Modeling Conversations in Social Media

Опубликовано 17 августа 2016, 3:24

Users of social media websites are engaging in public conversations at an unprecedented scale. This massive corpus of naturally occurring, open-domain conversations presents new challenges and opportunities for data-driven conversation modeling. In this talk I will describe work on automatically generating replies to open-domain Status Messages in Twitter. In order to leverage millions of naturally occurring conversations, we adapt techniques from Statistical Machine Translation (SMT), to build a models capable of ΓÇ£translatingΓÇ¥ an arbitrary status message into an appropriate response. Open-domain response generation could be useful for several applications including language generation in dialogue systems, and conversationally aware predictive text input. In addition I will describe work on unsupervised induction of dialogue acts in Twitter. By remaining agnostic about the set of dialogue acts, we are able to learn a model which provides insight into the nature of communication in a new medium.