D~3MOPSO:一种基于用户偏好的元搜索排序聚合演化方法

发布时间：2019-07-15 17:02

【摘要】：随着网络数据的爆发式增长和用户需求的多元化发展,现有元搜索排序聚合方法在精度和性能上面临着巨大挑战.以满足用户的多重需求和个性化偏好为目标,提出了一种新的元搜索排序聚合算法.通过重新定义多目标粒子群优化算法(multi-objective particle swarm optimization,MOPSO)中粒子的属性,调整速度变化因子,改进种群初始化和演化机制,设计新的存档与更新策略以及引导微粒选择策略,提出了一个基于支配分解的离散多目标优化(D~3MOPSO)算法,使其能根据用户的质量需求偏好在大规模离散解空间中快速准确地找出最优解集.在多个数据集上的实验结果表明:当数据规模较小时,D~3MOPSO算法的精度和性能接近机器学习排序聚合方法;在大规模数据环境下,其精度和性能优于机器学习方法以及同类多目标优化方法.
[Abstract]:With the explosive growth of network data and the diversified development of user demand, the existing meta-search sort aggregation methods are facing great challenges in accuracy and performance. In order to meet the multiple needs and personalized preferences of users, a new meta-search sort aggregation algorithm is proposed. By redefining the properties of particles in multiobjective particle swarm optimization (multi-objective particle swarm optimization,MOPSO), adjusting the speed change factor, improving the population initialization and evolution mechanism, designing a new archiving and updating strategy and guiding particle selection strategy, a discrete multiobjective optimization (D~3MOPSO) algorithm based on dominant decomposition is proposed. So that it can find the optimal solution set quickly and accurately in the large-scale discrete solution space according to the user's quality demand preference. The experimental results on multiple data sets show that the accuracy and performance of D~3MOPSO algorithm is close to that of machine learning sort aggregation method when the data scale is small, and its accuracy and performance are better than that of machine learning method and similar multi-objective optimization method in large-scale data environment.
【作者单位】：武汉轻工大学数学与计算机学院;武汉大学计算机学院;
【基金】：国家自然科学基金项目(61502350) 湖北省教育厅科研计划项目(Q-20161702)~~
【分类号】：TP18

【参考文献】