Journalism Representation Index (JeRI)

In partnership with NovaceneAI, computer science and journalism researchers from Toronto Metropolitan University will continue to develop an AI-powered product (Journalism Representation Index, or JeRI) that gives journalists and media consumers a tool with which to analyze the sources cited in journalism reports. The primary goal of the platform is to provide an evidenced-based structure for improved journalism that can show statistically significant under-representations of certain kinds of sources in journalism texts and corpi. Our tool, JeRI, exists in a prototype form, developed by the team with funding from a Google award to drive innovation related to equity and inclusion in the news ecosystem. Prototype testing of our pipeline, which was trained using a semi-supervised machine-learning approach with a BERT, case-base model at its core, achieved F1 scores for the aforementioned categories ranging from 0.58 to 0.78 but these scores need improving. In testing our platform, we hand-coded 60 articles for Indigenous sources in any one of our seven categories. That process yielded 24 articles with at least one identifiable Indigenous, First Nations or Métis source. Out of the 24 articles there are 30 identifiable such sources. JeRI categorized 5 out of
the 30 correctly. For example, the platform identified First Nations Chiefs in the authority category instead of the politician category in the story “Review critical of arrest of Indigenous man and granddaughter in opening bank account.” It Identified Marilyn Slett, chief councillor of the Heiltsuk Nation, asan Authority source. These miscategorizations need to be addressed specifically.

Faculty Supervisor:

Gavin Adamson

Student:

Partner:

NovaceneAI

Discipline:

Computer science

Sector:

Artificial Intelligence; Information and Communications Technology

University:

Toronto Metropolitan University

Program: