As the label is marked by the user according to their own understanding and preferences, the expression of the concept is fuzzy and there are a large number of noise tags, resulting in the low efficiency of the traditional label-based recommendation algorithm recommended. casein view of this problem, a tag recommendation algorithm combining the score information entropy is proposed. The algorithm determines the importance of the tag for the user in order to build the user's interest model for the rating of the label. The algorithm can effectively use the score weight and combine the information entropy to enhance the recommendation accuracy, and compared with the previous label-based recommendation algorithm, it can get a satisfactory recommendation effect.