Editor’s note: Seth Redmore is CMO of software firm Lexalytics, Boston.
Qualitative data reigns in the world of sentiment analysis. While quantitative data can give us the stats and numbers that look great on charts, qual gives us the reasoning and behaviors that lead to those stats. When we truly understand the sentiment behind someone’s communications, we can understand their motivations and predict their actions.
Marketing is all about knowing your customer and sentiment analysis makes that possible on a deeper level than typical customer analysis approaches. Sentiment analysis draws on a data set of written communications such as e-mails, tweets or op-eds. It then identifies someone’s motivation and potential subsequent behaviors using three interlinked, overlapping paradigms: sentiment, intention and context.
Let’s take a closer look at how bringing the three paradigms together can show us an individual’s feelings, potential actions and how businesses and organizations can respond.
Gleaning sentiment
Sentiment is the expression of how you feel about a situation, event or person. Sentiment analysis is about written tonality – the feelings you put on the page. It can be positive, negative or neutral or tied with an emotion like joy or frustration. It seems simple but sentiment is slippery. It shifts from sentence to sentence, paragraph to paragraph and subject to subject.
When it comes to analyzing sentiment, application is key.
Let’s look at text from the Deepwater Horizon oil spill. We can analyze it at various levels of granularity: the document level, category level or entity level. Each will tell us something different.
An oil spill is inherently negative. Think headlines like, “BP oil spill trashed more shoreline than scientists thought” or “’Deepwater Horizon [film] shows human side of infamous BP oil spill.” But the sentiment associated with companies or individuals embarking on a cleanup effort is likely to be positive. The former article talks about scientists “shedding new light” on the disaster and the latter uses terms like gutsy, culturally connected and joyous when talking about reenacting the event for a film. If taken at the document level, these terms could incorrectly skew the overall sentiment around an oil spill toward the positive.
By getting granular with the analysis, we can measure the ebb and flow of sentiment throughout a text – correctly associating positive and negative sentiment and attendant emotions with the relevant subjects. Because there’s nothing gutsy about an oil spill.
While we’re not all in the oil business, gleaning consumer sentiment from written communication can influence your understanding of any target audience.
Determining intention
Next up: intention. Understanding intention is key for identifying motivation. Assessing the validity and strength of an intention requires context and a knowledge of that person’s past behavior. Gathering that knowledge requires having sufficient information about someone, which is a tricky line to walk. If someone posts a status update on Facebook suggesting they might harm someone, the decision on whether to act may come down to an analysis of that person’s history and actions. The difficulty is that people may state something in a way that suggests intent but may be nothing more than a comment on a given situation. It might even be a joke.
Having access to that information raises privacy issues and acting on it requires assigning a certain amount of credibility to a statement that may be nothing more than an expression of frustration.
Take the case of Paul Chambers. Chambers was convicted over a joke tweet where he told an airport to get its act together or he’d blow it sky high. It took two years for the case to be overturned, with Chambers maintaining the entire time that he couldn’t believe anyone would take his joke seriously.
Commentary and humor may be couched in the language associated with intention but that language is only associated with intent. It doesn’t entail it. This gets more confusing when we switch things up: A throwaway comment about liking your new shoes might actually be an off-the-record statement that you’re heading out to buy another pair.
Such is the power of language – and implicature. So much of what we do isn’t captured in what we say and so much of what we say isn’t captured in what we do.
Determining intention requires an understanding of the nuanced ways consumers suggest and request things, along with the directionality of that behavior. With that understanding we can determine what someone’s going to do – and act appropriately.
Context
From the above, it’s obvious that sentiment and intention aren’t straightforward. That’s why we need context to get the whole story. Context tells us who, what or where is being discussed and when. It’s the recipe we need to narrow down our scenarios and dig deep into the granularity we’re after.
You can determine a lot about a state or an entity with who, what, where and when. And when it comes to people, context is key for giving insight into an individual’s personality, preferences and values, as well as what that person is actually talking about.
The biggest piece of the context puzzle – what – is also fuzzy. A “what” could be an object, a demographic, a way of speaking or a feeling. “Whats” cross categories and the way we name them will vary depending on who we are.
A scientist will use a precise technical term for something the rest of us use a generic term for. A car buff might refer to a 365. For the rest of us, that number refers to the number of days in a calendar year. Auto enthusiasts will know right away that they’re talking about a Ferrari 365 GTB/4.
At the opposite end of the precision spectrum are things that get referred to only obliquely. The familiar – your bae. The taboo – the powder room. Or the kitchen gadget we can’t remember – the thingamajig. These tell us something about the speaker.
Stripping back the fuzziness of a what is a challenge. But it ties into the who and the where we need to give us the when associated with intent and the how of sentiment.
Consumer motivation
Using sentiment analysis in your research helps you identify sentiment, intention and context – all of the facets that underpin motivation. It helps us determine if a Facebook update is as innocent as it seems or if someone’s thinking about splurging on a Ferrari.
If your current customer analysis doesn’t touch on sentiment, intention and context, it may be time to consider including sentiment analysis in your research process. Many researchers are turning to – or developing – machines that can read and crunch the data found in written forms of communication. With the understanding that comes with knowing someone’s motivation and the context around it, you can provide a viable solution or a necessary – and appropriate – course of action.
Because you wouldn’t want to try and sell someone a calendar if they’re after a car.