Track: Data Analytics and Big Data

Abstract

Sentiment analysis is one of the areas of application of text analysis and NLP (Natural Language Processing). It builds a system to identify and analyze personal information from textual sources, which are generally divided into positive and negative categories. Many case studies are chosen to measure the accuracy of the computational model of sentiment analysis. In this study, we use social media's perception of electronic wallets usage. The Naïve Bayes method was chosen. It can help classification because it is assumed to be an independent variable, and Lexicon is used to calculate the weight of each word. Although several studies have been conducted on sentiment analysis, none have used non-standard word correction. The use of non-standard word correction deals with non-standard words such as slang and word abbreviations. This research begins by digging up the required data, namely the keyword electronic wallet. The standard word dictionary normalizes non-standard words, preprocessing data with four stages: case folding, tokenizing, stopword, and stemming. The lexicon dictionary produces positive and negative labels, and nave Bayes is used to classifying. The data used in this study were 3878 tweets, with a distribution of 70% training data and 30% test data. Sentiment analysis obtained in this study shows that Twitter users in Indonesia are more likely to give negative comments to electronic wallets. The results of this study indicate that by using both methods and adding standard word corrections and testing using RapidMiner, the accuracy rate for classifying positive and negative sentiments reaches 88.56%. Further research can add Levenshtein Distance normalization to the classification results to better influence the accuracy value.