Abstract: Combined with the rating matrix of user-item and the correlation matrix of itenrcategory, a new hybrid rec- ommended model was proposed. First, a new correlation degree measuring algorithm was presented by using these two matrixes. This algorithm takes into account the feature information and dynamically adjusts the result based on the sparse situation of the rating data, truly reflects the degree of association with each other. Then, a new weighted two- layer graph model was constructed by using the item-item correlation degree and the user-item correlation degree as the weight. On this basis, starting from the global structure of the two-layer graph, the recommendation algorithm based on weighted two-layer graph was given by the random walk algorithm, to provide users with personalized item recommen- lotions and user recommendations. The experiments show that the algorithm compared to other recommended models in the references has higher accuracy.

Key words: Random walk, Hybrid recommendation filt}ring,Item category information,hwo-Layer graph