Today, social networks are very popular all over the world due to the possibility of communication between different people, and their use has become more widespread day by day. One of the main problems in analyzing these networks is predicting relationships between people. Therefore, link prediction in social networks has been considered by researchers. In this article, we have tried to briefly review the methods proposed in the field of link prediction. For this purpose, while studying the types of social networks, it categorizes the similarity criteria between the entities in the social network, in order to facilitate the selection of the most appropriate method of link prediction by recognizing these criteria. According to studies, non-supervisory methods are more accurate than supervised methods. Features used to determine the similarity between entities extracted from a network graph include local, semi-local, and global features, where local features have the advantage of speed and global features have the advantage of accuracy. Semi-local features are also provided to balance the accuracy and complexity of computation.