Abstract: Aiming at the problem that the traditional similarity calculation method only used the user’s rating information as the explicit feedback behavior to recommend, which led to the unsatisfactory recommendation effect, we proposed a new similarity calculation method to improve the accuracy of the recommendation by introducing implicit feedback information, such as weight adjustment mechanism and user behavior preference. Firstly, according to the anti-user frequency of negative sampling, the influence degree of global software engineering of popular items was reduced, and the minimum weight of the common scoring behavior was used to adjust the recommended accuracy deviation caused by few common scoring. Secondly, we proposed the definition of project preference words, and calculated users with common preference in project characteristics according to the matrix of project preference words. Finally, the experimental comparison and analysis on the MovieLens data set. The experimental results show that the improved similarity calculation has better MAE value and higher recommendation accuracy.

Key words: weight adjustment, user preference, collaborative filtering, recommendation accuracy