Research With Key Technologies Of Mobile Socialized E-commerce Recommendation

Based on the basic model of the mobile social e-commerce recommendation system,we focus on the key technologies such as high-dimensional feature information selection,similar trust relationship between items,similar trust relationship between users,direct and indirect trust communication relationship between users.In addition,the research results of feature selection algorithm in content-based recommendation algorithm and recommendation algorithm based on social collaborative filtering(including scoring-oriented and sorting-oriented)are deeply studied.Specific research contents and results are as follows:1 In the field of feature selection based on content recommendation of the mobile social e-commerce:related research on how to select highly relevant,delete-independent and redundant feature subsets from high-dimensional feature data,and the results of the research are as follows:(1)A maximum correlation minimum redundancy feature selection algorithm based on approximate Markov blanket is proposed,which includes two stages:feature correlation sorting and redundant feature deletion:In the first stage,the relevant correlation ordering is performed by using the maximum correlation minimum redundancy criterion,and the forward iterative search method is used to select the optimal features;In the second stage,the approximate Markov carpet method is used to analyze the relationship of the features and labels and the dependencies between features,and the features with high degree of interdependence between features are deleted,and the features with high discriminative ability are retained to form the optimal feature subset.Because the algorithm has the ability of selecting feature subset at the initial stage,it further improves the generalization ability of classification learning algorithm.(2)At present,many feature selection algorithms have the phenomenon of missing redundant and irrelevant features,which is leading to overestimation of some features.Moreover,more features will significantly slow down the speed of the mobile social e-commerce content recommendation and lead to classification over-fitting.Therefore,a new nonlinear feature selection algorithm based on forward search was proposed.The algorithm used the theory of mutual information and mutual information to find the optimal subset associated with multi-task labels and reduced the computational complexity.Compared with the experimental results of 9 datasets and 4 different classifiers in UCI,the algorithm is superior to the feature set selected by the original feature set and other feature selection algorithms.2 In the filed of the mobile socialization e-commerce recommendation algorithm based on social collaborative filtering,taking social trust relationship as the core,aiming at data sparsity and cold startup,a series of related research on recommendation algorithm of the mobile socialization e-commerce collaborative filtering oriented to scoring and ranking has been carried out.The specific research results are as follows.(1)Few studies currently consider using hidden core item trust relations to solve the data sparsity of the mobile socialization e-commerce.A new mobile socialization e-commerce recommendation algorithm for incorporating the implicit core item trust relationship into the probability matrix(Item Related Matrix Factorization,IRMF)is proposed in this paper.Firstly,through the activity and trust of the item,the core item set is identified,and the similarity between the items is calculated by the Pearson formula,and the similarity value is stored in the hidden core item trust matrix;Secondly,the solution process of the IRMF algorithm is constrained by sharing the feature space of the item;Finally,it is proved by experiments that the IRMF algorithm is superior to other classical recommendation algorithms in three different data sets.(2)The existing socialized recommendation algorithm only uses the trust network to model directly,and ignores the user feature information and connection relationship hidden behind the trust relationship.Therefore,this paper proposes a new mobile socialization e-commerce recommendation algorithm.Firstly,the algorithm constructs a similar trust matrix by calculating the credibility between users.Secondly,it shares the user space between the trust matrix and the similar trust matrix,and incorporates the direct and indirect trust propagation mechanism into the model.Finally,the algorithm completes the personalized recommendation to the target user.The experimental results show that the proposed algorithm not only can effectively improve the recommended Root Mean Squared Error(RMSE)and Mean Absolute Error(MAE)values,but also effectively solve the problem of recommendation quality degradation caused by data sparsity and user cold start.(3)In the the mobile socialization e-commerce,users are usually connected.At the same time,each user has different social influences and trusts different friends in different the mobile socialization e-commerce networks.In order to model the social influence and trust relationship in different social networks,a the mobile socialization e-commerce network recommendation algorithm based on different users of the social influence and trust relationship in different social network,which combine the user scoring matrix and social trust relationship,is proposed.Firstly,the algorithm uses the SocailRank algorithm to calculate the social influence of different users in social networks;Secondly,the algorithm uses in-link and out-link between the different users to calculate the trust relationship of each user;Finally,the social influence and trust strength relationship the relationship in different social networks is incorporated into the probabilistic matrix factorization model,which is a highly unified and credible mobile socialization e-commerce recommendation algorithm.The algorithm is compared with a variety of related algorithms on the Ciao and CiaoDVD datasets.The experimental results show that the proposed algorithm has a big improvement in the recommendation accuracy,compared with the other social recommendation algorithm.(4)The traditional social recommendation algorithm can not accurately grasp the user preference information for sorting recommendation.In this paper,we propose to optimize the matrix decomposition model by using BPR framework with low computational complexity,and further improve the precision and accuracy of items recommendation sorting by mining similar relationships between users and direct and indirect relationships between trusting users.In the Douban dataset and Filmtrust dataset,compared with the mainstream recommendation sorting algorithms such as SBPR,TBPR,BPRMF and MostPopular,the experimental results show that the mobile socialization e-commerce algorithm is superior to other sorting recommendation algorithms.