Research On Influence Maximization Algorithm Based On Genetic Local Search Optimization And User Interaction Representation

Owing to the prosperity of technological innovation,things in intricate systems can be naturally linked into intricate networks after being simplified.Social network as an indispensable branch of intricate network,it connects interrelated objects to form a network of social services functions.The massive information contained in social network provides resources for social data analysis.As the research focus of social network data mining,the goal of influence maximization is to design a reasonable evaluation mechanism to select some users with outstanding influence ability from the intricate network,so that the information can spread to the greatest extent.The Monte Carlo simulation based on greedy can guarantee the approximate optimal solution in theory,but due to the so many simulations,it’s time-consuming and out of the question to be extended to the large-scale networks.Some algorithms developed of heuristics may reduce time consumption,but in most cases they cannot achieve satisfactory solutions.Therefore,how to balance the solution performance and time consumption is still a challenge for influence maximization.Secondly,networks are not immutable.Limited to the fixed network structure and hypothetical diffusion model,some existing algorithms cannot be coordinated with real application scenarios.So as to enhance the reliability and make the algorithm better in universality,this article utilized two different methods of examining how to maximize influence.The specific work is as follows:（1）In the context of genetic algorithm to put forward a local search tactic.That makes the improvements in the problem about influential nodes gathering caused the restriction of influence diffusion.Firstly,uses the first-order inclusion-exclusion activation function to select high-quality nodes to generate initial population and construct key node sets.Then,in the evolutionary process,local search strategy is adopted,including the replacement of individual mutation genes by seed set and node similarity calculation combined with key node set in mutation operation,and the optimization of sequential crossover by node second-order degree in crossover.Finally,use a fixed model to assess the scope of influence.Experiment in six data sets indicates the capability of raised algorithm is close to CELF,and it can balance the performance and time consumption well.（2）In order to break away the constraints of fixed network and diffusion model,under the user interaction representation to raise an algorithm.Firstly,user interaction cascades are used to construct user context pairs,and Skip-gram model is used to learn user feature vectors.Then the candidate seed set is selected according to the active user,and the user feature vector is used to calculate the interactive relationship degree to select the best seed set.Finally,uses the interaction cascade to obtain the influence appraisal.The experiment in two authentic networks confirm the raised algorithm produces a better user sets which can bring about the scope of the interaction widely.