The feature extraction for classifying words on social media with the Naïve Bayes algorithm

Abstract

To classify Naïve Bayes classification (NBC), however, it is necessary to have a previous pre-processing and feature extraction. Generally, pre-processing eliminates unnecessary words while feature extraction processes these words. This paper focuses on feature extraction in which calculations and searches are used by applying word2vec while in frequency using term frequency-Inverse document frequency (TF-IDF). The process of classifying words on Twitter with 1734 tweets which are defined as a document to weight the calculation of frequency with TF-IDF with words that often come out in tweet, the value of TF-IDF decreases and vice versa. Following the achievement of the weight value of the word in the tweet, the classification is carried out using Naïve Bayes with 1734 test data, yielding an accuracy of 88.8% in the Slack word category tweet and while in the tweet category of verb 78.79%. It can be concluded that the data in the form of words available on twitter can be classified and those that refer to slack words and verbs with a fairly good level of accuracy. so that it manifests from the habit of twitter social media user.