New Filtering Scheme Based on Term Weighting to Improve Object Based Opinion Mining on Tourism Product Reviews

Abstract

Reviews are an essential thing in tourism industry. Opinion mining used for processing a massive amount of review data, so it can be more useful for the industry. The utilization of filtering can improve the feature extraction result from object based on opinion mining and can improve opinion classification result generally. However, there is no proven method yet to develop filter data automatically. This work applies several term weighting methods such as TF-IDF mTFIDF and BM25 to develop filter data automatically. The result from this research is the best term weighting method for developing filter data, that can improve the feature extraction and opinion mining relatively. TFIDF become the best term weighting method applied for filter data combined with the most frequent objects, The accuracy is 37.98%, the precision is 50.69%, the recall is 44,28%, and F-measure 47.27% for hotel data. Meanwhile, for restaurant data, the accuracy is 37.98%, precision is 50.69%, recall is 44.28%, and F-measure 47.27%.