About this Event
Measuring Demographic Biases and Stereotypes in Textual Data Using Computational Methods by
Women, ethnic minorities and members of the working class tend to be underrepresented in the media and political discourse: they speak, and are spoken about, less. When they do appear, they tend to be quoted or described in stereotyped terms. This course offers a pipeline for measuring both patterns at scale:
1. Census. Named-entity recognition, coreference resolution and quote extraction turn a corpus into a population of persons and utterances.
2. Inferring demographic category of "gender, class, origin” from first names, honorifics and occupations, and quantifying the accuracy.
3. Analysis. How large is each group's share, and how does it compare to general population? Where is it concentrated (women in culture pages, working class in sport)? Which verbs tend to introduce their speech, which vocabulary describes them? Can we build semantic axes that separate them (eg: are women quoted with softer/emotional verbs)? Do these biases fade over time?
Session Details
Time: 10.00 - 15.00 (12.00 - 13.00 Lunch break)
Date: 16 November 2026.
Mode: Hybrid: In-person at CON.1.01 and online via Zoom.
If you are an ESRC funded student then we require your name, email and course to report your attendance to your home institution, please click to see the privacy notice that this data collection is subject to.
Event venue & nearby stays
CON 1.01, Connaught House, LSE, 65 Aldwych, London, United Kingdom
Concerts, fests, parties, meetups — all the happenings, one place.