Contextual Affective Analysis
Our work takes a people-centric approach to understand subtle connotations in narrative text. By quantifying how people are portrayed in different contexts, we can analyze narratives for signs of bias. To quantitatively measure people’s portrayals, we draw from social psychology theories suggesting that sentiment (good-bad), power (dominant-submissive), and agency (active-passive) are the most important dimensions of affective meaning.
Our methodology involves using verb-annotations to score people along these dimensions. Our initial work analyzes media coverage of the #MeToo movement, revealing that women are often portrayed as less powerful than men in these narratives, despite the common perception of the movement as empowering (Field et al., 2019). In subsequent work, we incorporate noun and adjective annotations to analyze media bias more generally (Field and Tsvetkov, 2019). Furthermore, our current work involves gathering new data sets, facilitating analysis across languages, and allowing us to analyze how LGBT people are portrayed in different languages on Wikipedia (Park et al., 2020).
People
Anjalie Field
Chan Young Park
Related Papers
- Multilingual Contextual Affective Analysis of LGBT People Portrayals in Wikipedia, (Park et al., 2021), ICWSM
- Contextual Affective Analysis: A Case Study of People Portrayals in Online #MeToo Stories, (Field et al., 2019), ICWSM
- Entity-Centric Contextual Affective Analysis, (Field and Tsvetkov, 2019), ACL