Research On Information Core Extraction Problems In Recommender Systems Based On Evolutionary Computation Of Xidian University

Recommender systems are effective tools to deal with the information explosion problem.The information core of a recommender system is a group of core users who carry reliable,objective and helpful information for recommendations.The information core extraction problem of recommender systems is to find the group of core users who can best represent all users of the system.The research shows that the core users can get satisfactory recommendation results and improve the efficiency of online recommendations when they replace all users of the system in the recommendation process.The information core extraction problem is to study the relationship of users in recommender systems.The deep understanding of user relationship is also helpful to improve the performance of recommendation algorithms.Therefore,the information core extraction problem is of great significance in both theoretical research and practical application.In this paper,based on the existing information core extraction algorithms,the multi-objective,intra-domain and cross-domain multi-tasking optimization problems in recommender systems information core extraction problems are studied by using evolutionary algorithms:?1?An information core extraction algorithm for recommender systems based on multi-objective evolutionary algorithms is proposed.The existing algorithms for information core extraction problems are mainly divided into two categories:greedy algorithms and evolutionary algorithms with accuracy as the objective.Accuracy,coverage and diversity are all important indicators to evaluate the recommendation performance.Existing information core extraction algorithms only focus on accuracy,ignoring other evaluation indicators.In order to meet the multi-objective optimization requirements of recommender systems,two objective functions related to the accuracy and coverage of recommendation results are proposed,and a model of multi-objective information core extraction problems with constraints on the scale of information core is established in this thesis.Experiments are conducted to verify the conflict between the two objective functions.Then,based on multi-objective evolutionary algorithms,a new optimization algorithm,abbreviated as MOEA-IC,is proposed to deal with the multi-objective information core extraction problems.The effectiveness of MOEA-IC is validated by experiments on different open datasets.?2?An intra-domain multi-tasking information core extraction algorithm based on multifactorial evolutionary algorithms for recommender systems is proposed.The information core extraction problem is a complex application problem.According to different selection rates,a recommender system can extract information cores of different scales.Extracting information cores of different scales from the same recommender system at the same time is an intra-domain multi-tasking problem.In practical applications,intra-domain multi-tasking requirements exist in recommender systems.In order to study the intra-domain multi-tasking information core extraction problems,in this thesis,we model the intra-domain multi-tasking information core extraction problem,design a unified coding and task-specific decoding method for intra-domain multi-tasking problems,and propose an intra-domain multi-tasking information core extraction algorithm based on multifactorial evolutionary algorithm,abbreviated as MFEA-IC_IN.The effectiveness and advantages of MFEA-IC_IN are verified by experiments.?3?A cross-domain multi-tasking recommender systems information core extraction algorithm based on multifactorial evolutionary algorithm is proposed.There are intra-domain multitasking problems and cross-domain multi-tasking problems in information core extraction problems of recommender systems.Different recommender systems exist different tasks related to information core extraction problems,and the types of tasks may be the same or different.In this thesis,two kinds of cross-domain multi-tasking information core extraction problems are mainly studied,which are completing information core extraction tasks of different systems and completing information core extraction tasks and other tasks of different systems.A unified encoding and task-specific decoding method for cross-domain multi-tasking problems are designed.Based on multifactorial evolutionary algorithm,a cross-domain multi-tasking information core extraction algorithm,abbreviated as MFEA-IC_CR,is proposed.The effectiveness of MFEA-IC_CR is validated on open data sets.The experimental results show that the proposed method has certain guiding significance for solving and optimizing cross-domain multi-tasking information core extraction problems in recommender systems.