Research On Evolution Mechanism Of The User Interest Graph

Based on user behavior in e-commerce site and the user demographicinformation, personalized recommendation algorithm recommends commodities tousers they may be interested in, thus it has been widely used in enterprises. But withthe explosive growth of the quantity of commodities, data sparsity has brought a bigchallenge to the accuracy of recommendation algorithm. Personalizedrecommendation algorithm researches began to involve the information of users insocial networking sites, through modeling cross-domain user interests, provided userswith more diversified and more accurate recommendation results. Recommendationbased on user interest raised the accuracy of recommendation algorithm, andenhanced the efficiency of recommendation. The problem with the algorithm is that,the user interests are constantly changing over time.Existed researches on the change of user interest simply studied the effect oftime factors on user interests, and measure it with the linear time model, but theforgotten of user interests is an uneven process. Or it only studied the change ofinterests which was caused by influence of other users in social networking sites, butonly when the user receives the influence and responses to it (such as relaying,comment), it shows the user interest corresponding to that information changes. At thesame time, the spreading of user interests are affected by the weight of user influence,traditional way to calculate the influence of users in information diffusion onlystudied the scale and amount of the information users release, it is unable to fully giveout the user information. In addition, the change of user interests are also affected bytheir consumption behaviors in e-commerce sites, which means the user interest is atriples of its own behavior, time factor and network space, so the evolution of userinterests must also combine the three aspects.The main problems the evolution mechanism of user interest graph combineindividual, time and network space face are:(1) the influence of the user behavior inthe e-commerce sites to his interest, and how to associated the commodities with userinterests;(2) some user interest was affected and changed, it will also affect theinterest which has high semantic similarity with it, the traditional diffusion activationtheory fails to complete the process;(3) the information users publish or browse insocial networks needs to be associated with users’ interests in calculation of weight ofuser influence;(4) combining the user behavior, time, and network space together, how to express the actual process of the evolution of user interest graph topersonalization recommendation is a problem.On the basis of the above problem, this paper puts forward the user interest graphevolution mechanism integrated individual-time-space, including user interestevolution caused by individual behavior, the attenuation caused by time and evolutionin network space, improves the process of the change of users’ interests. The mainwork includes:(1) interest spreading activation model in online social networks.(2) based on semantic diffusion path, the spreading of user interest in its internalinterest graph ontology is divided into three categories.(3) setting up the user consumer behavior model, with the aid of the WordNetand Linked Open Data, taking the attributes and labels of commodities into account,associating the commodities with user interests, so as to obtain the evolution of theusers’interests.(4) considering the network topology of user social network, associating theinformation users published with user interests, proposing the calculation method ofuser influence in network space.(5) integrating the algorithm and model above, establishing a user interest graphevolution mechanism integrated individual-time-space, including the change ofuser interests caused by individual behavior, the social networks, and the change withthe time in internal interest graph ontology, which completes the whole evolutionprocess of user interest graph.(6) on the basis of the proposed algorithm, collecting user data, use the results oftraditional interest graph evolution algorithm based on spreading acivation torecommend commdities for users, at the same time, comparing with therecommendation results of the interest graph evolution algorithm in this article,proved that the proposed algorithm has greatly improved the accuracy and thediversity of recommendation results.