Revealing Implicit Relations in Russian Legal Texts

Download PDF
Open Webpage

Dmitry Devyatkin, Anna Sofronova, Vasiliy Yadrintsev

Published: 2020, Last Modified: 18 May 2026RCAI 2020EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: In this paper, new methods for detecting implicit links between legal documents are proposed. Those methods are based on approaches for building vector representations of words, such as Word2Vec, FastText, as well as vector representations of texts and sentences: BERT and Doc2Vec. In addition, as part of this study, we propose an approach to create a dataset for the detection of implicit links and provide such a dataset. The dataset contains more than 36K Russian legal documents. The experiments on that dataset show applicability of the proposed methods. Namely, BERT-based fine-tuned models show the best performance; however, they have the highest demand for memory and computational resources.