This is critically important in order to make an informed decision about a sentiment signal. When monitoring / listening to a stream of social media for positive or negative mentions of your brand (filtering), you frequently end up with a really strange collection of incomplete and ambiguous conversation snippets. In order to really know what is being said and the context behind it you need to fill the conversation gaps. This means going back to the data source and grabbing all related conversations, perhaps even going so far as to grab all conversations of the associated network so that a complete picture can be determined. In this scenario, more is definitely better. And this is of course where Big Data comes into its own as this is clearly not a problem you are going to solve on your laptop.
With the joy that is Big Data — in IBM we have our Infosphere BigInsights and Infosphere Streams products — for the first time we have the opportunty to grab all this deluge of information and make sense of it. To paint a picture not just of the content but also the opinions, perspectives, and agendas of the people creating that content.
Content is simply a set of signals we are putting out into the socialsphere to confuse the heck out of the software programs :-)