A Multitask Multimodal Framework for Sentiment and Emotion-Aided Cyberbullying Detection
Krishanu Maity, Abhishek Kumar, Sriparna Saha
Abstract
Cyberbullying has become more widespread, especially among teens with the growth of the digital sphere and advancement of technology. This article is the first attempt in investigating the role of sentiment and emotion information for identifying cyberbullying in the Indian scenario. From Twitter, a benchmark Hind–English code-mixed corpus called BullySentEmo has been developed as there was no dataset available labeled with bully, sentiment, and emotion. The developed dataset consists of both the modalities, tweet- text and emoji. In India, the majority of communication on different social media platforms is based on Hindi and English, and language switching is a common practice in digital communication. A multitask multimodal framework called MT-MM-Bert+VecMap based on BERT and VecMap embedding schemes with emoji modality, has been developed. Our proposed multitask-multimodal framework outperforms all the single task and unimodal baselines with the highest accuracy values of 82.05(+/- 1.36)%, 77.87(+/- 1.93)%, and 58.05(+/-2.78)% for the cyberbully detection task, sentiment analysis task, and emotion recognition task, respectively.