Research And Implementation On Personalized Service Recommendation Mechanism For Cold Start Users

As the number and variety of services in the network increasing,in order to help users find the services he needs from a large number of services,service recommendation technology emerges as the times require.Most of the traditional service recommendation methods find neighbors of the target user based on the users service using records and recommend the services which are similar to the services neighbors used.Other methods recommend the services which are similar to the services the target user used.However,these.methods can not recommend services for the cold start users who have no service using records.As a result,the cold start user problem has become one of the most urgent problems to be solved in the service recommendation field.A part of the existing algorithms to solve the cold start user problem require users to fill out the questionnaire in advance to obtain their preferences which increases the using complexity.The other part of the algorithms look for neighbors based on the registration information to predict the preferences of the cold start user,but these methods do not consider that different attributes of registration information have different influence on users'preference,which results in a decrease in the accuracy of recommendation.Therefore,for the shortcomings of the existing research work to solve the cold start user problem,this paper focuses on the research on the personalized service recommendation mechanism for cold start users.The main contributions of the paper include the following aspects:(1)An user similarity model based on registration information is established.The linear regression algorithm is used to calculate the influence weights of the registration information attributes to the users' preferences and it provides the basis for finding neighbors who have similar preferences to the cold start users.(2)A service functionality oriented personalized recommendation algorithm for cold start users is proposed.This algorithm uses the user similarity model to find the neighbors and predict the cold start user's function preferences by the neighbors' preferences.Then,all the services in the recommendation system are clustered based on their functional characteristics.Finally,calculate the matching degree between the functional characteristics of each service cluster and the cold start user's preference.and the services which have highest matching degrees are recommended.(3)Having the services recommended in the previous step as the candidate set,a service QoS oriented personalized recommendation algorithm for cold start users is proposed.This algorithm also uses the user similarity model to find the neighbors and predict the cold start user's QoS preferences according to the neighbors' preferences.Then,predict the cold user's QoS experience when he use the services in the candidate set according to his context information.Finally,recommend services to the cold start user based on how the service QoS meet the user's QoS preference.In summary,this paper deeply studies the mechanism of personalized service recommendation for cold start users.We build an user similarity model based on registration information,and propose a functionality-oriented personalized recommendation method and a QoS-oriented personalized recommendation method for cold start users.The simulation and system implementation have shown the feasibility and effectiveness of the theoretical results of this paper.