The improved neural network model in humor detection with traditional humor theory

Judging the degree of humor in natural language texts is a very important issue in natural language processing. This paper takes edited news headlines as the research object and studies how to use a neural network to predict the degree of humor. Our main purpose is to explore the method of combining neural networks and traditional humor theory, to be specific, is to automatically evaluate the humor degree of news headlines, which includes original news headlines and modified news headlines. This paper designs a prediction model of humor degree based on RoBERTa, which contains three information extractors. And the classic models in deep learning, LSTM, and CNN and the recent popular models, Bert and RoBERTa, were selected as comparative models for experiments. The results prove that our proposed method can effectively improve the prediction effect of the neural network model.