CASRank: A ranking algorithm for legal statute retrieval

Unlike the courts in western countries, legal documents of the Indian judiciary are unstructured, verbose, and noisy. In the justice system, statutes are written laws referred to by judges in support of judicial decisions. Retrieving relevant statutes for a given legal problem can be helpful to lawyers as well as the common man. Moreover, the dearth of publicly available annotated datasets of Indian legal documents limits the scope of legal analytics research. In this paper, we propose a ranking algorithm called CASRank to identify relevant statutes for a legal case query. We also develop a new dataset consisting of 858 Central Acts enacted by the Indian Parliament. Each Central Act is annotated with several attributes, like the act title, enactment date, act definition, chapters, sections, schedules, and footnotes. The first part of the experiment determines the best retrieval model suited for CASRank. The second set of experiments aims to identify the extent to which the attributes of the proposed Central Act dataset contribute towards the retrieval effectiveness of statutes. Experimental results show that the proposed approach obtains a MAP score of 0.0776 with a Precision@10 of 0.0420, showing a considerable increase in retrieval efficiency.

Automated summary

Unlike western courts, the legal documents of the Indian judiciary are unstructured, verbose, and noisy. Retrieving relevant statutes is useful for lawyers and the general public. This paper proposes a ranking algorithm called CASRank to identify relevant statutes for legal case queries. A new dataset of 858 Central Acts enacted by the Indian Parliament is also developed and annotated. The experimental results show a significant increase in retrieval efficiency.