A knowledge-centered framework for exploration and retrieval of legal documents

Automated legal knowledge extraction systems are strongly demanded, to support annotation of legal documents as well as knowledge extraction from them, to provide useful and relevant suggestions to legal actors (e.g., judges, lawyers) for managing incoming new cases. In this paper, we propose CRIKE (CRIme Knowledge Extraction), a knowledge-based framework conceived to support legal knowledge extraction from a collection of legal documents, based on a reference legal ontology called LATO (Legal Abstract Term Ontology). We first introduce LATO-KM, the knowledge model of LATO where legal knowledge featuring documents in the collection is properly formalized as conceptual knowledge, in form of legal concepts and relationships, and terminological knowledge, in form of term-sets associated with legal concepts. Then, we present the bootstrapping cycle of CRIKE that aims to progressively enrich the terminological knowledge layer of LATO by extracting new terms from legal documents to be used for enriching the term-set associated with a corresponding legal concept. Finally, to evaluate the results obtained through CRIKE, we discuss experimental results on a real dataset of 180,000 court decisions of the State of Illinois taken from the Caselaw Access Project (CAP). •The paper presents the knowledge-based CRIKE (CRIme Knowledge Extraction) framework.•CRIKE relies on a reference legal ontology called LATO (Legal Abstract Term Ontology).•The LATO ontology is conceived to support modeling of abstract, open-textured legal concepts.•CRIKE allows to progressively improve the automatic recognition of new terminology useful for document exploration and retrieval.•CRIKE aims at limiting he involvement of legal domain experts in manual annotation tasks as much as possible.