Precocious identification of popular topics on Twitter with the employment of predictive clustering

Neural Computing and Applications. 2017. Vol. 28. No. 11. P. 3317–3322.

Gromov V., Konev A.

The present paper outlines a novel approach to predict popularity of topics for social network Twitter; the method is designed to identify precociously the topics able to demonstrate “explosive” growth in popularity. First of all, the predictive clustering method ascertains real (not written in hash-tags!) topics of tweets and then predicts popularity rates for the topics. The same clustering algorithm is employed both to ascertain the real topic of a message and to cluster segments of time series (in order to predict topics popularity), namely, maximum likelihood adaptive neural system based upon modelling field theory. In the course of wide-ranging simulation, typical variants of “pre-explosive” dynamics were revealed; some of them were turned out to be equal to heuristic techniques to predict topics popularity well known for PR community collaborating with the network (“crab,” “Pesavento’s butterfly,” etc.).

Keywords: twitter, Twitter time series, Predictive clustering, Modelling field theory, Topic popularity