| The development of the Internet has brought about an upsurge in network data analysis,which has been widely used in many fields such as psychology,sociology,and computer science.Data-driven decision-making has become a magic weapon for success in various fields.From the perspective of enterprise development,the analysis and application of network data has become a "sharp weapon" for its long-term development;from the perspective of network users,it is possible to mine their preferences based on the information associated with users,so as to provide accurate information to network individuals.An important feature will appear in most networks,that is,the nodes in the network will form a community structure,forming multiple network communities,or clusters.For any two nodes,if they belong to the same community,the probability of connection between nodes is greater.Knowing the community structure can deepen the understanding of the network,so community detection is an important issue in network data analysis.Social network is the most common kind of network in daily life.We take it as the research background,mining the user characteristics in the network,realizing the scientific distribution of web page version,and helping e-commerce enterprises to make scientific decisions according to the click rate data of network users,so as to obtain more potential profits.Aiming at the research of network community detection,this thesis mainly analyzed the Popularity-adjusted Block Model,and established a network community structure detection based on modularity.By establishing the node popularity to capture the connection probability between network users,the likelihood module degree under this model is constructed.Considering the shortcomings of community detection by exhaustion method in terms of operation efficiency,two optimization algorithms are introduced to overcome them.Experimental results show that this Model can more flexibly model node popularity and show the relationships between network users than Degree-corrected Stochastic Block Model.Aiming at the analysis of page click rate,this thesis focuses on the Bayes A/B/C test problem under the multi-version delivery environment.A/B/C test is the multiversion form of A/B test.The traditional A/B test method can only evaluate the pros and cons of two web page versions,and can not measure the loss of decision mistakes.To this end,a Bayesian test model of user click rate under multi-version delivery environment is established.By considering Bernoulli modeling of click rate and combining with Beta conjugate priori,a posterior distribution of click rate is obtained.On this basis,by defining a loss function,a web page version selection model with a posterior expected loss as the criterion is further established.Experimental results show that this method can quickly obtain the optimal version of online delivery,and the error loss is small,the decision accuracy is high,and has a wide application prospect. |
