We discuss our experiments for COLIEE Task 1, a court case retrieval competition using cases from the Federal Court of Canada. During experiments on the training data we observe that passage level retrieval with rank fusion outperforms document level retrieval. By explicitly adding extracted statute information to the queries and documents we can further improve the results. We submit two passage level runs to the competition, which achieve high recall but low precision.
翻译:我们讨论了我们在COLIEE任务1上的实验,这是一个使用加拿大联邦法院的案件的法院案例检索比赛。在针对训练数据的实验中,我们观察到用排名融合的方式进行段落级别的检索优于整篇文档级别的检索。通过在查询和文档中显式地添加提取的法规信息,我们可以进一步提高结果。我们向比赛提交了两个段落级的运行,取得了高召回率但低准确率的成绩。