Category Based Sentiment Analysis for Basic Skin-Cares Using LDA and SVM Approach

Abstract

Aspect Based Sentiment Analysis (ABSA) is a sentiment analysis technique of identifying sentiment value towards a specific aspect associated with a given target. However, in some cases, several aspects can be classified into one category group. This is because some words have different spellings, even though they refer to the same aspect. The problem is the dataset for category-based categories is still limited, especially for low-resource languages such as Bahasa Indonesia (Indonesian Language). In fact, dataset generation is the most time-consuming step rather than the process of building the classification model. This study aims to build a category-based sentiment analysis model starting from the dataset generation, model classification generation, and model evaluation. The case study in this research is sentiment analysis for the basic skin-care category in Bahasa Indonesia. The dataset is generated from raw data, which is the user’s review on the female-daily website (www.female-daily.com). We implemented the Latent Dirichlet Allocation (LDA) algorithm to extract the aspect from each review and implemented TextBlob library to annotate the polarity of each review. The research uses a Support Vector Machine algorithm to generate a model classification. Based on the research, the LDA can extract and also can group the same aspect into the same category. The support Vector Machine method is a fairly accurate method for conducting sentiment analysis. In this study, the accuracy value of 86.9% for training data and 83.9% for test data from all categories were obtained using the K-Fold Cross Validation evaluation method with 5 folds.