Evaluation of data mining features, features taxonomies and their applications

• Main Authors: Noekhah, Shirin, Salim, Naomie, Zakaria, Nor Hawaniah
• Format: Article
• Language: English
• Published: Sulaimani Polytechnic University 2017
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/80657/1/NaomieSalim2017_EvaluationofDataMiningFeatures.pdf; http://eprints.utm.my/id/eprint/80657/; https://dx.doi.org/10.24017/science.2017.3.3
• Institution: Universiti Teknologi Malaysia

The World Wide Web has brought an enormous improvement in the lives of people, during the last couple of decades. E-commerce is a new area arisen during this evolutionary period and has changed the traditional trading approaches for selling products and services. It uses different techniques to discover a market trend and analyze the competitor’s activities by exploiting reviews’ information. On the other hand, potential customers, also, use the online opinion to make their purchase decision. Opinion mining and sentiment analysis are the most critical and fundamental domains of data mining which can be useful for variety its sub-domains such as opinion summarization, recommendation system and opinion spam detection. Opinion mining and all its sub-branches can be performed efficiently when there is a comprehensive understanding of the most effective features applied in those domains. To achieve the best results, we need to use the most proper set of features for different case studies in order to classification or clustering. To the best of our knowledge, there is no extensive study and taxonomy of variety range of features and their applications in opinion mining. In this paper, we do comprehensive investigation on various types of features exploited in variety sub-branches of opinion mining domain. We present the most frequent features’ sets including structural, linguistic and relation-based features as a complete reference for further opinion mining research. The results proved that using multiple types of features improve the accuracy of opinion mining applications.