A pragmatic approach for hate speech detection through applying machine learning

• Main Author: Devi, G. Malini
• Format: Conference Proceeding
• Language: English
• Subjects: Classification; Hate speech; Machine learning; Sentences; Social networks; Verbal communication; Websites; Words (language)
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0130174

Present trend of colossal increase in social exchanges seen in web social networks, an ensuing boost in vile activities are exploiting this infrastructure. Almost all websites proffer platform for people to chatting, sharing opinions and views, voicing their moral fiber in form of post(s), comment(s) and message(s) making them nearly not doable in controlling the content. In addition, given the diverse ethnicities, cultures and belief systems, subsequent masses incline towards belligerent and odious verbal communication when in discussion with people with contradictory views. In a conversation, sentences are interpreted into classification of: “strongly hateful (SH)”, “weakly hateful (WH)”, and “non-hateful (NH)”. An approach based on pattern detection is proposed in which pre-defined patterns are mined in realistic way from a training set and then optimized set of parameters are defined to collect the similar patterns. Beside the mentioned patterns another proposal is made on collection of words and expressions which imitate an offense tone and reveal hate and further using these words with patterns alongside features based on sentiment to detect a hate speech. Classification of sentences into three classes makes the distinction which reflects hate or just being offensive. Later taking a test set, a classification is run based on grammatical patterns and semantic features which revealed the score of 0.68, 0.88, 0.80, 0.76, 0.62, 0.82 for SVM and 0.99, 0.98, 0.97, 0.97, 0.99, 0.99 for random forest. For Random forest it out performed