Understanding Relationships & Interactions

Identity and Subjectivity in Relationships & Interactions: In this roundtable, we will talk about how identity and subjectivity are bound up with relationships and interactions. We will discuss how computational approaches to identity can shift to take a relationship/interaction-first perspective.

Understanding Conflict and Abusive Language: What knowledges and imaginaries are necessary to employ machine learning to create more responsive (rather than retroactive or reactionary) interventions to abuse. How might we direct AI to engage in conflict to foster user- and community-based empowerment – to understand the capacity of what emerges in conflict as a productive space to equitably intervene in and redirect abusive conversations?

Character networks for interaction modeling: A convenient approach to character-centric media content analysis is to construct a network of characters, where each character is a node and their pairwise relationship is encoded in the edge connecting them. Such representation offers several advantages in efficient retrieval and comparison of media content, and also can be used to generate insights about the storytelling process. The majority of computational work relies on co-occurrences to build such networks, i.e. two characters appearing on screen in the same or consecutive shots or having consecutive dialogs indicate interaction. This approach, however, cannot distinguish between mere co-occurrence from the strength of the relationship; nor can this identify the type of relationship between a pair of characters. This brainstorming session will engage in a discussion about new approaches to construct character networks and computationally viable ways of modeling character interaction along with their challenges.

The Ethics of Measuring Body Size: This session will unveil a new automated tool for measuring body size and discuss the ethics of how to gather and use this information. More specifically, we will discuss how automated body size measures may be used to advance fatphobia and discrimination, and ways to avoid these pitfalls.

Taxonomies for Relationships & Interactions: In this roundtable we will focus on how to represent social relationships and interactions in the machine, discussing taxonomies that can be used. For example, if we want to characterize an interaction, in some occasions we might be interested in understanding who is in control, and this will require a representation in the power domain. However, in other cases we might be interested in understanding whether the interaction was positive or negative, which can be encoded in a continuous dimension. In this roundtable we will discuss what categories and dimensions are relevant or necessary for the automatic understanding of social relationships and interactions in computational media, in order to characterize relationships and interactions in a rich and meaningful way manner.