基于多目标优化的约束模式挖掘方法研究

Mining interesting patterns from massive data, namely constrained pattern mining, is a very important area of research in data mining. However, most of constrained pattern mining algorithms require manual selection of constraint parameters, and also the generation of mined patterns easily leads to the combinatorial explosion problem. If the problem of constrained pattern mining could be transformed as a complex combinatorial optimization problem, then the above problems can be solved well by using the multi-objective optimization techniques. Thus, this project intends to model constrained pattern mining problem from the perspective of multi-objective optimization, and carries out key technologies of pattern mining based on multi-objective optimization studies. Firstly, study how to transform the pattern mining problem into a multi-objective optimization problem, building the mapping relation between constrains and objectives. Then, study how to propose novel evolutionary computation method for solving the converted multi-objective optimization problem by exploiting the property of bound estimation in constrained pattern mining. Finally, to verify these findings in a number of practical pattern mining applications. In summary, by exploring the mechanism of multi-objective optimization, this project will propose effective solutions for the problem of pattern mining based on the multi-objective optimization. In this way, not only could the application areas of the multi-objective optimization be expanded but also the performance of the constrained pattern mining could be enhanced. Thus, the research of this project has important theoretical and practical value.

如何从海量数据中挖掘出有价值的模式，即约束模式挖掘，是数据挖掘中一个非常重要的研究领域。然而，大部分约束模式挖掘算法需要人工设置约束参数并且模式生成方式容易导致组合爆炸问题。若能够把约束模式挖掘问题转换成一个组合优化计算复杂问题，利用多目标优化技术可以很好地解决上述问题。为此，本项目拟从多目标优化的角度来对约束模式挖掘问题进行建模，开展基于多目标优化的模式挖掘关键技术研究。首先，研究如何把约束模式挖掘问题转换成一个多目标优化问题，确定约束和目标之间的映射关系。然后，研究如何融入约束模式挖掘的界估计特性，提出高效进化多目标优化方法，求解转换后的多目标优化问题。最后，在一些实际的模式挖掘应用中验证上述研究成果。本项目通过探索多目标优化机制，拟提出基于多目标优化的约束模式挖掘有效解决方案，对拓展进化多目标优化的应用领域和提升约束模式挖掘的实际效果，具有重要理论意义和应用价值。

约束模式挖掘研究是数据挖掘中一个非常重要的研究方向。然而，大部分约束模式挖掘算法需要人工设置约束参数并且模式生成方式容易导致组合爆炸问题。针对于上述问题，本项目从多目标优化的角度来对约束模式挖掘问题进行建模，开展基于多目标优化的模式挖掘关键技术研究。首先，研究如何把约束模式挖掘问题转换成一个多目标优化问题，确定约束和目标之间的映射关系。然后，研究如何融入约束模式挖掘的问题特性，提出高效进化多目标优化方法，求解转换后的多目标优化问题。最后，在一些实际的模式挖掘应用中验证上述研究成果。此外，还在智能优化基础算法及其应用等主要方面取得了创新性研究成果。在本项目的支持下，已发表学术论文18篇（IEEE/ACM Trans系列4篇，中科院一区期刊4篇），申请发明专利3项。本项目通过探索多目标优化机制，提出了基于多目标优化的约束模式挖掘有效解决方案，对拓展进化多目标优化的应用领域和提升约束模式挖掘的实际效果，具有重要理论意义和应用价值。