Experimental of vectorizer and classifier for scrapped social media data

Abstract

In this study, we used several classifiers and vectorizers to see their effect on processing social media data. In this study, the classifiers used were random forest, logistic regression, Bernoulli Naive Bayes (NB), and support vector clustering (SVC). Random forests are used to reduce spatial complexity, and also to minimize errors. Logistic regression is a method with a statistical model whose basic form uses a logistic function to represent the binary dependent variable. Then, the Naive Bayes function uses binary elements and SVC which has so far given good results rivals other guided learning. Our tests use social media data. Based on the tests that have been carried out on classifier variations and vectorizer variations, it was found that the best classifier is a linear regression algorithm based on predictive adaptive compared to the random forest method based on decision trees, probability-based Bernoulli NB and SVC which work by clustering. Meanwhile, from the test results on the count vectorizer, term frequency-inverse document frequency (TFIDF), and hashing, the best accuracy is achieved on the TFIDF vectorizer. In this case, it means that the TFIDF vectorizer has a better value in presenting word feature dimensions.