Particle Swarm Optimization BP Neural Network In Collaborative Filtering Research In The Algorithm

With the explosive growth of the amount of network resources,it is more difficult for users to obtain effective information.In this context,the recommendation system emerges as the times require.Nowadays,the recommendation algorithm has been widely used in different application scenarios such as e-commerce.Different scholars have taken corresponding measures to improve the most widely used collaborative filtering recommendation algorithms in the current mainstream commercial recommendation system.The most critical step of the collaborative filtering algorithm is the user(or project)similarity calculation,but the traditional algorithm has lower prediction accuracy when the scoring matrix data is sparse.Therefore,in view of the above problems,this paper proposes an improvement based on the neighbor user scoring error.Collaborative filtering algorithms,the main contributions are as follows:1.The improved algorithm adopts reverse thinking,and considers the influence of the characteristic attributes of the project on the user's scoring behavior.In addition,considering the complexity of user scoring data,this paper uses BP neural network to deal with the advantages of complex nonlinear problems.By combining the characteristics of global optimization of particle swarm optimization algorithm,a scoring prediction model between user and project attribute features is constructed.Referred to as the PSO-BP scoring prediction model,the error of the neighbor user prediction score output by the model and the actual score of the same project history of the target user is used as a criterion to determine the degree of similarity between users,thereby improving the traditional recommendation algorithm in the scoring matrix.In the case of sparse data,the accuracy and reliability of predictive scoring are low,and the accuracy of scoring prediction of traditional algorithms in the case of sparse matrix data is improved.2.In this paper,the similarity matrix calculated by the custom similarity calculation function has lower dimension,which achieves the purpose of reducing the dimension,reduces the system memory overhead,and improves the algorithm scalability.In this paper,using MovieLens and InternetBook dataset,the improved algorithm and traditional collaborative filtering algorithm and other scholars' improved algorithms are tested on three different scaled datasets.Finally,the improved algorithm proposed in this paper is predicted.Both robustness and robustness are superior to traditional collaborative filtering algorithms.