Research On Model Of Viral Marketing Based On Trust Network

Viral Marketing, also named as Word of Mouth, which is previously referred to thecommercial marketing activity of distributing free samples to potential consumers. Those whohave used the new product would probably recommend it to their relatives, friends andcolleagues. And these new users in turn would transmit the product in exactly the same wayas those who receive free samples did. Therefore, the new product can be adopted by a largenumber of consumers. Viral Marketing based on online trust network transmits therecommendation of product by taking advantage of trust network. It takes the value of trust asthe influential ability between two nodes. It is also must be much more efficient than thetraditional way of Word of Mouth. Because, first, the real-time of the network, second, thehuge number of nodes of the online trust network, and last but not least, it is both natural andconvenient for users to send recommendation links to their neighbored nodes online.The study of Viral Marketing based on trust network is mainly three folds. First, themodeling of trust network, that is to establish the trust relationship between nodes and makefor the Viral Marketing. Second, the establishing of the network diffusion model. Thenetwork diffusion model describes how the recommendation was diffused on the trustnetwork. Third, designing of the selecting algorithm. That is to decide which nodes to chooseas the initial marketing objective so that the maximized number of adoption is gained. Thisproblem has already been proved to be NP hard.Firstly, this paper established the model of trust network by proposing a new trustevaluation method with dynamic punishment factor. The new model well follows the code ofchange of the value of trust that is hard to increase while easy to fall and resistant to maliciousattack meanwhile. Experiment also proved that the new model performs well when it comesto malicious attack, specifically loop attack. Secondly, this paper take a deep study of networkdiffusion model and proposed a cellular automaton based diffusion model (CAND). The newmodel was proved to be equivalent to the Linear Threshold model but much more efficientthan it in experiment. As for the selecting algorithm, this paper proposed an intelligentalgorithm based selecting method in which the Standard Genetic Algorithm, Particle SwarmAlgorithm, Differential Evolution as well as the Co-evolutionary Algorithms were used. Anintelligent algorithm interface was developed in JAVA to connect the NetLogo programmeddiffusion model. Experiment shows that in comparing with the newest selecting algorithms,Particle Swarm Algorithm performs best while keeping the running time in a relatively lowlevel which is also an indication of its scalability.