Improving Web 2.0 Opinion Mining Systems Using Text Normalisation Techniques

Proceedings of Recent Advances in Natural Language Processing, pages 491–495, Hissar, Bulgaria, 7-13 September 2013. Improving Web 2.0 Opinion Mining Systems Using Text Normalisation Techniques Alejandro Mosquera University of Alicante amosquera@dlsi.ua.es Paloma Moreda University of Alicante moreda@dlsi.ua.es Abstract A basic task in opinion mining deals with determining the overall polarity orienta- tion of a document about some topic. This has several applications such as detecting consumer opinions in on-line product re- views or increasing the effectiveness of so- cial media marketing campaigns. How- ever, the informal features of Web 2.0 texts can affect the performance of auto- mated opinion mining tools. These are usually short and noisy texts with pres- ence of slang, emoticons and lexical vari- ants which make more difficult to extract contextual and semantic information. In this paper we demonstrate that the use of lexical normalisation techniques can be used to enhance polarity detection results by replacing informal lexical variants with their canonical version. We have carried out several polarity classification experi- ments using English texts from different Web 2.0 genres and we have obtained the best result with microblogs where normal- isation contribution to the classification model can be up to 6.4%. 1 Introduction Nowadays, Web 2.0 applications provide some of the most popular forms of communication be- tween users on the Internet such as blogging, so- cial networks or short text messaging platforms. This large daily amount of generated information contains valuable insights about user opinions and sentiments regarding almost any topic. A basic task in opinion mining deals with deter- mining the overall polarity orientation of a docu- ment about some topic. The polarity information extracted from user comments and consumer feed- back from on-line product reviews can be used to increase the effectiveness of social media market- ing campaigns, discover new market threats and opportunities or react faster to customer issues. Also, microblogging platforms such as Twitter in- clude rich metadata about interactions which pro- vides a way to measure the reputation of their users based on the number of followers or the publication popularity by counting the number of times one message has been shared. However, the language used in social me- dia is very informal, containing elements such as misspellings, slang, lexical variants, inconsis- tent punctuations, URLs or emoticons (Thurlow, 2003). Also, the presence of genre-specific ter- minology such as, RT for re-tweet and #hashtags can make any Natural Language Processing (NLP) task challenging. For this reason, a way to handle such challenges is needed in order to automatically understand the opinions and sentiments that peo- ple are communicating on the Internet. The use of lexical normalisation techniques has recently been the subject of research applied to short and noisy texts such as tweets or SMS, improving the perfor- mance of NLP tools that need to extract contextual and semantic information from this type of infor- mal texts. Moreover, not all Web 2.0 genres have the same level of informality, microblog posts have to be short so they tend to contain …

Content truncated to 500 words