Furthering the development of AI based Historical Homicide files analysis with the help of Large Language Models

In this project, we are working on a large repository of historical homicide case files, which is majorly unstructured text. We are building a framework for knowledge graph extraction with the help of existing information extraction models further augmented and made robust with the help of existing state-of-the-art publicly available LLMs’ zero/few shot capabilities. We believe, our system will achieve more accuracy and coverage in terms of entity and their relationship detection embedded in those unstructured files and work as a foundational data structure to help build efficient Query and analysis tools for the domain experts, such as detectives to solve their criminal investigation roadblocks. In future, we will use the generated ideas from this project to extract structure from other types of unstructured or semi-structured documents to aid for advanced analysis of investigative files.

Faculty Supervisor:

Randy Goebel

Student:

Partner:

Edmonton Police Service

Discipline:

Computer science

Sector:

Public administration

University:

University of Alberta

Program: