Trust Evaluation Model In Network Environment And Its Application In Recommendation System

The rapid development of network application has put forward new technical requirements for the corresponding network security mechanism.Trust can provide a relatively flexible security measurement mechanism in a complex network environment by analyzing and evaluating the potential trust information between various network nodes,which has become a hot spot in the field of network security research and application.Trust evaluation is the basis of trust technology research.However,due to the subjectivity and fuzziness of trust,how to build a high-precision and low-load trust evaluation model is still an important and challenging research topic.At the same time,the popularity of various online social networks has changed the way users access information.Users will not only get information based on their personal interests or hobbies,but also the recommendations of other users who have social relations with them will influence users' choices.The social recommendation system integrates the social attribute factors that influence the user's behavior preference with the collaborative filtering algorithm.It effectively improves the accuracy of the recommendation,and makes the recommendation system more in line with the social characteristics of human beings.Trust relationship is the most widely used social relationship in the current social recommendation system.How to apply user trust relationship to the recommendation system is an important topic in the social recommendation system.This paper studies the trust relationship in the network environment.It proposes a new trust evaluation model based on artificial intelligence.In addition,it tries to integrate the user's attribute preference and trust preference to make a comprehensive social recommendation.The main work is as follows:(1)the traditional trust model has poor performance in large networks because it relies on the propagation and iteration of trust in the network.This paper proposes a trust evaluation model based on machine learning.However,when a supervised neural network algorithm is used to build a trust evaluation model,there will be problems such as insufficient label data and unbalanced misclassification cost.In order to solve these two problems,a cost sensitive semi-supervised learning method is used for training.This algorithm is verified and tested on several data sets containinguser trust data.The experimental results show that this model can better identify the distrusted agent while ensuring the accuracy of the model.(2)In view of the data sparse,cold start problems of traditional recommendation algorithm,this paper proposes a social recommendation algorithm that integrates users' attribute preference and trust preference: first,this algorithm factorizes the rating matrix to get user latent feature vector and item latent feature vector,then uses the user latent feature vector to calculate the similarity of preferences between users.In addition,in order to solve the common problem of cold start in recommendation system,this algorithm introduces trust relationship.It fully explores the connection between rating information to measure the trust relationship with user preference.At last,dynamic weight is used to balance the recommendation weight among users.The algorithm is tested and compared with other recommendation algorithms on Film Trust and Epinions datasets.Experiments show it gains a higher accuracy of recommendation.