Is Twitter Text Analytics Descriptive or Predictive?
I think we are still in the early stages of mining the wealth of knowledge and insights that can be gained from text analytics for predictive purposes. Text analytics and especially natural language is more complex when compared to typical machine learning analysis of numbers and models used to classify or cluster. Dialects and nuances in speech vs writing makes it hard to understand sentiments.
We are presently in the former phones of text analytics mainly focussed on descriptive methods, which help us understand the lay of the land in terms of the ideas or themes presented in the text and engaging insights (Struhl). Descriptive methods use cases could be in areas of customer support, marketing strategy development, product development, etc. where insights from text analytics could be leveraged.
Predictive methods focus on seeking to forecast patterns or trends based on the combination of words, and sentiments. To leverage the power of predictive text analytics, one must have knowledge of context and concepts, the people who created the text, their intentions, related entities and events, and possibly access to additional information to understand if holistic predictions can be sufficiently made with the textual data at hand (Struhl).
Twitter Analytics — Descriptive or Prescriptive?
Since its launch in 2006, Twitter has now become a ubiquitous part of our daily reads for most of us either as followers or tweeters. What’s trending on Twitter could range from current political to sports events, celebrity news, and world events. In my opinion the study of trending tweets can be leveraged for both descriptive as well as predictive purposes; rather than categorizing into any one of these given categories.
From the few images that follow, it is clear that trending tweets and/or hashtags can provide a glimpse into the cultural tastes, attitudes and sentiments. These trending topics can range from funny to serious, informative to useless, keeping us entertained and aware.
Trending topics can vary from state to state and differ between countries, though there are certain trends that have been global in nature like #covid, #blacklivesmatter, #brexit, #thedress (remember the blue and black debate). Some that have gained traction have been hitting the emotional chords like #cupforben which touched the world trying to find a cup for Ben, who suffers from severe autism and refused to drink from any other cup than his sippy cup. Tommee Tippee company ultimately got this message and make 1000 cups just for Ben. Some topics like #talkpay become popular in 2015 for a short burst and achieve their purpose to generate awareness about gender wage gaps especially amongst minorities. Anonymous posts could further give insights into any reservations or fear factors of community members. On the other hand, some trending tweets could be just plain like wishing celebrities #happybirthday.
Analysis of Twitter trending topics can help understand audience sentiments of any local, national or world events and what topics are being amplified and the reasons behind it — what’s inspiring, what’s worrying, what’s humorous, what’s innovative, etc. Twitter Text Analytics could be used to infer information, for example, ratings of TV shows and movies based on proxy data (descriptive use). Given that a majority of celebrities have Twitter handles, success of movies and potential movie sales can be predicted based on the cast, movie reviews, fan following, viewer sentiments to predict if the movie will be a blockbuster or a flop (predictive use).
References
https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.1066.1810&rep=rep1&type=pdf
http://www1.cs.columbia.edu/~gravano/Papers/2011/jasist11.pdf
https://blog.hubspot.com/marketing/best-hashtags-twitter
https://buffer.com/resources/trending-topics/
https://money.cnn.com/2015/05/01/technology/talkpay-twitter-pay-transparency/index.html
https://www.quirks.com/articles/a-look-at-predictive-and-descriptive-text-analytics
https://bluetoad.com/publication/?i=379852&article_id=2700120&view=articleBrowser
