Abstract

Social media posts have a wide variety in terms of content and the language use. The automated analysis of the language use has various practical use cases and incurs challenges. In this talk, two sub problems will be elborated on: detection of ironic content and improper posts. The first one can be considered as a complementary task to sentiment analysis, as the irony basically includes a constrast between the sentiment of what is said and what is meant. Therefore it is useful for understanding the lingustic styles of users well as for improving the sentiment detection. The second subproblem, detecting improper posts, can have different forms under different contexts, such as fake news, hate speech or use of adult language. In this talk, we will discuss the problem within the scope of a use case where the social media posts submitted about a program are answered/read live. In such a setting, order to filter out the improper posts, the efficient and correct detection becomes vital.