Via online social interactions, users in social networks can form their personal attitudes toward other users. Some of the personal social attitudes will be expressed explicitly, which are represented as the signed social links from the initiators to the recipients. In this paper, we will study the 'social Attitude exPression prEdiction' (APE) problem, which aims at inferring both the expression activities and the expressed social attitudes simultaneously. The APE problem is very challenging to address due to two reasons: (1) extraction of useful features for predicting social attitude expression activities and the expressed social attitudes, and (2) the prediction model needs to incorporates the correlations between these two tasks covered in the APE problem. To address the APE problem, a novel real-time 'Bayesian network based Integrated Social Attitude exPression inference' framework BI-SAP is introduced in this paper. Framework BI-SAP extracts a set of features for the expression activities and social attitudes from the networks based on various social closeness measures and social balance theory respectively. In addition, with the extracted features, the integrated social expression prediction framework BI-SAP is built based on the Bayesian network model, in which the dependence relationships between these two tasks covered in APE can be effectively represented and the parameters can be updated in real time. Extensive experiments conducted on real-world signed network datasets have demonstrated the effectiveness of BI-SAP in addressing the APE problem.