PReLCaP : Precedence Retrieval from Legal Documents Using Catch Phrases

Precedence retrieval is the process of retrieving similar prior case documents for the given current case document in the legal domain. Referencing the prior cases is important to ensure that an identical situation is treated similarly in all the cases. Concise representation of case documents using catch phrases facilitates the practitioners to avoid spending more time on reading the whole documents for finding the prior cases. The existing approaches for precedent retrieval in the legal domain use either statistical or semantic similarity features to find the prior cases. However, the substruction similarity features that consider the context of the statement helps to correctly identify the prior cases. Further, the existing approaches consider the whole document while extracting the similarity features, which is time-consuming. In this paper, we propose to use a combination of statistical, semantic, and substruction similarity features that are extracted from the catch phrases of the legal documents. The catch phrases from legal documents are extracted by utilizing Sequence-to-Sequence deep neural network with stacked encoder-decoder and Long Short Term Memory (LSTM) as the recurrent unit. The substruction similarity features are obtained using a convolutional neural network. The IRLeD@FIRE-2017 dataset is used for evaluating our approach. The experimental results show that considering catch phrases reduces the retrieval time without reducing the retrieval performance. The k -paired t -test also shows that the improvement in performance of the model by using substruction similarity features that are extracted from the catch phrases is statistically significant when compared with other models. The PReLCaP outperforms state-of-the-art approaches with the MAP score of 0.632 on test data.