Indonesian Police in the Twitterverse: A Sentiment Analysis Perspectives

Sentiment analysis is extracting information in the form of emotional tendencies based on a series of data. One data source that can be used is tweets from users on Twitter social media. The biggest challenge of text-based sentiment analysis is the Text's multiple representations or meanings. In addition, the rules for writing non-standard texts require special handling. This study analyzes public sentiment towards the police by crawling Twitter tweets with the hashtag #percumaLaporPolisi. Process stages in this study include data collection, pre-processing, feature engineering, Naive Bayes classification, and evaluation. In the data collection process, manual labelling was carried out into three classes: positive, neutral, and negative, resulting in 633 data. At the pre-processing stage, the dataset will be normalized by Text, namely changing the words of slang and typos to synonyms of more understandable words. For feature engineering, researchers use TF-IDF for feature extraction algorithms and chi-square for feature selection. This study uses Naive Bayes as a classification algorithm because of its ability to classify documents by paying attention to the number of word occurrences. The study results show that people tend to be negative toward the police. Meanwhile, the naive Bayes classification model developed produces an accuracy of 74.8%.