Knowledge based intelligent legal system for information retrieval

Abstract

This research applies methods from machine learning and natural language processing (NLP) newlineto demonstrate intelligent information retrieval for the legal domain. It proposes several newlineschemes for legal document similarity estimation using the automatic extraction of prominent newlinelegal concepts within a legal document. Thus, each of the proposed methods for estimation newlineof legal document similarity also proposes a method for the extraction of prominent legal newlineconcepts from a legal document. newlineSimilarity estimation using legal text segmentation proposed in this research exploits legal newlinedomain knowledge to perform the task of segmentation to identify logical subdivisions within newlinea legal document (court judgment). In addition to a single value of document similarity, this newlineapproach also has the advantage of providing the segment-wise similarity values for a deeper newlineunderstanding of relevance. Legal citation-based similarity estimation demonstrated in this newlineresearch proposes the construction of citation co-occurrence networks to capture the newlinerelevance among citations and in turn the court judgments (cases). Legal citation networks newlineare extremely huge, complex and sparse. This research addresses these challenges by newlineleveraging on the recent advances in the field of neural embeddings, node2vec, for the vector newlinerepresentation of nodes. newline newlineThough document summarization is researched extensively for wide-ranging applications in newline newlineinformation retrieval, it is relatively less explored for legal documents due to the non- newlineavailability of sufficient labeled training data. This research proposes an entirely data-driven newline newlinescheme for automatic labeling data for performing legal text summarization to mitigate the newlineproblem of training data availability. The approach further demonstrates legal document newlinesummarization using deep learning techniques by utilizing the labeled data as generated and newlinedocument similarity estimation using generated summaries subsequently. newlineTo provide a multi-dimensional abstraction of the similarity among legal documents, this newlineresearch presents a novel approach of presenting similarity estimation as multi-criteria newlinedecision making problem. Frequently interacting concepts within a document are extracted newlineusing a graph based method intrinsically without relying on external inputs. The document newlinesimilarity is then estimated based on the extracted concepts using the ordered weighted newlineaverage (OWA) operators. This approach also facilitates intuitive visualization that aids in newlinethe interpretation of the similarity between a pair of documents. newlineThe effectiveness of the proposed methods is demonstrated through experiments performed newlineon real word datasets. newline