This paper proposed an optimized solution for user recommendation in research social network. The major method applied was PSO with time window combined with K-means clustering. The time window was brought in to raise the accuracy of optimal solution searching and locate better cluster centers. The test showed that the PSO with time window improved the accuracy by 95% although 18% more time were cost. The empirical experiment indicated that this integrated algorithm could identify the similar user accurately, and it's valuable in practice.