In order to improve the impact of data sparseness on the recommendation effect in the traditional recommendation system, a recommendation algorithm for improving asymmetric similarity and associated regularization is proposed. According to the asymmetry relationship between different users and different projects, an improved correlation calculation formula is proposed for predicting the score. At the same time, due to the difficulty in obtaining the implicit relationship of socialization, this paper uses the traditional similarity to obtain the neighborhood set as the user social relationship, and the regularization is used to constrain the matrix decomposition objective function to alleviate the data sparse problem caused by user information asymmetry. Finally, the algorithm is validated in some real dataset. The experimental results show that the algorithm can predict the actual score more effectively than the mainstream recommendation algorithms.

%U http://cea.ceaj.org/EN/10.3778/j.issn.1002-8331.1907-0280