Research On Improving Top-N List Diversity Algorithm In Book Recommendation System

With the rapid development of information technology,massive data is reflected in people's eyes.The advent of recommended technologies helps people quickly find preferred items.The huge amount of information and increasing user demand have contributed to the development of recommendation algorithms.The recommendation system has received the attention of experts and scholars and is widely used in industry.However,the traditional recommendation system pays too much attention to the accuracy of the recommendation and ignores other indicators of the recommendation system.Sometimes it is more important to pay attention to the overall satisfaction of the user and a recommendation system with high accuracy is not necessarily a good recommendation system.In recent years,experts and scholars have gradually increased their attention to the diversity of recommendation systems.More and more people are turning their attention to improving the diversity of recommendation systems,but the increase in general diversity will bring greater losses to accuracy.Therefore,how to balance these two indicators is still a difficult problem to solve.In this paper,a recommendation algorithm for diversity improvement that balances the relationship between accuracy and diversity is proposed.The algorithm is applied to the book recommendation system.Based on this paper,the following work and contributions are mainly completed:1.A candidate set algorithm for improving accuracy is proposed.Accuracy,as a key indicator of user satisfaction,plays an important role in the recommendation system.It also plays an important role in the diversity promotion algorithm.This paper proposes a Deep Dense Crossing Network(DDCN),an improved algorithm based on the NFM algorithm.The algorithm combines the good characteristics of deep learning and feature crossing.The embedding layer is used to map the input sparse data,then the mapped feature vectors are used as feature intersections,and the first-order feature vectors with weights and second-order features are intersected.The vectors are spliced into a vector as input to the subsequent deep network,which facilitates the learning of higher order cross features.The deep learning network part uses the densely connected structure of computer vision and transforms it into a deep neural network structure suitable for the recommendation system.Such a structure greatly enhances the reusability of features and enhances gradient propagation and slows down gradient vanish problem with the increasing of network layers.The improvement of the above algorithm is effectively improved on the two data sets.2.A diversity lifting algorithm U-MMR is proposed.This algorithm is an improved version of the MMR algorithm.Based on the MMR algorithm,the U-MMR algorithm replaces the accuracy calculation part with the user book preference score calculated by DDCN,and calculates the user diversity preference index as an adaptive parameter for diversity and accuracy balance improvement solving the problem of lack of user diversity preference brought by the MMR algorithm.At the same time,this paper applies the unsupervised text similarity calculation algorithm used in natural language processing to recommender system of book,which can accurately measure the similarity of books.Finally,the U-MMR algorithm proposed in the above modified part proves to have good performance.3.The overall design and implementation of the book recommendation system is introduced.The candidate set algorithm and the diversity promotion method are applied to the mobile phone reading platform provided by"Mi Gu",and the overall framework of the system is introduced.