Attribution of professional interests of social network users based on subject-oriented groups and personal pages

The present study is part of an interdisciplinary project that unites the efforts of sociologists, linguists, mathematicians and IT specialists. The authors' ultimate aim is to create a system for predicting the career choice of prospective students that would use automatic text analysis of freely available and topically unconstrained data produced by school leavers in social networks. The current stage presents the testing of a hypothesis that user data of prospective students in the social network VKontakte contain information about their interests for a certain subject, and this information can be formalized and become the basis of automatic models for determining the career choice of text producers. The main source of data for the prospective student's linguistic portrayal is the wall of the personal profile which contains both original posts and reposts that are shared by the page owner. The first stage of analysis consists in studying posts and reposts on the walls of subject-oriented communities in VKontakte which were matched to three subject areas: mathematics, humanities and natural science. This content was used to determine which lexical markers could differentiate between texts of different subject areas and serve as their markers. At the second stage of analysis, the authors used the textual content of freely available user walls that belonged to TSU first-year students. The principles of organizing primary textual sources consisted in dividing them into the three aforementioned subject areas. Downloading all the texts from a user's wall was highly important, since the material was used to test the hypothesis that the set of lexical markers discovered at the first stage could be used to automatically classify the texts of thematically unconstrained communication of VKontakte users. The statistical analysis showed that it was possible to apply the markers in text attribution tasks according to the three subject areas the prospective student might be most interested in. Among the classifiers tested using the markers derived in the first stage, Logistic Regression proved to be the most successful in dividing the texts into three classes: humanitites, natural science and mathematics; this also proves that the subject-area thesauri used functioned correctly. Overall, the study shows the effectiveness of using automatic text classification of thematically unconstrained communication in VKontakte with keywords derived from texts belonging to particular subject areas. The problem that awaits its resolution is the relatively low discriminatory power of humanities keywords, probably due to their widespread usage in social network communication on the whole.

Keywords

Authors

References