Optimizing Support Vector Machine in classifying sentiments on product brands from Twitter
Jao Hallen L. Banados, Kurt Junshean Espinosa · 2014
This paper involves giving a better solution in optimizing Support Vector Machine in classifying sentiments towards a product brand. Sentiment analysis rose to solve the problem of classifying sentiments and classifying as to positive or negative feedback towards a certain product brands. Using the Support Vector Machine learning algorithm, this study aims to improve the algorithm's accuracy through choice of kernel and proper tuning of SVM hyper-parameters as core factors in contributing to SVM accuracy, having a huge amount of training sets in order to widen the hyper plane of vectors and strong support vectors. The sentiments are gathered using the Twitter API and are pre-processed to filter unnecessary words. To be able to use the given tool, the pre-processed sentiments are converted to SVM format. By the given default parameters of the SVM tool used, with radial basis function as kernel type. The SVM type used is C-SVC, a multi-class classification. A training set is produced and is used as the training model for test sets and as of the initial results. The model produced an accuracy of 63.54% using SVM with the said default parameters and used 3768 tweets for training set.