Using Machine Learning to Detect Explanatory Strategies

Project Lead

Dr. Robert Clapperton

Emerging Researcher(s)

Amin Mirlohi

This project is working to design and develop a machine learning model for the detection of elements such as temporal focus (deliberative, forensic, epideictic), evaluation, and stance in journalistic texts. The deliberative, epideictic, and forensic modes of rhetoric are strategies that journalists can use to shape their content, ultimately impacting the audience’s understanding of the situation being discussed. These modes are temporal in nature, with forensic rhetoric focusing on establishing facts about the past, epideictic rhetoric stating proclamations about the present, and deliberative rhetoric outlining possible futures while attempting to convince audiences to work towards (or against) these outcomes. 

Dr. Clapperton’s team is developing a custom IBM Watson model that can detect whether a sentence within a particular article features deliberative, epideictic, or forensic rhetoric. The goal is a model that can differentiate aspects of explanatory journalism from informational reporting or opinion-editorial types. The model will be trained to identify and learn explanatory strategies.

This project is partnered with Dr. Catherine Schryer’s team and is working to operationalize their codebook for uptake of explanatory journalism. The team is annotating an initial set of articles, identifying each word of each sentence in certain grammatical categories to establish a base for the machine learning model to work from. From there, the team will keep “teaching” the model, with the ultimate goal of it being able to find the major explanatory strategies used in any Conversation article given for analysis. This will be an effective tool in understanding how rhetoric functions in explanatory journalism and in The Conversation in particular. Once we understand the rhetorical functions, we can then begin to understand the audience impact that these functions may have.