With the start of the new decade, I am making a resolution ; I will help further NLP analyses with the goal of better understanding customers. One NLP analysis that has me particularly excited is the entity sentiment analysis. Until now, all of the sentiment analyses and the sentiment tools on the market focused on the overall sentiment of a consumer’s feedback. Because feedback often has positive and negative elements, the results of most consumer interactions converged toward neutrality. As an example, how would you classify the following statement ?
“I really love this brand but this season’s sales rep is rather bad.”
Since there are very strong positive and negative adjectives, the sentiments cancel out and the feedback gets classified in the very unhelpful neutral category. But what if there was a way to automate understanding this sentence on a more granular level ?
What is so revolutionary about this new type of sentiment analysis ? Well, by combining entity recognition with a sentiment dictionnary, these algorithms are able to break up the sentence into its many parts and retain only the adjective-noun pairs that are closest to each other. So instead of giving the entire feedback a neutral score, an entity sentiment analysis algorithm would return a very positive score for brand and a negative sentiment score for the sales rep.
As a result, when the aggregated results of hundreds and thousands of customer feedbacks come in, the marketing department will have a tallied performance score for all of the aspects of the business, as opposed to a very undefined overall consumer sentiment.
To make good on my resolution, the first step that I will take is to analyze and compare these different types of sentiment analysis. By analyzing the tweets of a large brand, I will show the difference in actionability of the two different types of analyses. Is there any type of brand in particular that you would like me to look at ?