Investigating Methods and Representations for Reasoning About Social Context and Relative Social Power

摘要

Social context has a profound effect on how people interact with each other, and should have important ramifications for how intelligent systems interact with people. However, social context has received comparatively little attention in research on context-aware systems. This paper begins by highlighting possible dimensions for descriptions of social-interactional context, based on social science research. An important component is the interactants' place in the social hierarchy, and especially their relative social power. The remainder of the paper presents results on using machine learning methods to learn cross-domain classifiers for predicting relative social power. An experimental evaluation of cross-domain learning between three domains suggests that the important features for determining whether or not one interactional domain can be used to predict the relative social power of interactants in another are which social power dimension has the most influence in a given domain.