Much has been written about humor and even sarcasm automatic recognition on Twitter. Nevertheless，the task of classifying humorous tweets according to the type of humor has not been confronted so far，as far as we know.　　This research is aimed at applying semi-supervised classification algorithms and other NLP algorithms to the challenging task of automatically identifying the type of humor appearing in messages on Twitter.　　The different methods，algorithms，tools and classifiers used are discussed，as well as the specific difficulty encountered due to the very subjective nature of humor and the informal language applied in tweets.　　It is shown that the discussed methods improve the accuracy of classification by up to 5％above the baseline which is ZeroR，the algorithm that classifies all instances to the majority class.