Enterprise-Oriented Policy Push Algorithm

In the traditional collaborative filtering recommendation algorithm, the similarity calculation of users is only based on cosine similarity; in the rating prediction link, only the direct neighbors of users are used for prediction. Therefore, under the circumstance that the rating matrix of enterprises on policies is highly sparse, traditional collaborative filtering has the problem that it cannot accurately predict the attitudes of enterprises towards policies and implement policies to corresponding enterprises in a timely manner. This paper proposes an enterprise-oriented policy push algorithm, which incorporates the extreme attitudes and characteristics of enterprises into the similarity calculation process. When the rating matrix is highly sparse and cannot be predicted accurately by relying on direct neighbors, iterative prediction is performed by referring to indirect neighbors and using z-score to eliminate rating bias. The experiments are carried out on the enterprise-policy dataset collected in the article and the film-trust dataset commonly used in recommender systems. The experimental results show that the algorithm reduces the mean absolute error by 5.67% and 1.54% respectively compared with the iterative rating prediction algorithm, which shows that the algorithm has achieved good optimization in the recommendation accuracy.