Research On Key Technologies Of High-quality Information Dissemination In Social Networks

With the development of communication technologies and the popularization of smart devices,social networks are becoming an increasingly important platform for users to consume and share information,with many important applications in areas like daily entertainment,viral marketing and social ideology etc..Among these application scenarios,how to ensure the highquality dissemination of information in social networks is of great importance to promote users’experience and improve the operation of networks.In practice,the high-quality information dissemination is often embodied in three aspects1,i.e.,faster speed,broader coverage and higher popularity.To elaborate,(1)the information should be transmitted to network users within a short time(fast speed);(2)the information should be transmitted to a broad range of network users(broad coverage);(3)faced with multiple information competition,the target information should be received by a high proportion of users(high popularity).With this motivation,focused on the high-quality information dissemination in social networks,this dissertation will conduct researches from the above three dimensions to guarantee the dissemination quality and solve practical problems faced by the promotion of dissemination quality in each dimension,in order to reach the goal of faster speed,broader coverage and higher popularity and finally provide theoretical guidance and technical support for the efficient operation of practical social networks.Specifically,the research content,challenges and contributions are summarized as follows.First,regarding the dissemination speed,this dissertation studies how to analyze and improve the time for information to reach all network users.In practical network operation,social networks are often observed to be evolving over time:new users constantly join in and establish new connections with previous network users.On the one hand,the evolution of networks brings about an increasingly larger network scale and more complex network structure.Thus,the information needs to reach more users and handle more complex user connections during its dissemination,often resulting in longer dissemination time.On the other hand,numerous existing researches indicate that the evolution of networks also brings about distinctive structural features,such as shrinking diameter and edge densification.To some extent,the densifying structure increases the probability for the information to be known and requested by users,which could help improve the dissemination time if properly leveraged.Thus,this dissertation first conducts quantitative analyses on the densifying structure of evolving networks,and derives the upper bound and lower bound of the multi-hop neighborhood size of arbitrary users,which turns out to be a substantially larger value than non-evolving networks.Leveraging this property,this dissertation proposes an efficient transmission mechanism based on neighborhood searching which provides users with higher probability to find an information holder for request.Meanwhile,the algorithm imposes necessary restrictions on neighborhood searching to prevent users from receiving too many simultaneous requests which would make users wait for a long time.Finally,further theoretical analyses find that the proposed algorithmachieves a sub-linear dissemination time of O(?),by properly leveraging the evolution properties,where n is the number of users,i is the searching range,D is the network diameter,γ and τ are parameters of users’ degree.Finally,experiments on real networks indicate that the proposed algorithm could achieve 1.38 times faster dissemination speed than traditional algorithms.Second,regarding the dissemination coverage,this dissertation studies how to optimize the information coverage in social networks.To obtain a large information coverage,people often distribute some welfare like samples and coupons(called "seeding")to attract influential users to become information sources(called "seeds")to propagate information.However,in real social networks,users’ information in the social network may not be always valid.For example,the registered phone number may not be frequently used,and the address information may not be updated in time.Therefore,it is hard to distribute welfare to any user in the social network as one wishes.Instead,only a small subset of users is initially accessible for seed selection.Within a small set of users,influential users are even scarcer,which could only bring about limited information coverage.To break the access limitation and promote the information coverage,this dissertation proposes a two-stage seeding framework based on the friendship paradox phenomenon(i.e.,the degree of neighbors is larger than a user himself in expectation),where initially accessible users are first seeded to reach their neighbors,and then the welfare is further allocated to the neighborhood to expand the information coverage.Following this framework,this dissertation first designs a convergent two-stage coordinate descent algorithm based on the traditional non-adaptive idea.Moreover,to deal with the uncertainty in the seeding and diffusion process,a series of adaptive algorithms are proposed with constant approximation ratios,where the best one achieves 0.468 times of the optimal coverage in theory,better than the state-of-the-art algorithm.Finally,extensive experiments are carried out on large scale real networks and the results indicate that the proposed algorithms achieve a larger information coverage than baseline algorithms,with a maximal elevation of 116%.Third,regarding the dissemination popularity,this dissertation studies how to promote the proportion of users occupied by the target information during information competition.As an ideal channel for information distribution,social networks have attracted large amounts of publishers to distribute their information.Faced with the intense competition,how to capture the competing dynamics accurately and reveal key impact factors are of great instructive importance for promoting the popularity of the target information(i.e.,the proportion of users that adopt the information).However,in practical network operation,the depiction of competing process is subject to challenges from both the network and users.To elaborate,it is often hard to derive the accurate social network topology that information propagation relies on,and users could receive too much information that exceeds ones’ handling capability(i.e."information overload"),leading to variable user behaviors.Accordingly,this dissertation adopts the network embedding technique that is often applied in link prediction to abstract user similarity,based on which we further recover possible network connections.Moreover,user’s behaviors before and after information overload are captured in mathematical expressions,to take account of the impact of information overload on the diffusion process.On this basis,this dissertation obtains the critical time of information overload which is closely related to the network topology,user characteristics etc..And explicit expressions are further derived to describe the dynamics of user proportion before and after overload.The results indicate that before information overload,the proportion of users could be elevated to 100%theoretically by enhancing the infection ability of information.After information overload,the proportion of users could be further improved by strengthening user’s ability of handling information and restraining the formation of user connections.Further experiments indicate that,by leveraging the revealed competing dynamics,the proportion of users could be improved by 98.2%.Based on the above content,this dissertation has conducted in-depth researches on three typical aspects of high-quality information dissemination(i.e.,speed,coverage and popularity),and has derived a series of theoretical and practical results,promoting the high-quality information dissemination in social networks.In the future work,this dissertation will investigate other aspects relating to the high-quality dissemination of information in practical applications,to further optimize the operation of social networks.For examples,how to guide the information to the interested users with high precision and how to avoid the distortion of information during its dissemination between users.