Sentiment Analysis is when a piece of text is categorized as either positive, negative, or neutral. Traced back to the 1950s, this form of analysis has continually evolved to mirror the ever-advancing changes in technology. Yet, despite its benefits to computer science, it has tainted the text analytics industry. Why? Because sentiment analysis is not always accurate and cannot reveal actionable insights. If it does not deliver actionable insights, then it does not deliver true value.
- Text analytics companies struggle to get over 85% accuracy because, as individuals, we rarely agree on the sentiment of a particular passage of text.
- There are 171,476 words in the Oxford English Dictionary. As such, there are many millions of permutations of phrases that can be generated. Not to mention phrases that include different meanings of words, grammatical differences, or misspellings. Creating rules to categorize these permutations will not, inevitably, catch everything. Which will lead to inaccurate findings.
- The word ‘Bad’ is predominantly a negative word. But the phrase ‘not bad’ is not. In fact, ‘not bad’ may even be viewed as a neutral or positive comment. Not all sentiment analysis tools take into consideration the negation of how words encase another. Neither does this form of analysis take into account how these words alter the meaning of the content.
- Sentiment Analysis can be applied to large amounts of text. But even when applied to a single sentence, sentiment can swing between positive, neutral, and negative words several times. As a result, those looking at Sentiment Analysis are often not convinced in the findings and lose confidence in the tool.
Sentiment Analysis remains a good high-level metric. But while metrics, such as ‘80% positive sentiment’, are useful, people don’t buy products on the basis of sentiment metrics. They buy products on a sub-set of sentiment. This being emotion.
The phrase “This product is good I found it useful” has two positive sentiment words (good and useful). But this does not mean that the product will be a success and bought again. On the other hand, the phrase “This product is exciting, and I loved it” has two positive emotion words (exciting and loved). These emotion words are far more enticing to a customer. This is because we all make decisions based on our emotional response. Sure, other factors influence our decision to buy something. But unless you are psychopathic, your emotional response will have a large part to play.
But, while we make decisions based on our emotional responses, few of us actually use them when leaving a review/comment. As such, there are less of them to be found. But they are highly worth finding.
To show how few emotional responses there are in comparison to sentiment comments, Pansensic conducted a study of the airline industry. We analyzed 1.25 million sentences, which is over 8 million words. Whist 70% of the sentences contained a positive or negative sentiment, only 26% of the same sentences contained a positive or negative emotional response.
As such, while sentiment is everywhere, emotions are the rare jewels to search for if you are trying to obtain actionable insights that provide a deeper understanding of your business, product, or service.