Cross-Domain Aspect Detection and Categorization using Machine Learning for Aspect-based Opinion Mining

ABSTRACT: There is an increase in the development of social media and electronic commerce sites day by day. In order to express their opinions about the products purchased user's write comments, messages and reviews. The reviews present in the e-commerce sites are also increasing. Users find difficulty in getting appropriate information about the right topic from this large data. Aspect-based Opinion Mining (ABOM) helps users in this regard. In many real-world applications ABOM is used to get the details about the aspects of entities, where the opinion is expressed for those aspects and entities. One of the key elements of ABOM is Aspect extraction. Unsupervised Machine Learning approach has been used to extract aspects from the reviews as it does not require pre-labelled data. In this regard Latent Dirichlet Allocation (LDA) and Latent Semantic Analysis (LSA) are two most commonly used unsupervised Topic Modeling approaches. The topics are extracted from three different datasets such as Amazon Mobile Reviews, Hotel Reviews and IMDb Movie Reviews using LDA and LSA algorithms. These extracted topics are aspects of our interest. The results of topic modeling algorithms are quite difficult to be interpreted by the common user. The different visualization methods are used to display the results of topic modeling algorithms in an interactive way. Two different multi-class classifiers such as Multinomial Naive Bayes (MNB) and Support Vector Machine (SVM) have been constructed for aspect categorization. These classifiers are evaluated by considering the evaluation measures such as Precision, Recall and F1 score. As a result, SVM classifier has good performance than MNB classifier for aspect categorization task of aspect-based opinion mining