Intelligent Platform for Concern Assessment Based on Micropost Utterance Classification by Topic of Interest

摘要

Means of communication on the Web have become widely spread and extremely used, e.g. social networks such as Twitter. One of the purposes of this use is to inform, discuss and give an opinion on political topics and social issues. However, relevant information from this system is fuzzy and surrounded by an important amount of noise, which makes it difficult to assess concern from the population toward social issues on the Web. Popular trends do not cluster around social issues, and tools such as social search engines do not provide the required functions. Thus, we propose a system that can be able to process the noise and collect the important information by classifying utterances for later concern assessment in the form of clustering and retrieval. We deal with scalability in the processing of microposts by taking advantage of hardware specifications, and test three algorithms to classify and sort relevant utterances by topic. The tests show that contrary to initial belief CNB does not perform better than NB for per-topic tweet classification, and that topic refinement improves results overall, while logistic regression stays fairly unaffected by this feature clustering.