Application-Layer multicast (ALM) is an important supplement to IP multicast. However, unlike in IP multicast, the participating nodes in ALM are selfish and strategic host users. In order to improve their own interests, selfish host users might not strictly obey the ALM protocols, because of which the overall performance of the multicast session could be affected. To design robust and trustworthy ALM protocols, it is necessary to study the selfishness in ALM. This paper surveys the recent research trends in this area, and classifies the researches into three categories according to the working steps of ALM protocols, that is, the selfishness in maintenance of control structure, the selfishness in collection of node information, and the selfishness in construction of data structure.