The objective of our work is to detect hate speech in the Indonesian language. As far as we know, the research on this subject is still very rare. The only research we found has created a dataset for hate speech against religion, but the quality of this dataset is inadequate. Our research aimed to create a new dataset that covers hate speech in general, including hatred for religion, race, ethnicity, and gender. In addition, we also conducted a preliminary study using machine learning approach. Machine learning so far is the most frequently used approach in classifying text. We compared the performance of several features and machine learning algorithms for hate speech detection. Features that extracted were word n-gram with n=l and n=2, character n-gram with n=3 and n=4, and negative sentiment. The classification was performed using Naïve Bayes, Support Vector Machine, Bayesian Logistic Regression, and Random Forest Decision Tree. An F-measure of 93.5% was achieved when using word n-gram feature with Random Forest Decision Tree algorithm. Results also show that word n-gram feature outperformed character n-gram.
The objective of this research is to investigate the benefit of utilizing hashtags to determine sentiment polarity of tweets in the political domain. We used the sentiment polarity of hashtags as the features in classification, proposed rules for automatically annotating dataset based on the number of positive and negative hashtags in the tweets, and proposed a method to enrich terms in the tweet by extracting hashtag terms. We named the number of positive and negative hashtags as SentiHT feature. The experiments and evaluation show that sentiment classification using SentiHT feature and the automatically labeled dataset using SentiHT has a very good accuracy of more than 95%. Moreover, SentiHT outperforms unigram feature when combined with Naïve Bayes, SVM or Logistic Regression algorithms, but the opposite occurs when using Random Forest algorithm. Based on computing time to build the model, we recommend using SentiHT feature combined with Naïve Bayes algorithm.
Instagram is one of social media hype, potentially used to spread hatred. The objective of our study is to conduct hate speech detection on Instagram comments for Indonesian language. We used FastText as the classifier and word representation. The experiment results showed that FastText is better than Random Forest Decision Tree and Logistic Regression. The highest result achieved when FastText is combined with bigram feature with F-measure of 65.7%.
Named Entity Recognition (NER) plays a significant role in Information Extraction (IE). In English, the NER systems have achieved excellent performance, but for the Indonesian language, the systems still need a lot of improvement. To create a reliable NER system using machine learning approach, a massive dataset to train the classifier is a must. Several studies have proposed methods in automatically building dataset for Indonesian NER using Indonesian Wikipedia articles as the source of the dataset and DBpedia as the reference in determining entity types automatically. The objective of our research is to improve the quality of the automatically tagged dataset. We proposed a new method in using DBpedia as the referenced named entities. We have created some rules in expanding DBpedia entities corpus for category person, place, and organization. The resulting training dataset is trained using Stanford NER tool to build an Indonesian NER classifier. The evaluation shows that our method improves recall significantly but has lower precision compared to the previous research.
1 January 2017 in Journal of Physics: Conference Series
This research intends to develop a model to evaluate the quality of e-services on e-government. The proposed model consists of seven dimensions: web design, reliability, responsiveness, privacy and security, personalization, information, and ease of use. The model is used to measure the quality of the e-registration of BPJS Kesehatan, an Indonesian government health insurance program. The validation and reliability testing show that of the seven dimensions proposed, only four that suitable for the case study. The result shows that the BPJS Kesehatan e-registration service is good in reliability and responsiveness dimensions, while from web design and ease of use dimensions the e-service still needs to be optimized.