Simulation of marketplace customer satisfaction analysis based on machine learning algorithms

Twitter can be a source of public opinion data and sentiment. Such data can be used efficiently for marketing or social studies. In this research addresses this issue by measuring net sentiment based on customer satisfaction through customer's sentiment analysis from Twitter data. Sample model is built and extracted from more than 3.000 raw Twitter messages data from March to April 2016 of top marketplace in Indonesia. We compared several algorithms, and the classification schemes. The sentiments are classified and compared using five different algorithms classification. There are, K-Nearest Neighbor, Logistic Regression, Naïve Bayes, Random Forest, and Support Vector Machine. The five machine learning can be applied to the Indonesian-language sentiment analysis. Preprocessing on the stages of tokenization, parsing, and stop word deletion of word frequency counting. Frequency of the word of the document used weighted by TF-IDF. The Random Forest, Support Vector Machine, and Logistic Regression generate better accuracy and stable compared with the Naïve Bayes and K-Nearest Neighbor. The results showed Support Vector Machine has accuracy 81.82% with 1000 sampling dataset and 85.4% with 2000 sampling dataset. This shows that the more the number of training data will improve the accuracy of the system. The Net Sentiment score for marketplace in Indonesia is 73%. This results also showed that customer satisfaction has average Net Promoter Score (NPS) 3.3%.