Online Text-Based Humor Detection

Artificial intelligence is replacing humans and their employment in different fields in today's technological environment. Researchers are trying to create virtual assistants and robots to mimic human characters as much as possible. Out of many impressive human characters, a sense of humor is a charming one. A virtual assistant or a robot with a great sense of humor will be a better replacement for an actual human. Moreover, natural language processing plays a vital role to capture the sense of humor from online texts. In this paper, we detect humor text from online media with help of a generalized autoregressive model. In specific, we fine-tuned the XLNet base to outperform other models in the same humor detection task with a 200k formal texts dataset. The proposed model applies context dependent features to capture the sense of humor. Our result analysis shows that our proposed work achieved an accuracy of 98.6% which is higher than existing models.