TWITTER, SENTIMENT ANALYSIS AND PRODUCT DEVELOPMENT: HOW MUCH USERS ARE EXPRESSING THEIR OPINIONS?
Leandro Matioli Santos, Ahmed Ali Abdalla Esmin, André Luiz Zambalde, Frank Mendes Nobre · International Conference on Information Systems, Technology and Management · 2010
This paper aims to perform an investigation using mathematical procedures about how the micro-blogging service and social network known as Twitter can work as a way to make direct contact with products' clients. Using a machine learning method called SVM (Support Vector Machine), it intends to estimate how much Twitter users are expressing their opinions and sentiments about a certain product on it and perceive if Twitter really serves to do such research on people's receptivity about one specific product. In this work, the new operating system launched by Microsoft(R) called Windows 7(R) was chosen as a study case. A message gathering from Twitter containing the word in several languages was done and after that, a filter was build to obtain only messages in English containing, specifically, the word windows 7. The SVM was trained by a set of messages randomly selected from the document collection and they were manually classified as being neutral or containing opinions. Test with a set built in the same manner was also performed, resulting in an accuracy of 80% in the messages automatically classified. Finally, remaining messages were submitted to the classifier already trained and separated in two distinct classes (those containing opinions or sentiments about Windows 7(R) and those containing neutral messages about the operating system). It was concluded that most part of what is posted on Twitter doesn't express the user's opinion, being necessary a considerable amount of messages to perform a poll only with those containing sentiments related to the product in case.