In many segmentation applications, data objects are often clustered based purely on attribute-level similarities. This practice has neglected the useful information that resides in the link structure among data objects and the valuable expert domain knowledge about the desirable cluster assignment. Link structure can carry worthy information about the similarity between data objects (e.g. citation), and we should also incorporate the existing domain information on preferred outcome when segmenting data. In this paper, we investigate the segmentation problem combining these three sources of information, which has not been addressed in the existing literature. We propose a segmentation method for directed graphs that incorporates the attribute values, link structure and expert domain information (represented as constraints). The proposed method combines these three types of information to achieve good quality segmentation on data which can be represented as a directed graph. We conducted comprehensive experiments to evaluate various aspects of our approach and demonstrate the effectiveness of our method.