In the big data environment, by establishing association relationships and constructing network structures, the deep rules behind the data can be mined. This paper uses graph network structure to represent the sparse clustering relationship within the asset portfolio. The least-squares strategy with network structure and low rank sparseness is used to effectively target the index, in order to explore the effective feature factors and optimize the continuous-time asset portfolio. Finally, better Smart Beta performance was obtained. Studies have found that the sparse network structure based on link prediction can better capture the non-linear dependencies between assets and achieve effective asset classification results. Sparse decentralized regression and network structure feature extraction methods can profoundly reveal the inherent characteristics of assets. The robust regression strategy based on the least-squares method can adaptively optimize the tracking strategy, effectively solve the optimal investment decision problem in an incomplete market from the perspective of Alpha and Beta separation, and improve the overall performance of the investment portfolio. It has important guiding significance for index portfolio management