FVEC-SVM for opinion mining on Indonesian comments of youtube video

Support Vector Machine (SVM) has long been used in opinion mining social media website including YouTube, the most popular video sharing based media social in the world. However, the preprocessing approach and use of kernel functions in SVM requires precision in the selection of appropriate kernel functions in order to get high accuracy. Thus, this research focuses on proposing FVEC approach for preprocessing and finding the best kernel function in term of accuracy, for opinion mining on Indonesian comments of YouTube video. Four types of kernel functions have been investigated, namely linear, poly degree 2, poly degree 3, and RBF. The experiment uses 13,638 Indonesian comments of YouTube videos that review about smartphone products of various brands. The comments can contain sentiments that refer to how the video is delivered or the product itself, or even irrelevant to both, so this study classifies comments into seven classes. From the experimental result show that FVEC-SVM using linear kernel function is outperformed than others on accuracy term, i.e. 62.76%.