Look Who’s Talking: Inferring Speaker Attributes from Personal Longitudinal Dialog

We examine a large dialog corpus obtained from the conversation history of single individual with 104 partners. The consists half million instant messages, across several messaging platforms. focus our analyses on seven speaker attributes, each which partitions set speakers, namely: gender; relative age; family member; romantic partner; classmate; co-worker; and native to same country. In addition content we conversational aspects such as time messages are sent, frequency, psycholinguistic word categories, linguistic mirroring, graph-based features reflecting how people in mention other. present two sets experiments predicting attribute using (1) short context windows; (2) larger messages. find that all leads gains 9–14% over message text only.