Lexicon Addition Effect on Lexicon-Based of Indonesian Sentiment Analysis on Twitter

Opinion numbers in social media such as Twitter are so widespread that it is not possible to read all of it sentiment (positive, negative, or neutral). Sentiment analysis is one method that can be used to overcome these problems. One of sentiment analysis approach is lexicon-based approach which is highly dependent on the completeness and diversity of sentiment lexicons. Therefore, this study conducts lexicon addition to the sentiment lexicon to improve performance. The datas used in this study were tweet data on the West Java 2018 Governor election, 2019 Presidential election, and COVID-19 pandemic. The results of classification are determined by the highest frequency of occurrence of words based on positive and negative sentiment lexicons. The result of lexicon addition thus being compared to previous work which is Lailiyah method and Saputra and Nurhadryani method. The lexicon addition has proven to improve the accuracy of both Lailiyah and Saputra and Nurhadryani methods on all data with an increase respectively: 6.09% and 5.07% on the 2018 West Java Governor election data, 9.16% and 5.9% on the 2019 Presidential election data, 15.74% and 15.48% on the COVID-19 pandemic data. The lexicon addition could improve the weighted f1-measure on both Lailiyah and Saputra and Nurhadryani methods on all data with an increase respectively: 4.85% and 2.09% on 2018 West Java Governor election, 6.89% and 2.26% on 2019 Presidential election, and 12.18% and 5.10% on COVID-19 pandemic.