基于协同学习进化多目标优化的网络结构分析

After a deep analysis of the limitations brought by the single objective optimization based community detection methods for solving multi-resolution network community detection problem, the multi-resolution network community detection problem will be modeled as multiobjective optimization problems. The local learning tactics, the adaptive collaborative learning strategies and the parameter adaptive learning models will be established, and multi-resolution network community detection methods based on multiobjective optimization will be proposed. By analyzing the effect of different objectives on the performance of community detection in networks, the influence of different coding strategies on search space and the effect of different local search and adaptive collaborative learning strategies on the performance of the proposed algorithm will be studied. Network topology optimization methods based on multi-resolution community detection will be proposed to improve the robustness of networks, and personalized recommendation algorithms based on multi-resolution community detection are devised to improve recommendation accuracy and diversity.We will publish 12-15 papers in related leading journals and conferences, apply 2-3 patents, and bring up 5-6 graduate students.

深入分析基于单目标优化的网络社区检测方法在解决多分辨网络社区检测时的局限性，将多分辨网络社区检测建模为多目标优化问题，建立适用于多分辨网络社区检测的局部学习及其自适应协同策略，建立适用于多分辨网络社区检测的参数自适应学习模型，提出基于进化多目标优化的多分辨网络社区检测方法。分析不同目标函数对社区检测结果的影响，分析不同的编码策略对搜索空间和求解结果的影响，分析局部学习和自适应学习对算法性能的影响。进一步研究基于多分辨网络社区检测的网络拓扑结构优化方法，提高网络的鲁棒性；研究基于多分辨网络社区检测的个性化推荐算法，提高推荐算法的准确性和多样性。研究成果拟在本领域国际主流刊物和知名会议上发表论文12～15篇，申报国家发明专利2～3项，培养博士、硕士5～6人。

本课题严格按照研究计划进行，主要研究成果包括：深入分析基于单目标优化的网络社区检测方法在解决多分辨网络社区检测时的局限性，将多分辨网络社区检测建模为多目标优化问题，研究了基于进化多目标优化的多分辨网络社区检测方法。设计了相互冲突的多分辨目标函数，将网络社区发现建模为多目标优化问题，设计了基于网络拓扑的个体学习策略，提出了用于多分辨社区检测的进化多目标优化算法，算法求得的非支配解集合代表不同分辨率下的网络划分结果。针对个性化推荐中的推荐准确性和多样性两难问题，使用社区检测算法将有相似偏好的用户聚类到同一社区中，使用进化多目标算法平衡推荐准确性和推荐多样性。本课题共发表SCI检索论文17篇，授权国家专利20项，培养博士4人，硕士研究生10人，圆满完成了预期目标。