With the widespread of smart terminals, there appear a lot of applications to satisfy all kinds of needs in users' daily lives. Existing systems of smart terminals manage the background applications through an LRU-based task killing policy. LRU algorithm only considers applications' recent using logs instead of users' application using habits, it'll increase the application restart ratio and also lead to problems like increasing application launch delay, increasing energy consumption and losing state. In this paper, it designs and implements a Bayesian network-based application management policy named BNLP, which analyzs users' application using behavior to predict the probability of launching each application in the near future. Then BNLP will manage the background applications according to these probabilities. Experiment results show that our BNLP model can decrease the restart ratio of LRU by 17.2%, which greatly reduces launch delay, energy consumption and improves user experience.