Comparison of Deep Learning Technologies in Legal Document Classification

Predictive Coding in legal document review, also called Text Categorization in machine learning, has been widely used in the legal industry. By leveraging machine learning technologies such as logistic regression and support vector machines (SVM), each document is assigned a probability score of its relevance to the legal case and the probabilities of documents are used to prioritize the documents to be reviewed so to improve review efficiency and cost. In recent years, deep learning technologies have been successfully applied in many text classification tasks. In predictive coding, studies were also shown better performance in some applications. Several different deep learning technologies have been used in text classifications, but there are few studies in comparisons of these technologies, especially in predictive coding. These deep learning technologies include Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), as well as RNNs with attention mechanism. This paper reports our preliminary results in comparison of different deep learning technologies in predictive coding. Specifically, the authors conducted experiments using these technologies in three open source legal document review datasets and the experimental results show that CNNs perform better than other models.