Research On PMF Recommendation Algorithm Combining EMD Distance And User Interest

With the rapid growth of data information,the phenomenon of information overload appears.In order to effectively alleviate the problem of information overload,researchers put forward the concept of personalized recommendation and carried out in-depth research and a large number of practical applications,and achieved good economic benefits.Collaborative filtering recommendation technology as an important branch of personalized recommendation is one of the most widely used and successful personalized recommendation technologies.Collaborative filtering algorithms can be divided into project-based collaborative filtering,user-based collaborative filtering and model-based collaborative filtering.Modelbased collaborative filtering is the most popular collaborative filtering type.However,the probability matrix factorization model is widely used in collaborative filtering recommendation algorithm because of its advantages of easy programming and high recommendation accuracy.However,the basic probability matrix factorization model has cold start problem,the research found that the integration of user social relations into the process of probability matrix factorization recommendation can alleviate the cold start problem in the traditional recommendation system to a certain extent.In the actual recommendation environment,the social relationship between users is difficult to obtain.The sparsity of data needs to be considered in the construction of user social relations.The traditional similarity calculation method mainly calculates the final similarity between users through the user common scoring items.The more common scoring items,the more accurate the calculated similarity.When there is a data sparsity problem,there are few common scoring items between users,which can reduce the accuracy of similarity calculation results to some extent.In order to solve the problem that users' social relations are difficult to obtain,this paper proposes to use user similarity and trust factors to construct a user social relationship based on potential interest and trust.In order to solve the problem of sparse data in similarity calculation,a hybrid similarity algorithm based on EMD distance and user interest is proposed in this paper.The specific work of this paper is as follows:First,First,a hybrid similarity algorithm(I-EMD)is proposed,which combines EMD distance and user interestIn order to solve the problem of data sparsity,firstly,EMD method uses linear programming to add all users' scoring data to the process of similarity calculation,which changes the traditional method to calculate the similarity by only relying on users' common scoring items,and calculates the EMD similarity between users.Then,by using the user interest degree,the user's preference for each item attribute can be mined,so that the user interest element matrix with less sparsity of the item attribute can be constructed,and the user interest similarity between users can be calculated through the user interest element matrix.Finally,the weights ? are used to combine the two similarity values to obtain the comprehensive similarity of EMD distance and user interest between users.Through comparative experiments,it is found that the average absolute error(MAE)and p @ 10 values of the collaborative filtering recommendation algorithm based on I-EMD hybrid similarity are better than those of the traditional collaborative filtering recommendation algorithm based on cosine similarity.Second,A PMF recommendation algorithm combining similarity and trust is proposed(ST-PMF CF)In order to solve the problem that the social relationship between users is difficult to obtain,this paper proposes to use the similarity and trust between users to construct the social relationship between users based on potential interest and trust,in the process of similarity calculation,I-EMD hybrid similarity algorithm is used.Finally,the social relationship between users is added to the probability matrix factorization(PMF)model to complete the recommendation.Compared with the traditional collaborative filtering algorithm and the recommendation algorithm based on probability matrix factorization,the average absolute error(MAE)value and p @ 10 value of ST-PMF CF are better than the former two.