Analyzing bandung public mood using Twitter data

Nowadays, the number of internet users in Indonesia is increasing rapidly. This condition leads to possibility to use and analyze the data gathered from the internet users to show the big picture of the specific condition in certain region. Currently, there are some research on analyzing public mood, such as happy, sad, anger, fear, and neutral. The analysis itself consists of mood classification and topic clustering. However, there is still difficulty to find the best combination of algorithm and feature in the mood classification and topic clustering. In this paper, we propose to compare several algorithms, namely preprocessing (url elimination, word normalization, and stop words elimination), attribute selection (CfsSubsetEval and ClassifierSubsetEval), classification algorithm (Naive Bayes Classifier and Support Vector Machine), and clustering algorithm (EM and simple k-means), to find the best results. Experiments have been done by using 949 tweets from Bandung data. The best classification result was 89.04%, achieved by the combination of using normalization in the preprocessing, CfsSubsetEval in the attribute selection and Naïve Bayes algorithm.