OFFENSIVE LANGUAGE DETECTION FROM SOCIAL MEDIA TEXT USING MACHINE LEARNING CLASSIFICATION METHODS

Abstract

The utilization of social networking platforms has significantly surged in recent years, leading to a substantial rise in user-generated content across the web. This information predominantly appears in unorganized and somewhat organized forms. Numerous social media platforms face the issue of hate speech, which takes on different forms including aggressive language and the development of visual content like memes. This research focuses on employing Twitter data to identify offensive speech online. Nowadays, techniques for ML which is machine learning and NLP which is natural language processing) have been increasingly utilized for detecting hateful content on the internet. This study specifically addresses the issue of offensive speech detection in textual data by applying _machine learning techniques. Prior to utilizing the dataset with machine learning models, feature selection was conducted. Various machine learning algorithms were applied to an openly accessible Twitter dataset. Offensive speech can be defined as, use of such text or words which are aggressive, violent, or abusive in nature and directed towards a certain group or individual who shares a gender, ethnicity, set of beliefs, or place of residence. The suggested model can automatically identify hateful content on Twitter. This method relies on the TF IDF where TF is known as term frequency and IDF is known as inverse document frequency methodology and a bag of words. Machine learning classifiers are trained using these features. Thorough tests are carried out on the available Twitter dataset, and by comparing 5 different models based on their performance we can conclude that Random Forest Classifier algorithms works best with highest accuracy of 95.22%.