Neural Jugement Prediction in English

Legal Judgement is the task of predicting the outcome of a court judgment, given the textual description concerning the case. The data contains cases of European Court for Human Rights (ECHR) violations. Our focus is on a Neural approach to predict the violation of human rights based on the case description. The previous approaches do not explore the explicit property of text data gathering information from their neighbourhood and reflecting similar structure as a graph shows in the same feature space. But, the dominant graph neural networks heavy relied on graph links, and while training such graph models, they lead to over-smoothening and suspended animation problems. A fair Graph-BERT approach is proposed to overcome these issues, which does not consider the meta-information of the case description but tries to capture information from some neighbours and eliminate the over-smoothening and suspended animation problems while predicting the verdicts.