Local Community Structure Discovery Based On Evolutionary Algorithm And Its Application In The Recommender System

At present,with the increasing size of the network and the network is a dynamic or a increasing network,the complete information of the network is often difficult to obtain,such as web pages,research papers and Facebook users,to get the integrity community division information from these networks is not realistic,in addition,some nodes that play the key role often exist in the network,if the nodes are found,it is of great help to the analysis of the network structure,if all the nodes in the network are taken into account for network division,on the one hand it will cost a lot of time,on the other hand,it will get a lot of redundant information,which is not conducive to the network structure analysis.In this case,the local community detection is more meaningful,the local community structure can provide us with a lot of micro analysis,which can be an auxiliary analysis of macro analysis.However,the majority of local community detection is sensitive to the initial point,the choice of initial point may influence the final result,this thesis proposes the local community detection algorithm based on evolutionary algorithm,the first step is to find the core nodes in the network,and then to expand based on the core nodes,thus solving the shortcoming that an initial different position nodes can affect the community result.With the development of computer network technology in the real life,the importance of the network as the electronic and commercial transaction media is increasing.This has prompted the development of recommender systems.This paper mainly studies the local community detection problem based on evolutionary algorithms,and using the local community detection method based on the discrete particle swarm optimization for the recommendation system problem,the main works are summarized as follows:(1)We proposed a memetic algorithm for extracting the tightest social network community.Memetic Algorithm is an evolutionary algorithm,which is a combination of global search based on population and local heuristic search based on individual.In order to get the most intensive community structure,the algorithm makes a global search for the network connection diagram,which overcomes the shortcomings of the traditional discovery of community structure from a single node,at the same time,we design a heuristic algorithm for local search.Through the experimental comparison,the algorithm can get better community structure.(2)we put forward a multi-agent genetic algorithm for local community detection by extending the tightest nodes.Multi agents genetic algorithm is a swarm intelligence algorithm,the agent can act on the environment and change the environment,agents communicate with each other,which makes the multi agents be easier to collaborate to search,and the population toward the optimal direction of development,in order to get the most intensive nodes in the local area of the network,we call the most intensive nodes is the "core" of the network,we design a neighborhood crossover operator,the operator makes the algorithm have a better global search ability in the local range,which solves the random error of the initial node that will influence on the final result,after finding the "core",then to expand based on the "core",in the experiment,several traditional discovery algorithms are compared,the experimental results show that MAGA-LC has good performance.(3)Many practical recommendation systems use collaborative filtering techniques to make recommendation,but when the number of users is less with respect to the number of projects,the user score for the project is relatively small,at this time the collaborative filtering techniques can not applied to recommendation system,but using the network technology with the similar users in a community,the collaborative filtering recommendation based on the community can solve the above problems.A local community detection algorithm based on discrete particle swarm optimization is designed,which is denoted as DPSO-LC.In the experiment,the algorithm firstly carrys out the community detection experiments,the algorithm can get higher normalized mutual information index value,then makes the collaborative filtering recommendation experiment,and two collaborative filtering recommendation algorithm are compared.The experimental results show that our algorithm gets a better result.