Information Core Optimization Based On Evolutionary And Network Propagation Algorithm In Recommendation System

With the advent of the era of big data,the massive growth of information has forced people to face the dilemma of information overload.As an effective technical means to solve information overload,the recommendation system has attracted the attention of many scholars.As one of the most commonly used algorithms in the field of recommendation system,collaborative filtering algorithm gradually exposes some inherent problems with the rapid growth of data.Among them,the issue of scalability is one of the main obstacles hindering its development.In order to alleviate this problem,scholars have proposed a variety of methods,and the recommendation method based on information core is one of the novel methods.At present,the information core is built mainly through some heuristic methods,and there is a certain loss in the recommendation quality compared with the traditional collaborative filtering algorithm.To solve this problem,this paper proposes an information core optimization method based on evolutionary and network propagation algorithms,which are used to reduce online recommendation time to alleviate the scalability problem and improve recommendation quality.(1)An information core optimization method based on multi-subpopulation evolutionary algorithm is proposed.Firstly,three constraints are proposed according to the degree of the user to divide the population into three sub-populations.Then,each constraint and information core are encoded in the individual of the population,and a better degree threshold is automatically searched through evolutionary iteration.Finally,the elite retention strategy is used to retain the offspring with strong competitive ability to form the next generation population.Under the condition that the total number of population individuals remains unchanged,the size of the sub-population with strong competitive ability will gradually increase,while the size of the sub-population with weak competitive ability will gradually decrease or even be completely eliminated.The experimental results show that the method can effectively find the information core,and thus obtain better recommendation quality than other comparison methods.In addition,the time for online recommendations is greatly reduced.(2)A virtual information core optimization method based on evolutionary and clustering algorithm is proposed,which aims to improve the utilization rate of users' information in the system and construct a virtual core user with more information.Firstly,a simple "average method" is proposed to merge the information of multiple users to form a virtual core user.Then,the problem of searching virtual information core is modeled as a combinatorial optimization problem,and was solved by the proposed evolutionary algorithm.Finally,to make offline optimization more efficient,we propose two improved strategies.One is to introduce a simple similarity measure based on dimensionality reduction and clustering to save time of similarity calculation,and the other was to use dimensionality reduction and clustering to construct a smaller scale of training set and validation set.The experimental results show that the virtual information core optimization method based on evolutionary and clustering algorithm further improves the recommendation quality,and its online recommendation time is shorter than the comparison methods,and has stronger ability to alleviate scalability problems.(3)A virtual information core optimization method based on clustering and network propagation algorithm is proposed,which aims to reduce the offline optimization time of virtual information core,and further utilize the user's information in the system to build the virtual information core with better performance.Firstly,in order to reduce the time of repeated clustering and obtain better clustering effect,we use the t-SNE algorithm to get the user's low-dimensional data.Then,the user's low-dimensional data is repeatedly clustered,and the cluster center of each cluster is calculated in the user's high-dimensional data,thereby obtaining the virtual users.Finally,the improved network propagation algorithm is used to select some virtual core users from the virtual users to form the virtual information core.The experimental results show that the virtual information core constructed by this method further improves the recommendation quality and greatly reduces the offline optimization time of the virtual information core.