Aspect detection in sportswear apparel reviews for opinion mining

Abstract

As a result of the growth of social media sites and e-commerce websites, most of these websites provide platforms for people to express their opinion about their products or services. Main purpose of these platforms is to improve customer shopping experience. Moreover, these websites can use customer reviews to improve their products or services. In the sportswear apparel industry, almost all e-commerce websites provide these platforms for customers to leave their feedback. Since manual analysis of huge number of reviews is practically impossible, the automated approach of sentiment analysis/opinion mining has got the attention. Sentiment analysis can be classified into 3 categories such as document-level sentiment analysis, sentence-level sentiment analysis and aspect-level sentiment analysis. Document-level or sentence-level sentiment analysis does not give the complete information as reviews consist with multiple entities and may have different opinions for different entities. This issue has inspired the aspect level opinion mining. There are two core tasks involve with aspect level opinion mining. Those are aspect extraction and aspect sentiment analysis. This research aim at the first task of aspect level opinion mining which is aspect extraction task for sportswear apparel reviews as none of pervious works consider a clothing review dataset. A new data set will be produced with manual annotations by domain experts. This study used different deep learning models and achieved state-of-the-art performance for sportswear apparel reviews. It serves as the baseline for future research.