Broadcast Message Prioritization

Broadcast message is defined as a message that can be sent to a group of subscribers at once.Popular types of broadcast messages include broadcast emails,tweets,broadcast short messages and so on.With billions of broadcast messages sent and received everyday,they are fundamentally impacting the way people live,work and communicate.However,there comes a curse with the broadcast messages,the broadcast message overload problem.The majority of broadcast messages are usually unimportant or irrelevant and people have to waste a large amount of time handling them,causing a trillion-level economy loss in productivity.The serious situation leads to a thriving research field,personalized broadcast massage prioritization,which aims to predict the importance label for broadcast messages.In this thesis,I focus on broadcast message related research questions on two popular types of broadcast messages,broadcast emails and tweets.I first work on the mention recommendation problem,which tries to recommend the optimal set of users to be mentioned in a tweet to expand its diffusion.Considering tweet prioritization(users' preference of the tweet)and user influence at the same time,I propose the first framework to handle the mention recommendation problem by designing a novel learning to rank model.In the second work,I focus directly on the personalized broadcast email prioritization task.I proposed the first broadcast email prioritization framework that takes collaborative filtering into consideration.To handle the complete cold start item challenge,a novel active learning framework is proposed,considering unique characteristics of our task,like one-class implicit rating and time sensitive feedback.In the third work,I continue to work on broadcast email prioritization problem while considering the fact that there exist large numbers of mailing lists in a real system.A cross domain recommendation framework is proposed to transfer extra knowledge from other similar mailing lists.I propose the first cross domain recommendation approach that can automatically select the optimal set of source domains from large numbers of candidate domains.All of our research works are evaluated on real life datasets.I not only compare them with the state of art baselines,but also evaluate how different new features proposed by us affect the final performance.Comprehensive experiments show the effectiveness of our proposed methods.