A segregational approach for determining aspect sentiments in social media analysis

摘要

Aspect-based sentiment analysis is an emerging field of research that evaluates people's views, ideas or sentiments. It is a subtask of sentiment analysis that is used to identify text sentiment orientation towards different aspects of a mobile phone such as camera and screen resolution. During the last decade, research community focused on identifying and extracting aspects like the most common methods used for aspect extraction to identify the main features of an entity only. These techniques are corpus or lexicon based and domain specific. Some approaches for aspect extraction are based on term frequency and inverse document frequency. Such approaches are quite good if aspects are associated with predefined categories and may fail if low-frequency aspects are concerned. The heuristic-based approaches are better than frequency and lexicon-based approaches in terms of accuracy, but due to the different combinations of features, they consumed time. The researchers have already implemented machine learning techniques to analyse sentiments present in the given document. But, execution time for these techniques increases due to the increasing aspects in a set of data. Also, irrelevant and redundant aspects participate in determining the sentiment of the given document, thereby varying the accuracy of the algorithm. In this research, we present a segregational approach for aspect identification that is based on aspect and opinion words disentangling and aspect refinement using concept similarity. To obtain better accuracy, we also built a set of part of speech tagger and integrated it with our proposed technique. The experimental analysis reveals that our proposed technique outperforms the existing counterparts.