Lexicon-based Sentiment Analysis for Persian Text

Proceedings of Recent Advances in Natural Language Processing, pages 9–16, Hissar, Bulgaria, Sep 7–9 2015. Lexicon-based Sentiment Analysis for Persian Text Fatemeh Amiri1, Simon Scerri2 and Mohammad H. Khodashahi3 1Media Informatics, RWTH Aachen University, Germany 2Fraunhofer IAIS, Sankt Augustin, NRW, Germany 3Institute of Computer Science, University of Bonn, Germany amirii.fatemeh@gmail.com simon.scerri@iais.fraunhofer.de mh.khodashahi@gmail.com Abstract The vast information related to products and services available online, of both ob- jective and subjective nature, can be used to provide contextualized suggestions and guidance to possible new customers. User feedback and comments left on differ- ent shopping websites, portals and social media have become a valuable resource, and text analysis methods have become an invaluable tool to process this kind of data. A lot of business use-cases have ap- plied sentiment analysis in order to gauge people’s response to a service or prod- uct, or to support customers with reach- ing a decision when choosing such a prod- uct. Although methods and techniques in this area abound, the majority only ad- dress a handful of natural languages at best. In this paper, we describe a lexicon- based sentiment analysis method designed around the Persian language. An eval- uation of the developed GATE pipeline shows an encouraging overall accuracy of up to 69%. 1 Introduction In comparison to other more popular and widespread language, few research efforts have sought to provide text analytics services targeting Persian text documents on the Web. As the offi- cial language of Iran, Afghanistan, and Tajikistan and an estimated 110 million people, we feel that the Persian language has not been given the at- tention it deserves. Besides attaining merit from a purely linguistic point of view, providing tech- nologies for Persian text analysis has also busi- ness implications in the regions where the lan- guage remains a preferred working language. In particular, sentiment analysis has a high poten- tial in providing insights for several Persian on- line communities and social media. Most of the limited available techniques have employed Ma- chine Learning (ML) algorithms, such as Sup- port Vector Machine-based (SVM) methods. In contrast, our approach is based on a manually- created lexicon enriched with sentiment scores; coupled with hand-coded grammar rules. In tack- ling our objective, we are faced with language- specific challenges and constraints. In the Persian language there is typically a large difference be- tween formal and informal writing styles. There is also a high level of complexity due to the fre- quent morphological operations. Besides a com- plex morphology, Persian has some other distinc- tive features, such as lexicon intricacy, a high con- text sensitivity of the script, and a free words or- der due to independent case-marking (Hajmoham- madi and Ibrahim, 2013). Therefore models used in approaches behind other languages, or even as- pects of which, can hardly be used in Persian text analytics methods. In this paper, we describe how we approached the language-specific challenges when designing and implementing a lexicon-based sentiment anal- ysis method for Persian text. An evaluation of this method is …

Content truncated to 500 words