A New Deep Transfer Learning Model for Judicial Data Classification

For judicial analysis, usually there are enough labeled instances for one domain, but there are few or even no labeled instances in target domain. Therefore, to bridge the gap between these two domains and use the sufficient source information for target analysis is important. In this paper, we focus on developing a new deep transfer learning model to translate the source domain information for target data classification. The proposed model integrates the deep neural network (DNN) with canonical correlation analysis (CCA) to derive a deep correlation subspace for associating data across different domains. Moreover, a new objective is designed to train the whole network jointly. When the deep semantic representation is achieved, the shared features of the source domain are transferred for instance classification in the target domain. Experiments on several datasets present that the proposed method is superior to the state-of-the-art methods for deep transfer learning, which is promising for judicial data classification.