Dataset and additional files/softwares required for the paper "LeSICiN: A Heterogeneous Graph-based Approach for Automatic Legal Statute Identification from Indian Legal Documents"

This dump contains all files and softwares required for running the codes for the paper "LeSICiN: A Heterogeneous Graph-based Approach for Automatic Legal Statute Identification from Indian Legal Documents". Specifically, these codes are available at https://github.com/Law-AI/LeSICiN.

LeSICiN is a deep neural network for the task of Legal Statute Identification which also uses graphical properties of the document-statute citation network for training and predictions.

We have three datasets --- train, dev and test. These are all .jsonl files with each instance dict per line; each instance dict contains the unique id, list of sentences and cited labels of the particular instance. Also, there is a fourth file --- secs.jsonl, which stores the text of all the statutes in similar format.

schemas.json list out the metapath schemas for fact and section type nodes, while type_map.json maps the id of each node to its type (Act/Chapter/Topic/Section/Fact). 

label_tree.json and citation_network.json list out the edges for the two parts of the network in the format of a 3-tuple ('source id', 'relationship type', 'target id')

"ils2v.bin" is the pretrained sent2vec vectorizer that can generate a 200-dim vector for each sentence