Advance Search Engine for Legal Texted Ordinances

Abstract

In the framework of legal information retrieval and knowledge representation, this thesis addresses the subject of how to improve the efficiency and accuracy of accessing legal clauses in Pakistan’s Trademark Ordinance 2001. The aim of this research work is to identify the NLP models and ontological structures to improve the legal analysis and decision-making process. This thesis proposes two approaches for improving legal information retrieval and knowledge representation within the scope of Pakistan’s Trademark Ordinance 2001. The first methodology uses transformer models to analyze semantic similarity: FT-Legal-BERT, Legal-BERT, and LegalRoBERTa. By assessing precision scores and employing the Mean Average Precision (MAP) metric, this approach is commonly used in searching tasks to rank relevant sections and effectively separate non-related ones. The proposed similarity score automates the identification of relevant sections based on semantic content, with FT-Legal-BERT outperforming the other transformers and achieving the highest precision score while ranking the top five relevant sections. The MAP scores for the transformers were 0.68681, 0.58974, and 0.47627 for FT-LEGAL-BERT, Legal-BERT, and LegalRoBERTa, respectively. However, when our framework is used to rank only the top three relevant sections, the MAP scores increase significantly to 0.8585, 0.6827, and 0.5790, respectively. The second methodology focuses on the development of an ontology for the Pakistan Trademark Ordinance 2001. The ontology is built using a UML class diagram and defined in OWL language, offering a structured representation of legal provisions, concepts, and relationships. This ontology is a helpful resource for legal researchers, practitioners, and policymakers in Pakistan’s trademark law sector, enabling decision-making and analysis. Furthermore, the thesis delves into the optimization of Legal-BERT’s performance by fine-tuning it to make it more efficient while identifying hidden patterns in legal content. While the methodologies show promise, the difficulty of developing an assessment dataset needs extensive quantitative analyses and comparisons with alternative methods. Future work will include broadening the scope of the ontology, fixing limitations, and guaranteeing frequent updates to reflect changes in the Pakistan Trademark Ordinance.