Research On Commodity Recommendation Algorithm Based On Graph

With the rapid development of information technology and Internet,more and more people choose online shopping giant,commodity business platform information is more and more complicated,the explosive growth of commercial users into information trek,more and more difficult to quickly and accurately retrieve the goods.Personalized recommendation algorithm plays an important role in solving this problem.Among the many recommendation algorithms,graph based recommendation algorithm has come to the fore,more and more scholars begin to study,but there are still some problems.In this paper,two kinds of graph based recommendation algorithms are proposed to solve the problem of "interest drift","long tail effect" in the recommendation process,and the problem that the performance of the graph is not fully considered in the recommendation process.In order to solve the problem of interest drift,we combine the forgetting theory with the three graphs which are composed of the user,the commodity and the keywords which can represent the goods,and the forgetting mechanism is used as the basis for the calculation of the weight of the three graphs;In order to solve the problem of long tail effect that often occurs in the process of recommendation,we combine the theory of material diffusion and the three graph weighted by forgetting mechanism.Analysis of the current main requirements of the recommendation algorithm,the accuracy of the recommendation diversity and recommend sorting the list as the evaluation indexes of the performance of the algorithm were collected on the Movielens-10 M data set,verified the rationality and validity of the algorithm.In order to solve the traditional graph construction method recommended map only consider the goods 22 similarity or only through the commodity attribute of simple integration build recommendation network diagram of network object complexity and dependence is not considered in the problem of low accuracy of recommendation,we use a mixed graph composed of various object business platform in the construction process without considering the heterogeneity of the object,the transfer process of nodes will be system logs,user behavior and businesses pay information,both the user and user behavior history and current demands of interests of businesses,platform into account.Increase the click through rate and sales of goods,but also to enhance the satisfaction and loyalty of businesses and users on the site,increase platform revenue.The algorithm is applied to the local electricity supplier website and the data are taken to verify the algorithm,which proves the accuracy and recall of the algorithm and the superiority of the harmonic value.The two improved algorithms do not need the user's active participation,and will not interrupt the user's current operation.