Estimating the Legal Basis from Court Files Using Deep Graph Methods

Legal texts are increasingly becoming a hindrance for lawyers in the judicial system of many countries today. The increasing number of cases every year, new laws, and the intermittent changes in the jurisprudence of the courts make it difficult to follow and classify the files. The fact that precedent cases and references in law provide clues to lawyers for current cases requires careful examination of these files. In this study, deep learning architectures were developed to find which articles of the European Convention on Human Rights the cases are related to using the files of the European Court of Human Rights. A dataset containing Turkish cases was created using the online resources of the European Court of Human Rights. To create a comprehensive case vector, each case was matched with the contract article it referred to, and a bipartite graph was designed. The vectors of the nodes in the graph were obtained using the node2vec algorithm. At the same time, the Doc2Vec algorithm was used to obtain the content vector of each case. Both representation vectors were tested on LSTM, CNN, and their hybrid architectures both together and separately to predict which contract article the cases are related to, and the results were compared and shown.