Exploring the Knowledge of Citations in Legal Information Retrieval

A legal judgment is a document which provides an explanation of the court order concerning a legal case. The legal data is being digitized and is being made available in digital format for access by legal practitioners and the general public. Also, the legal data is continuously growing with the judgments being delivered. In common law based legal systems, legal practitioners refer to the previous judgments of the court to prepare the new judgment. Extracting relevant legal judgments for a given input judgment is useful to legal practitioners. The inherent complexity of a legal judgment requires effective and efficient ways to extract relevant legal judgments for a given input judgment. Legal judgments contain text and citation information which can be exploited for finding relevant legal judgments for a given input judgment. In the literature, efforts have been made to exploit the text and citation information for finding similarity between the pairs of judgments. In this thesis, we show the potential of citation information in the judgments, in finding similar judgments through cluster analysis and for ranked retrieval of relevant legal judgments. Firstly, we analyzed how the citation information could be helpful in finding similar legal judgments. Link information has been exploited extensively in the web domain for search and retrieval systems. Similarly, citation information in the legal judgments can be exploited for building efficient search systems in the legal domain. We developed a clustering approach to cluster judgments having only citation information. We have conducted experiments on real world legal judgment dataset delivered by Supreme Court of India. The clustering results indicate that the citation information is helpful in finding clusters of judgments. As the naturally available citations are sparse, we also exploit the notion of paragraph links to increase the number of citations in the judgments and reported the corresponding clustering results. Further, we have made an effort to show the potential of citation information in the ranked retrieval of relevant legal judgments for a given judgment. Efforts are being made to build efficient search systems that extract legal judgments for a given input query by extending the traditional information retrieval approaches. Intricate legal concepts are discussed at a granular level in the legal judgment documents. In addition to citation information, we exploit the paragraph-level information of the judgments for ranked retrieval of relevant legal judgments for a given input judgment. Instead of considering the entire text document for comparison, we find the aggregate relevance between judgments by computing the relevance at paragraphs of the judgments based on the Okapi retrieval model. User evaluation study on