Abstract

Abstract: One of the important targets of community-based questionanswering (CQA) services, such as Yahoo Answers, Quoraand Baidu Zhidao, is to maintain and even increase the num-ber of active answerers, that is the users who provide answersto open questions. The reasoning is that they are the en-gine behind satised askers, which is the overall goal behindCQA. Yet, this task is not an easy one. Indeed, our empir-ical observation shows that many users provide just one ortwo answers and then leave.In this work we try to detect answerers that are aboutto quit, a task known as churn prediction, but unlike priorwork, we focus on new users. To address the task of churnprediction in new users, we extract a variety of features tomodel the behavior of Yahoo Answers users over the rstweek of their activity, including personal information, rateof activity, and social interaction with other users. Severalclassiers trained on the data show that there is a statisti-cally signicant signal for discriminating between users whoare likely to churn and those who are not. A detailed featureanalysis shows that the two most important signals are thetotal number of answers given by the user, closely relatedto the motivation of the user, and attributes related to theamount of recognition given to the user, measured in countsof best answers, thumbs up and positive responses by theasker.