Construction of spam filter for Twitter social listening and advertising operation

Abstract

This paper proposes a method for constructing a filter that removes various types of noise and spam posts, which are issues in implementing Web marketing measures using Twitter. Advertising media costs using SNS are increasing year by year, and there is a need for more efficient targeting methods. As a prerequisite, securing an information source with low noise is indispensable in conducting an analysis. Twitter, also known as microblogging, is widely used for word-of-mouth analysis and content marketing. On the other hand, unclear sentences, unique terms, and non-sentences are more likely to appear than ordinary blogs, Wikipedia, and Web sites, so there are parts that cannot be handled by ordinary filtering. The authors constructed a filter to classify tweets that would be the noise of analysis based on their characteristics and to detect them by type, with social listening and promotion using Twitter in mind. In addition, we evaluated the accuracy of the spam filter constructed by the experiment and confirmed that tweets unnecessary for analysis were removed with an accuracy of about 90% as a whole. This has reduced the work involved in performing social listening, and has enabled higher quality analysis.