Customer reviews which include subjective opinions for the product or service in online store have been generated rapidly and their influence on customers have become immense due to the widespread usage of SNS. In addition, a number of studies have focused on opinion mining to analyze the positive or negative opinions and get a better solution for customer support and sales. It is very important to select the key terms which reflected the customers’ sentiment on the reviews for opinion mining.
In this study, we used the subjective movie, book, game, and music reviews from Amazon to extract suitable terms and then constructed the sentiment classification model based on the terms in document level. Also, to improve the performance of sentiment classification model, we proposed ISCOM model that compared to the previous models which mainly used the existing data mining techniques. The experimental results are presented as follows.
First, this paper validated the fact that sensitive information can be represented by using various parts of speech(POS). Also, this study compared the difference between the result of only using adjectives and the result of using all the four kinds of POS(adjective, adverb, verb, noun). The difference can be identified by using sentiment classification model.
Second, the variable selection of most previous studies are inefficient, for the reason that these models are usually extracted extensive terms and some terms may be very repetitive and some may not contain any sensitive information.
To make up deficiency in previous model, we extract the significant terms by using POS tagging for the four kinds of POS separately.
In addition, we utilize various kinds of feature selection approaches such as Information Gain(IG), Document Frequencey(DF), Term Frequency Inverse Document Frequency(TFIDF), Chi-squared Statistic(CHI). The significant terms are extracted by using each technique and then are applied to the sentiment classification model.
Third, to the best of our knowledge, we first apply the novel techniques(SVM+, SVM+MTL) to opinion mining and construct the novel sentiment classification model and also validate its validity.
Another notable feature of our paper is that we applied real data of Amazon to our proposed sentiment classification model and validated the performance.
Finally we employed SVM, SVM+, and SVM+MTL as the benchmark model, and proposed the ISCOM (Integrated Sentiment Classification for Opinion Mining) model. Finally, we proposed an optimal model by comparing the results of these models.
The results showed the out performance of ISCOM model to each separate model.
ISCOM model is worth to apply to solve sentiment classification problems based on customer reviews. Our proposed opinion mining model is expected to improve customer service and gain competitive advantage in online store.