Editor’s note: Darren Borne is business development manager at Document Capture Co., a London-based data automation firm.
Text and sentiment analytics is nothing new. It has been a widely-used method of information-mining for some time, viewed by many as a fail-safe data capture process for producing valid insight from any environment where a large amount of free-written data is available.
Utilizing semantic technologies to mine insightful data from social networks, however, is a relatively new area, which marketing researchers in particular are keen to tap into. Social media, perhaps once viewed as the folly of Generation Y, has rapidly transformed into the port-of-call for brands wanting to unearth the latest consumer opinions, attitudes and habits.
With billions tweeting worldwide, posting and sharing information, the body of raw data is unfathomably immense. If that wasn’t enough, the ever-changing algorithms of the biggest search engines mean that tracking down exactly what we want can be a challenging and potentially laborious task.
So how might we go about separating the valuable data from the day-to-day ramblings? Some believe that traditional methods of data discovery are proving somewhat inept when it comes to sifting through social dialogues, due to the dynamic nature of the language being used by those participating in it. This might indeed be the case, if we wanted to analyze every piece of social dialogue ever written, or even published within 24 hours.
I believe that, in fact, we researchers are the ones who need to adapt our practices to get them in line with changing technology.
Where once we may have only had the body of data collected, tirelessly, by ourselves, we now have a relative bottomless pit of raw information at our disposal. What this means is that, although we may get less valuable data from a much bigger sample than we would when analyzing our own research data, the process can be repeated as many times as we choose, as the material at our disposable is unlimited.
Using an engine which can pick up sentiments and emotions from text by considering each word separately, linking associated words that refer to the same subject and then rating the sentiment on negativity to positivity scale is an approach that can help transform social dialogue into valid data. The sentiments relating to each subject, as well as an overall view, can then be presented on a dashboard using the usual visual techniques such as charts, heat maps and word-association clouds.
Tighter constraints should be employed that limit the amount of data collected, such as only those who mention a brand at a particular time of day or night (which may help to give clues as to location) and the semantics analyzed from there. Where available data is an endless stream, this may be akin to dipping a bucket in at different locations/times and seeing what’s pulled out, rather than trying to analyze the entire water flow.
This process gives still gives decision makers a true insight into what end users are saying and, most importantly, feeling.
Text and sentiment analytics can, in my opinion, be utilized effectively within almost any sphere in which vast amounts of free-text data harvesting is occurring. Because sentiment is everywhere, including on social media sites, review/comparison sites, blogs and even on a business’ own Web site, text and sentiment analytics can extract feedback from each and all of these mediums. These can then be collated within an analysis dashboard, allowing an organization to view feedback from all angles. Private and public sector businesses can bring such feedback into a semantics engine virtually in real time; the moment a person tweets or comments, organizations will be able to respond much faster.
Let’s look at three sectors and explore some ways that companies and organizations within them could use this type of data.
The service sector includes business such as hotels, bars, restaurants and cinemas. In this sector, customers will typically make judgments based upon the quality of service more than any tangible product – though products obviously play a significant factor.
In any sort of review, a customer will typically discuss a number of different factors and the sentiment analytics engine must be intelligent enough to decipher between these. At a restaurant, for example, customers will typically make judgments based on the quality of food, waiting times, hospitality, cleanliness, environment (lighting, temperature and so forth), comfort and parking facilities, amongst others. If that restaurant then collects comment cards from its customers after they have finished their meal in order to gain feedback, the analytics software provides a sentiment of their overall experience, as well as separate analytics on the sentiment felt towards each individual factor.
This method provides a restaurant with a well-rounded view of what its customers are saying and can provide the basis for a host of actionable insights in order to improve service in the future. Typically, it is most appropriate to represent information in the form of a word-association cloud, which facilitates a snapshot review alongside individualistic evaluation of restaurant services.
Within the retail sector, a number of major supermarkets have already found text and sentiment analytics to be very fruitful. Because data can be fed into the processing engine as and when it is collected, there are no restrictions on when the data can be viewed, which allows immediate action to be taken where needed.
While word-relationship clouds are useful to retailers, large superstores with multiple departments and customers may find that there is initially too much data to view in this manner. Instead, a heat map can provide an effective view into how customers are feeling throughout different areas of the store. Ideally, the heat map can be split by department and color-coded to represent the different levels of sentiment to give an overall view. From here, more detailed information can be accessed through clicking on the area of interest, such as an area with poor feedback, in order to give greater insight into the reasons behind such opinions.
The potential benefits of text and sentiment analytics in public and private health care are clear to see. Because health and well-being is all about how patients are feeling, it goes without saying the sentiment engine would bring to life their emotions in different aspects of care. Potential uses could be found within patient experiences and outcomes, hospital complaints and feedback from friends and family of patients, as well as other varying free-text data sets.
Already, government organizations and service providers in the U.K. such as libraries and councils run various consultations, collecting “have your say” forms and feedback on a variety of different subjects. Plugging such data into the sentiment engine would revolutionize the way in which such data is viewed.