In order to fit the diverse scenes in life, more and more people choose to join different types of social networks simultaneously. These different networks often contain the information that people leave in a particular scene. Under the circumstances, identifying the same person across different social networks is a crucial way to help us understand the user from multiple aspects. The current solution to this problem focuses on using only profile matching or relational matching method. Some other methods take the two aspect of information into consideration, but they associate the profile similarity with relation similarity simply by a parameter. The matching results on two dimensions may have large difference, directly link them may reduce the overall similarity. Unlike the most of the previous work, we propose to utilize collaborative training method to tackle this problem. We run experiments on two real-world social network datasets, and the experimental results confirmed the effectiveness of our method.