Resumo

HERNANDEZ-PAJARES, Beatriz; PEREZ-MARIN, Diana  e  FRIAS-MARTINEZ, Vanessa. Visualization and Multiclass Classification of Complaints to Official Organisms on Twitter. TecnoL. [online]. 2020, vol.23, n.47, pp.107-118. ISSN 0123-7799.  https://doi.org/10.22430/22565337.1454.

Social networks generate massive amounts of information. Current Natural Language techniques allow the automatic processing of that information, and Data Mining enables the automatic extraction of useful info. However, a state-of-the-art review reveals that many classification methods only distinguish two classes. This paper presents a procedure to automatically classify tweets into several classes (more than two). The steps of the procedure are described in detail so that any researcher can follow them. The accuracy and coverage (instead of only coverage as usual in the literature) of two automatic classifiers (SVM and Random Forests) were analyzed in a comparative study. The procedure was applied to automatically identify more than two types of complaint from 190,000 tweets. According to the results, Random Forests should be used because they achieve an average accuracy of 81.46 % and an average coverage of 59.88 %.

Palavras-chave : Text Mining; Multiclass Classification; Social Networks; Twitter.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )