Prior Case Retrieval for the Court of Cassation of Turkey

This work introduces the first prior case retrieval models for Turkish courts. We investigated the rulings of the Court of Cassation of Turkey. Since law professionals have to find texts they are interested in from enormous legal databases and computers can process large amounts of text swiftly, information retrieval algorithms are helpful for law professionals. The information retrieval algorithms utilized in this work are recurrent neural network autoencoders, and the combinations of recurrent neural network autoencoders with BM25 algorithms. The combination of the long-short term memory autoencoder with ATIRE BM25 achieves the best scores on our dataset.