Research On Trust Management Mechanism For Social Networks

With the fast development of computer and Internet techniques,social networks have grad?ually come into people's daily life.The fabulous service content and intimate social links resources brought by social networks are increasingly changing the life style and social com?munication module of people.Compared with traditional off-line network,social networks have advantages over high speed,swift and convenient reach;on the one hand,more and more companies such as Huawei,Tencent,Alibaba,etc.,are more prone to establish their new productions or service to attract consumers' attention in time through social networks.On the other hand,users can also promptly find their desired items or services via social networks and share their personal experience including shopping,ratings and other events experiences through social platforms,e.g.,Weibo,Wechat and Twitter.these instant information is also useful to help other users acknowledge service in time.However,the facility brought by social networks is also accompanied with trust problem.Currently,consumers can obtain service information at home via social networks,but they also lose the oppor?tunity to directly bargain with service providers,which impose the awkward situations that the consumers select service via simply information instead of direct experience.In order to tackle this situation,trust management mechanism comes into effect,it combines service information,personal feedback,social relationship,fair supervision from a third trusted ad?ministration,etc.,and computes the final trust value.The high speed in computation and accuracy in evaluation help it win popularity,and now,trust management mechanism has been deployed in many famous e-commerce platforms such as ant financial,JD finance,micro payment,etc.However,there exist several drawbacks in current trust mechanism:Firstly,the judgement depended on trust value is easy but also dangerous,which provides a shortcut for the illegal.Some service providers attract other users via trust value accumulated in the previous trans?actions,then provide fake services to obtain excessive profit;Secondly,trust management mechanism excels in judging the trust degree of information while it is inefficient in estimat?ing its integrity,which leads to the difference between seller show and buyer show.Thirdly,current trust mechanism cannot conclude a synthetic verdict for a group of users because of their various preferences.Thus in this thesis,we mainly focus on several research related with trust management on social networks as follows:(1)We first consider the trust fraud problem about high trust value entities and propose a trusted trust evaluation scheme,called RHT(Recommendation from High Trust Value Entities),to evaluate the trust degree of services provided by high trust value entities.Our insight is motivated by a common reality:people are more inclined to select the services from high trust value entities.In fact,not all the high trust value entities should be trusted,sometimes,they also provide fake or low quality services to consumers,which may damage the properties of consumers.In addition,there also exist some malicious nodes who always perform attacks to interfere users' service selection.RHT contains two parts:One is trust nodes selection,the other is trust computation,where the former part includes similarity computation between nodes and resistance mechanism.When a high trust value entity provide a service,there exist other nodes who experience this service afford ratings or opinions.This information is crucial to a target node's selection on this service.We first compute the similarity between target node and those nodes who rate service to choose trusted nodes.During this process,we must consider the existence of malicious nodes and design a resistance mechanism to prevent malicious nodes from being chosen as trusted nodes.Second,we design a trust computation method to calculate the trust degree of service provided by a high trust value node.The experiment results show that the proposed scheme do efficiently withstand four common attacks.Additionally,the proposed scheme has higher accuracy on trust computation compared with current classical works.(2)we consider the service risk problem in the online social networks,which originates a fact of unequal information between service provider and consumers and consider two risk facets,i.e.,Sense Drop and Blue Joy.we propose a scheme,called RecRisk(Recommendation with Risk account),which can simultaneously guarantee the recommendation accuracy and minimize the risk facets.The proposed model first combines trust element with heat equation to accomplish service computing,in the same time,we design a modern portfolio theory(MPT)-based model to estimate the risk via considering the satisfaction and risks of service.Our designed scheme not only satisfies users' preference,but also control the risks as soon as possible.The sufficient experiments have also demonstrated that RecRisk attains higher recommendation accuracy and involves risky facets.(3)we mainly study the trust problem of a group of users whose preferences are distinctly various and propose a trusted service recommendation scheme for group setting.The proposed scheme contains two parts,the first part is multi-facet probabilistic graph model,which aims to model the users' implicit interaction to achieve service selection;the second part mainly accounts for trust degree,i.e.,each user in a group hopes his or her preference should be priorly satisfied.For more persuaded,we consider users' activeness and integrate resource allocation strategy with coalition game theory,and finally design a coalition game strategy based on users' activeness to rank the recommended services.The final results derived from our scheme are trusted because our strategy is built upon practical resource allocation principle.Our experimental results on the real data sets show that our scheme not only guarantee the trust of service computation,but also attain higher accuracy compared with other state-of-the-art approaches.