How easy would it be if all consumer-posted comments online could fall into simple categories, positive and negative? If they were this black and white, social analytics software companies would not have the difficult task of reading into all of the grey areas of consumer sentiment. The truth is, however, there is a wide range of consumer sentiment that puts even the most sophisticated natural language processing technologies to the test.
According to Catherine van Zuylen, VP of products at the social analytics sofware vendor Attensity, there are seven main difficult comment-analysis problems.
First, there is the false negative which include words such as “crying” or “crap” which would imply negativity. However, when a consumer uses these words in phrases such as “crying with joy” or “Holy crap!” the software can be fooled.
What about a compound sentiment? These mixed comments such as “I love the style, but I hate the color options of these work shorts” can leave analytics software stumped. Likewise, a conditional sentiment such as, “The service was sub-par, but I got a great deal” can be equally as difficult to read.
There are scoring sentiments which leave vendors with the difficult task of determining how positive a sentiment is. How much better is “I really like it” than “I like it”? How does “I love it” measure up?
Sentiment modifiers like emoticons can often trick software as they can be used to enhance positive or negative comments or, in some cases, they can imply sarcasm.
Another challenge are sentiments that come from other countries and different cultures. People from different cultures are likely to comment differently or even use different emoticons, which can make reading already complicated comments even more difficult.
Advanced software systems can be designed to work through these types of problems, but no system can ever achieve perfection at this time. That being said, it is best to focus our efforts on analyzing and reacting to clear-cut sentiments.