Extracting personal knowledge from users' utterances in conversational text, e.g., hobby:climbing. We mainly focused on implicit assertions, requiring the identification of hidden cues, which are then utilized for identifying the relevant personal attribute values. Investigated personal knowledge includes basic demographic attributes (age, gender and family status), and more challenging ones like occupations and hobbies.
A Large Reusable Dataset of Reddit User Traits. We envision further use cases of RedDust for providing background knowledge about user traits, to enhance personalized search and recommendation as well as conversational agents.