Summary: Most of the existing literatures on the application of complex network theory to the stock market have studied the topology of stock network, but the combination with portfolio theory is scarce. In this paper, from the perspective of complex network theory, two kinds of community detection algorithms are used to divide the stock network into communities, and block matrix is used to describe the community structure of the stock network, so as to construct a new portfolio. Empirical analysis on the SSE180 index shows that when the expected return is fixed, the risk of the new portfolio constructed in this paper is lower than that of Pearson portfolio, and the effective frontier is to the left of the effective frontier of Pearson portfolio. The effectiveness of the new portfolio model is verified by empirical evidence.