Identifying Actionable Messages on Social Media

摘要

Text actionability detection is the problem of classifying user authorednatural language text, according to whether it can be acted upon by aresponding agent. In this paper, we propose a supervised learning framework fordomain-aware, large-scale actionability classification of social mediamessages. We derive lexicons, perform an in-depth analysis for over 25 textbased features, and explore strategies to handle domains that have limitedtraining data. We apply these methods to over 46 million messages spanning 75companies and 35 languages, from both Facebook and Twitter. The models achievean aggregate population-weighted F measure of 0.78 and accuracy of 0.74, withvalues of over 0.9 in some cases.