Text Analytics: Integrated sentiment analysis
Technique hears much more than the voice of the customer.
By Fiona McNeill
Just as you follow someone’s gaze to know what they are looking at, you can follow someone’s words to better understand them. Words allow us to express how we organize, interpret and perceive. With electronic expressions of words (i.e., unstructured text data) such as that from social media, we describe how we feel, what we are trying to avoid, what we are paying attention to and what we are thinking about.
Sentiment analysis, the investigation of how someone feels based on their words, is more than listening – it is an analytic method. Before analyzing the burgeoning “big data” volume of unstructured text, the issue many organizations struggle with is separating out what may be appealing to understand from what is necessary to know. Analytics can help.
In approaching sentiment analysis, the first question to ask is: Will sentiment have an impact? Will behavior change based on what is expressed? And will that behavioral change be from you (the reader), the author or both? Will those sentiments affect what someone else does?
Based on insights from sentiment analysis, a key influencer could be engaged, a new communication might be sent or a production change can be escalated. Alternatively, knowing the sentiment may not change any behavior.
Reputation management is critical. Positive sentiments could heighten visibility and lead to cult-like followings. Negative sentiments could run a stock trend down or cause undesired scrutiny. The question becomes: What is the tolerance threshold for your organization? At what point will some new action be taken and by whom?
However, some people don’t complain when they’re unhappy with service or products. They just walk away. Without expressing dissatisfaction, others can’t know what they plan to do – particularly from what they don’t say. Sentiment isn’t a good predictor of churn in those cases.
The second question to think about is whether you care about some segments in the overall population of sentiment more than others, such as influencers on specific channels. Analytics can help you determine whether a person’s influence or assigned value warrants a response or action.
Identifying social networks is an analytic exercise, a well-practiced method used in the determination of fraud rings and directly transferable to social communities. Making the connection between the online world and a customer base is often not as straightforward, however, as a link between online identity, and a customer number typically doesn’t exist. There are ways, though, to use the information to get a pulse on localized sentiment at the segment level. And, as those of us with backgrounds in database marketing can attest, analysis within a segment is typically more effective than analysis en masse – but again, only if has a purpose.
Is the channel the message, as Marshal McLuhan would have us believe? Is the sentiment expressed in an isolated forum where birds of a feather gather to reinforce their common interests? Does that have any impact on what you do? Or is the sentiment from a source front-and-center in the eye of your shareholders, such as commentary on NASDAQ? And to what extent is it between these two extremes?
Having the flexibility to look across unstructured text information sources – including, but not limited to, those from social media, and changing perspectives to the data by variations in channel, by audience or for different topics – will significantly extend the benefits of the investment made in sentiment analysis technology.
A third key element to keep in mind is which aspect you want to investigate – say the sentiment or variations of sentiment, such as mood, degree of threat, extent to which someone cares, etc.
Sentiment Technology Feels Out New Issues
To address variation, sentiment technology is most useful when it can be adapted to examine new issues, allowing you to answer a question framed in a number of different ways or ask completely different questions of the same data.
When you can examine a question from different perspectives, you are likely serving a multitude of needs, and often you will want to share analytic results in a variety of ways – say by integrating with your current alert processes, workflows and monitoring reports. Analytic applications have the flexibility to examine all the available information in any number of ways, use different techniques to solve a problem and alternate methods to deliver the derived insights.
For analytic problems like sentiment, you need technology that doesn’t restrict you to examining the question in only one or two ways, with a single definition or with only some of the data. With an integrated approach to analytics, you have the power to get the answers from the data – answers to questions you already have, as well as those you haven’t even thought to ask yet. If you use instead a highly specialized method – for example, subscribe to a social media report that monitors a trend for you – you are not gaining any kind of competitive differentiation. When the technology is integrated so you can use different methods and approaches, your analytical environment supports innovation.
Having a variety of analysis capabilities at your fingertips is best for solving sentiment questions and, for that matter, any other analytic question. For example, by including pre-defined categories that identify specific segments, you can describe where attention is focused, based on aspects of the business that you can affect. Assessing how words are being used to describe these concepts can provide insight into the degree of attention. This kind of extraction provides details on specific features, offering much more detailed analysis. With more detailed insight, you can be more specific in defining priorities and any corresponding action.
For example, considering the tense of a verb can help identify a person’s priorities or intention. The use of emotional words, including their frequency, can identify the degree to which someone cares.
Some describe this as using both the art and science – or the human and machine – aspects of analytics. The good news is that with an integrated system, you have both. You can use your custom entities (human-defined) in your text mining (machine-learned). If the system is fully integrated, you can combine your structured and unstructured analysis simply by dragging and dropping a predictive model algorithm on discovered topics to see if they affect the likelihood of churn, for example.
Integration gives you choices in how to develop analytics and how you deploy the results. This kind of flexibility is crucial to your operations as you become more responsive to your market, your different audiences and your different topics of investigation. Most important, when you approach analysis with open-ended questions, the benefits go far beyond just hearing the customer voice. Open-ended analysis can lead to cost savings by helping you: reduce reliance on third-party consultants and shorten time to insight, resulting in more efficient business operations; scale investigative understanding without linear cost; and simplify portfolio complexity based on market interest and sentiment. It can also benefit service and sales, allowing you to: improve product design; find undiscovered latent demand; identify competitive threats and benchmarks; improve retention and offer acceptance; and identify key market drivers.
To experience the many benefits that come from analyzing social media, you need open, flexible and integrated analytic technology that can tell you how sentiment matters and in what context. Organizations leveraging the full benefits of sentiment analysis are improving results with performance benefits that go far beyond reputation management to cost savings and increased sales.
Fiona McNeill, global product marketing manager at SAS, oversees product marketing of SAS Text Analytics. In addition to working with a wide range of industries, McNeill has defined marketing strategy and corporate relationships at SAS, received multiple innovation awards and was identified as a pioneer and one of the most influential people by CRMPower in 2000. Before joining SAS, McNeill was a member of IBM Global Services.