Identificação de padrões em textos de mídias sociais utilizando redes neurais e visualização de dados

Abstract

At present there is the generation of a large volume of textual data, and part of this volume is generated by so-called social media, where people connect, exchange information and experiences.These data contains valuable implicit knowledge, which can be extracted and analyzed according to the media selected and the type of knowledge wanted. The objective of this work is to demonstrate how to use data mining resources, analytical tools and neural networks of the type Self Organizing Maps (SOM) to perform analysis on textual data and knowledge generation. There will be two approaches: knowledge for the educational area (with data from Question and Answer sites, or simply (Q&A))) and trend identi_cation (with posts in microblog Twitter). Both sources are similar in that they have an unstructured text format. Based on an array of terms generated through Text Mining techniques, originated in a base composed by unstructured text, the posts were the basis for training a SOM network, and with this trained network it was possible to generate visualizations that allow to perform semantic analysis of the terms and questions grouped together and use them to identify the desired knowledge. The results obtained were: to demonstrate that questions about similar subjects can be grouped by their similarity of terms, and to visualize these groupings in the form of word clouds, allowing the semantic analysis on the grouped questions.