Semantic legal information retrieval for tax law using AI and ontology-based methods

Abstract

This paper aims to propose a semantic and artificial intelligence (AI)-driven approach to legal information retrieval that enriches documents with contextual metadata and supports intelligent search. To address the complexity of legal information retrieval, a socio-semantic approach was adopted, integrating semantic and AI within a design science research framework. Through a collaborative knowledge engineering process, experts and engineers co-developed an ontology structured as a semantic network for tax law, integrating international standards such as the European legislation identifier and the legal knowledge interchange format. AI techniques were used to extract domain-specific entities and instantiate a knowledge graph with enriched metadata. The resulting semantic artifacts serve as a framework that aligns extracted data with a formal knowledge organization system, implemented via a knowledge-as-a-service architecture. This enables reasoning mechanisms that support advanced, context-aware search and retrieval tailored to the evolving and heterogeneous nature of legal corpora. A custom named entity recognition (NER) model was developed, achieving an F-measure of 53% for specialized legal entities, significantly outperforming standard baseline models in domain-specific tasks. A case study query showed that semantic retrieval returns precise statutory articles and jurisprudence, while keyword search produced excessive, irrelevant results, demonstrating a precision improvement from 0.27 to 0.85 in contextual queries. The approach reduces research costs, supports decision-making and improves access to justice for lawyers, judges, students and policymakers by breaking down information silos. The novelty of this approach in the integrated and socio-semantic methodology it adopts for legal information retrieval, specifically characterized by the following innovative elements: (i) socio-semantic knowledge modelling; (ii) the use of ontology-based knowledge representation alongside natural language processing and NER techniques exemplifies a hybrid AI approach that bridges structured semantic networks with unstructured legal corpora. This dual-layer architecture – ontology + machine learning – enables a richer semantic contextualization of legal texts than traditional systems; (iii) ontology-driven instantiation of a legal knowledge graph; (iv) domain focus on tax law with potential for generalization; and (v) application to an evolving, real-world legal platform (Lexit).