Abstract: A new method for sentence sentimental analysis based on Markov logic network is proposed. With the combination of Markov logic network and deep learning methods, it could realize the cross-domain knowledge migration. By the function of Markov logic network that could combine discourse information with other sentiment features of sentence, the proposed method could also realize the sentence sentiment orientated analysis. Experimental results on COAE data show that, compared with SVM method, this method could improve the precision considerably and achieve the high precision for implementing cross-domain sentimental analysis task.