| With the development of technology,the informationization of library management has gradually replaced the traditional management mode,and due to the expansion of the range of majors covered by universities and the increase of university library collections,university teachers and students have more needs for libraries,and they hope to be able to choose more suitable books among the complicated books more efficiently,so the introduction of personalized recommendation algorithms in university libraries has become particularly important.However,the more mature personalised recommendation algorithms generally suffer from high data sparsity,low recommendation accuracy and high reliance on rating data.In this dissertation,we propose a pre-population algorithm to reduce the sparsity of data,and a hybrid algorithm of collaborative filtering and association rules to reduce the reliance on rating data.The main work of this paper has the following aspects:(1)A hybrid pre-population algorithm that incorporates the combined similarity of attributes is proposed.The algorithm introduces user attributes and item attributes respectively,calculates the combined similarity of items by using the item attributes and user rating matrices,and makes rating predictions based on the item nearest neighbours,fills the rating matrix with high sparsity for the first time,and then calculates the combined similarity of users by using the first filled rating matrix and user attributes,makes rating predictions for items based on the user nearest neighbours,and weights the two predictions The scores obtained from the two predictions are then weighted and the scoring matrix is populated twice.Through comparative experiments,the hybrid pre-population algorithm incorporating the combined similarity of the attributes proposed in this paper can make the rating matrix more realistic compared to the general default and mean-population methods,while reducing its sparsity and laying the foundation for subsequent recommendations.(2)A hybrid recommendation algorithm based on collaborative filtering and association rules is proposed.The algorithm mixes collaborative filtering and association rules using a parallel strategy.The collaborative filtering algorithm makes recommendations based on the user rating matrix and item attributes,and the rating matrix is pre-populated by the mix to improve the accuracy of the algorithm’s recommendations.The prediction scores generated by the two algorithms are weighted together to form Top-N recommendations.Through experimental verification,the hybrid recommendation algorithm based on the hybrid pre-populated rating matrix has significantly improved the overall recommendation accuracy and its coverage by reducing the sparsity of data and mining the relationships between attributes.(3)A university library book recommendation system is designed and implemented,and the proposed hybrid recommendation algorithm after the hybrid pre-population scoring matrix is applied to the recommendation system,proving that the hybrid pre-population algorithm and the hybrid recommendation algorithm have good application value and can meet the needs of users and provide more targeted personalised book recommendations for users. |
