Amrita_CEN at SemEval-2022 Task 4: Oversampling-based Machine Learning Approach for Detecting Patronizing and Condescending Language

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 515 - 518 July 14-15, 2022 ©2022 Association for Computational Linguistics Amrita_CEN at SemEval-2022 Task 4: Oversampling-based Machine Learning Approach for Detecting Patronizing and Condescending Language Bichu George, S Adarsh, Nishitkumar Prajapati, Premjith B, and Soman K.P Centre for Computation Engineering and Networking (CEN) Amrita School of Engineering, Coimbatore Amrita Vishwa Vidyapeetham, India b_premjith@cb.amrita.edu Abstract This paper narrates the work of the team Am- rita_CEN for the shared task on Patronizing and Condescending Language Detection at Se- mEval 2022. We implemented machine learn- ing algorithms such as Support Vector Machine (SVM), Logistic regression, Naive Bayes, XG Boost and Random Forest for modelling the tasks. At the same time, we also applied a fea- ture engineering method to solve the class im- balance problem with respect to training data. Among all the models, the logistic regression model outperformed all other models and we have submitted results based upon the same. 1 Introduction Discriminatory language on the social media is lately creating hostile environment towards the vul- nerable communities especially women and minori- ties. These are reflected in day to day conversations happening on popular social media sites. It is a high time now to build a technological solution to counter the discrimination against vulnerable com- munities. Here in this task, we consider one such issue known as "Patronizing and Condescending Language (PCL) Detection". When someone’s lan- guage conveys a pompous attitude toward others or portrays them or their circumstances in a compas- sionate manner, eliciting feelings of sympathy and compassion, they are patronising or condescending. This is why it is important to develop a computa- tional model to predict whether there is patronizing content in social media or not (Pérez-Almendros et al., 2020). This challenge can be solved by the applying Natural Language Processing (NLP) con- cepts. The Social media platforms reaches a huge audience, which might contribute to increased ex- clusion and inequity among vulnerable groups. De- spite the fact that harmful language behaviour (such as hate speech, abusive language, fake news, ru- mour propagation, or disinformation) (Sreelakshmi et al., 2020), (Sreelakshmi et al., 2021) has been extensively investigated in NLP, PCL has remained a neglected field of research. We implemented seven machine learning mod- els which include three classical machine learning algorithms and four ensemble models: Support Vector Machine (SVM), Logistic regression, Naive Bayes, XG Boost and Random Forest for mod- elling the tasks (Soman et al., 2009), (Premjith et al., 2019), (Premjith and Kp, 2020). The class imbalance problem was dealt by a minority over- sampling technique called SMOTE and compara- tive analysis of our algorithm was done by various evaluation metrics such as precision, recall and F1 score. The remaining parts of the paper are described as follows: Section 2 contains dataset description along with works related to that. Section 3 de- scribes the system overview. Section 4 explains the experimental setup. Section 5 discusses result and the paper is concluded in Section 6. 2 Related …

Content truncated to 500 words