metzzo/deephumor

DeepHumor tries to find out whether Deep Neural Networks are capable of understanding human humor by learning Gary Larson's cartoons. This diploma thesis consists of pre-processing an existing collection of more than 2000 cartoons, training multiple convolutional neural networks to understand the visual component and also training multiple recurrent neural network to detect punch lines in the text. A final classifier is used to merge the results. Additionally a detailed analysis regarding different configurations and an evaluation of the performance is included.

JavaScript