Research And Application Of Recommendation Algorithm Based On Matrix Decomposition And PSO Collaborative Filterin

With the rapid development of artificial intelligence and Internet,data search technology has brought great convenience to people’s lives.The huge amount of network data makes people enjoy new technology,but influences the efficiency and quality of data use.On the one hand,people can retrieve the information they need through surfing the Internet.On the other hand,when using retrieval technology to obtain relevant information,people need to provide more accurate information,such as keywords,otherwise they cannot get more accurate results.Information recommendation refers to providing the information people need actively through certain intelligent algorithms according to some matching rules.Recommendation algorithm research is the core technical support to realize this function.With the rise of intelligent information services,the research of recommendation algorithm and its application has attracted more and more attention.At present,the research on recommendation algorithm is mainly focused on e-commerce,video and music,and it has been successfully applied in the field of e-commerce,such as e-commerce information push,headline information push service and so on.However,there are still some problems that need to be solved through further research on recommendation algorithms,such as sparsity problem,high-dimensional,large data volume pushing efficiency and so on.The analysis and research of these problems can help to improve the theoretical system of intelligent recommendation algorithm,and it is really related to whether the commercial value of recommendation algorithm can be further explored.To deal with the clustering recommendation of unlabeled items,this thesis introduces PSO and k-means clustering algorithms from the perspective of semi-supervised learning.It also proposes a collaborative filtering recommendation algorithm based on the clustering of mixed labeled and unlabeled samples,which improves the accuracy while not increasing the sample labeling workload in the recommendation algorithm as much as possible.Meanwhile,to address the sparsity of user-item association matrix,a collaborative filtering hybrid recommendation algorithm based on matrix decomposition and PSO optimization clustering("MFPSOSSC-CF")is proposed by further mining the same or similarity of users’ ratings on product attributes and product attributes,and constructing a user-itemindicator association feature matrix.The proposed MFPSOSSC-CF algorithm is experimentally studied and the experimental results of the algorithm are compared with the corresponding metrics of the traditional User-CF.The results show the effectiveness of the algorithm studied in this thesis.Based on this,the personalized recommendation application system built to validate the validation algorithm is applied in simulation.