Study Of Information Propagation And Immunization Strategy On Complex Networks

This dissertation firstly introduces the research status of the complex network theory in information dissemination and immunity, the theoretical basis of complex networks and common propagation model and basic immunization strategy, then explores the propagation model for topic of online social network, the information dissemination on weighted networks with two pieces of news and immune strategy for suppressing harmful information, and gets some innovative research results. The specific research work and conclusions are as follows:1. Considering the characteristics of topic’s diffusion in online social networks, we propose a topic propagation model which is based on the online social networks. Individual differences among users described by random variable μt,, users’memory effect about information decribed by function P(k), inherent properties of information described by parameter λ and the character of topic propagation described by exponential decay function e-βt are fully taken into account in this model. Then the impact of the model’s parameters on spreading is studied on both BA network and the data sets of Facebook. The study finds that, when the mean value of μ, that every user accepts the topic is the same and equals to 0.5, different probability distributions of μi will affect the scope of spreading. When the average degree of network is larger,μi which obeys to normal distribution will have greater topic diffusion scale than which obeys to uniform distribution. While the network’s average degree and the topic’s inherent infection rates are all small,μt which obeys these two probability distributions will bring no obvious influence on the scope of spreading. According to the relationship between the frequency of a user reading a same topic and participation probability, we structure function P(k). Changing the value of P(k) function’s peak and balance will affect topics’propagation, and the influence brought by the balance value is greater than the peak. When the P(k) function’s peak and balance value change simultaneously, the influence of topic spreading scale brought by them can be compensated. The study also finds that increasing β which is the regulator for the spread decay rate, topic’s infection rate will decay with time rapidly, and it will shorten the spreading time of the topic in the network, and finally result in lower peak and small dissemination scale. When β and P(k) are working together, the spreading scale bought by β is significantly greater than P(k).2. The dissemination with two pieces of competitive information, the one has dominant ability to communicate called S1 and the other is vulnerable in spreading called S2, are studied in weighted networks. We focus on studying the propagation characteristics of S2 from the two aspects of its initial numbers and the network’s structure by simulating in GBBV weighted networks firstly and verifying the conclusions in five real datasets secondly. Study finds that when the network’s average degree is about larger than 4, the percentage of the population spread S2 will raise into a peak then fall down into a steady state with the increasing of time. And the larger the average degree is, the shorter the peak arrival time uses, and as well as the faster the spreading stable reaches. From the perspective of network’s structure, we find that average degree has largest effect on the propagation scale of S2. When the average degree is about less than 8, S2’s spreading scale can be greater than S1. While <k>≈4, the most conducive to the spread of S2. The above conclusions are not sensitive with the network’s scale, clustering coefficient and average edge weight. The study also finds that average edge weight of network can inhibit the whole network’s disseminative range, so that the final number of people who don’t spread information increases, and it can also play a role in accelerating the spreading speed. From the perspective of the initial spreaders’number, although the increasing number of initial S2 can augment the propagation scale’s peak of S2, as well as the peak of its propagation velocity, but it can’t make significantly impact to the scale of S2 in the spreading steady state, especially when the average degree of the network is relatively large, the increase of S20^ can hardly affect the steady size of So3. Studies Immunization strategy. We present a cluster immunization strategy and study its immune effect on scale-free network with tunable clustering in a modified classic rumor propagation model. Study shows that when the average degree of networks is small, the connection of networks is sparse, effect of cluster immunization becomes better with increasing of network clustering coefficient. While the network is connected more closely, the average degree of the networks is larger, the cluster immunization will be failure. Then several immunization strategies including target immunization, betweenness immunization, closeness immunization and cluster immunization are compared in the sparse connected network. It shows that betweenness immunization is always the best regardless of network clustering. As a network clustering coefficient is relatively great, effect of cluster immunization is better than that of closeness immunization and close to target immunization. With further increasing network clustering coefficient, cluster immunization exceeds target immunization and approaches to betweenness immunization. So the conclusion is that cluster immunization is suitable for scale-free networks with large clustering coefficient and sparse connection.