KMI-Coling at SemEval-2019 Task 6: Exploring N-grams for Offensive Language detection

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 668–671 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 668 KMI−Coling at SemEval-2019 Task 6: Exploring N-grams for Offensive Language detection Priya Rani Dr. Bhimrao Ambedkar University Agra, India pranijnu@gmail.com Atul Kr. Ojha Jawaharlal Nehru University New Delhi, India shashwatup9k@gmail.com Abstract In this paper, we present the system descrip- tion of offensive language detection tool which is developed by the KMI−Coling Group under the OffensEval Shared task. The OffensEval Shared Task was conducted in SemEval 2019 workshop. To develop the system, we have explored n-grams up to 8-gram and trained three different systems namely A, B and C system for three different sub tasks within the OffensEval task which achieves the accuracy of 79.76%, 87.91% and 44.37% respectively. The task was completed using the data set pro- vided to us by OffensEval organisers, which was the part of OLID data set. It consists of 13,240 tweets extracted from twitter and were annotated at three levels using crowd sourcing. 1 Introduction The very first question which arises in one’s mind when one starts working in the area of computa- tional sociolinguistics research related to the lan- guage usage in social media and networking sites is what is offensive language and its related terms such as hate speech, aggression and all? The sec- ond question arises is related to the definition of this terminology. We would suggest that offen- sive language is still not a very well-defined phe- nomenon. As we know that the natural language is productive in nature. These aspect of the lan- guage use has always existed as part of the speech repertoire of the speaker. In the area of scientific study, we need to move forward with a definition. Therefore we will go by the definition given by Jay and Janschwetiz which states that Offensive lan- guage is vulgar, pornography and hateful language (Chen et al., 2012). But even this definition does not incorporate the many more structure which is neither vulgar nor pornography nor hateful but are definitely offensive. Such type of structure is what leads to challenges in the detection of offensive language in the discourse. With the increase in the culture of social media and social networking sites, the use of offensive language has increased rapidly. Moreover, it has also given a very good platform to conduct different research in the given area. 2 Literature review This section gives a brief outline of the existing literature and approaches that are available for of- fensive language detection. Lots of research works are being done to detect offensive language and there has been significant progress over time. Lex- ical Syntactic based framework was used for sen- tence offensive detection and user offensive detec- tion by Chen et al. (2012). Another study by Xiang et al. (2012) which uses keyword matching tech- nique that performed very well in literature do- main. Razavi et al. (2010) uses auxiliary weighted repository by matching the text to …

Content truncated to 500 words