A Comparative Analysis of Social Network Pages by Interests of Their Followers

Social networks provide people with an opportunity to form social clusters that share interests not only sporadically, but on a regular basis (circles of fans of different music, books, kinds of sports, etc.). Every circle communicates these interests creating lots of linguistic data to attract new followers and support interests of the existing ones. In the present article, we suggest a linguistic model of classifying VKontakte pages by interests of their followers and test its effectiveness with such classifiers as Naive Bayes, SVM and some other. The task of cross-classification within three particular groups sharing a specific interest (football, vegetarianism, historical reenactment) is solved with F1-score reaching 0.99 for some classes. We also suggest a clustering procedure to retrieve texts of these groups from a collection of up to 1,000 random texts. When we enrich linguistic data with the help of DBpedia Spotlight, clustering becomes more effective.