Refworld is a UNHCR repository that contains much of the legal documentation on humanitarian affairs and refugee law, and many use it for easier access to legal precedents and court decisions in advancing their legal research. The UNHCR Division of International Protection was awarded a grant from the UNHCR Innovation Fund in 2019 to optimize Refworld navigation for the end-users by using AI to automatically extract metadata, citations and tags, related to references in case law to other legal and policy documents.
Asylum decision makers conduct legal research to draft for example a legal decision, a brief, to argue why someone should (or should not) be determined to be a refugee. For this, they need to be able to develop their reasoning by using and referring to other sources, such as case law and UNHCR policy and guidance. Refworld provides a lot of this information but the decisions are in general not linked with each other and its current design does not necessarily allow for a user-friendly manner in which the documents can be linked.
Whether the ease and speed of navigation between the different legal and (UNHCR) policy documents available on Refworld can be optimized for the end-users by the use of artificial intelligence in order to extract metadata in an automated manner, in particular with regard to references in case law to other legal and policy documents?
Data innovation solution
Performed market scoping to see if there were already solutions available that could be customized for our purposes. This resulted in two (hybrid) models used:
- commercial software (LUIS) from a technology company (Microsoft) and customized it with the assistance of an external partner (ELCA)
- open source software (spaCy / Blackstone spaCy) customized with the assistance of the Hive for USA for UNHCR
Project used the two models of two different partners to allow for benchmarking and optimized problem-solving.