A Novel Convolutional Neural Network For Statutes Recommendation

In recent years, statutes recommendation has been a popular research subject of artificial intelligence in legal domain. However, the existing statutes recommendation systems are more oriented to professionals, such as judges and lawyers, and are not suitable for general public who have no legal knowledge and cannot independently extract key points. We use deep learning to solve the ambiguity and variability of general public's linguistic expressions about cases. We propose a novel Convolutional Neural Network ( CNN) architecture to obtain the relations between statutes and cases. Unlike previous works, in order to utilize the semantics of statutes, we also put statute content as model input besides case description. Moreover, different from the Top-k method, the numbers of statutes recommended by our model varies among cases. In addition, all the features of the case statements and statute contents are extracted automatically without any human intervention. So, the approach for training the model can be easily applied in different types of cases and laws. Experiments results on the juridical document corpus of the proposed CNN model surpass those of previous neural network competitors.