An Analysis of Opinion Mining Research Works Based on Language, Writing Style and Feature Selection Parameters

An Analysis of Opinion Mining Research Works Based on Language, Writing Style and Feature Selection Parameters Jasleen Kaur Assistant Professor Research Scholar Shroff S. R. Rotary Institute of Chemical Technology Uka Tarsadia University, Ankleshwar, Gujarat, India Bardoli, Gujarat, India [email protected] Dr.Jatinderkumar R. Saini Associate Professor Research Supervisor Narmada College of Computer Application Uka Tarsadia University, 1 Bharuch, Gujarat, India Bardoli, Gujarat, India [email protected] --------------------------------------------------------------ABSTRACT--------------------------------------------------------------------Different writing styles, either formal or informal, can be adopted to present the written text. A Piece of text may contain a lot of emotional states, feelings or ideas presented through the means of words and means of Language. Various techniques and methods are present in the field of Opinion Mining and Sentiment Analysis to extract the emotions from text. This paper presents analysis of Formal and Informal text pieces written in different International Languages in the field of Opinion Mining and Sentiment Analysis. This paper presents a study and analysis of differences of approaches used for Opinion Mining and Sentiment Analysis for both cases. Formal and Informal Text Pieces are present in 8 different International Languages (English, Chinese, Arabic, Malaysian, Spanish, Turkish, Persian , Korean).In this study Formal text pieces ,in form of poetry, proverbs ,essay and documents, and Informal text in form of micro blogs, chats, emails and SMS, are analyzed. Maximum performance for Opinion Mining in case of Informal Text is achieved in Arabic Language and for formal text; maximum accuracy is obtained in Persian and Turkish Language. In this study 4 different Feature selection parameters (IG, TF-IDF, n-gram, MI and MMI) were analyzed in order to find emotional states associated with written text. It was found that parameter, IG and TF-IDF, were experimented by Researchers maximum number of times and IG outperformed all other feature selectors.