Opinion mining in heterogeneous data sources, automatic aspect extraction and sentiment analysis

Today while social networks, forums, e-commerce web sites and blogs are covering most of our lives, textual data generated by users is exponentially growing. User opinions in user reviews or in other textual data are very important for retailers, manufacturers and providers of these products and services. Therefore, opinion mining and sentiment analysis have emerged as important research areas recently. In mining user reviews, Latent Dirichlet Allocation (LDA) which is the most popular topic modeling algorithm is a significant method that is used in extracting product aspects in aspect based sentiment analysis. However, LDA algorithm is not so effective on short texts like user reviews or social media messages because of lack of co-occurrence patterns and data sparsity problem. Adaptation of LDA algorithm for short texts is an hot research area in literature. In this study, Sentence Segment LDA (SS-LDA), which is a novel method for aspect based sentiment analysis and aspect extraction is proposed. SS-LDA has been proposed for aspect extraction from short texts and product reviews. In this study, user reviews about smartphones are used as the dataset. The user reviews have been collected from www.hepsiburada.com, which is a popular e-commerce site in Turkey. Experimental results prove that SS-LDA is quite competetive in extracting products aspects.