定性指标偏好感知进化优化及在个性化搜索中的应用

The qualitative optimization problems are universal but difficult to be solved due to the lack of explicitly defined indicators. The interactive evolutionary algorithm (IEA) is a good choice for such problems, however, it involves a user to directly assess those candidate solutions, which evidently brings large evaluation burden and causes user fatigue. Therefore, the existing IEAs are greatly hindered in solving complicated qualitative problems. To deal with these critical issues, based on some simple user-computer interactions, this project proposes a novel evolutionary algorithm framework with induced-assessment for optimizing qualitative problems and expects to obtain the following achievements: (1) revealing the relationships between the interactions and the assessments on the candidate solutions;(2)building the computable preference model of the user to serve as a fitness function based on the relationships; (3)developing a novel evolutionary optimization framework by articulating the preference model to assess the candidate solutions and enhance the optimization performance;(4)applying the proposed theory to a specific E-commerce system to achieve personalized and fast search. The project, at the cross edge of such disciplines as automation, computer, mathematics and management, is novel and challenging, and is urgently demanded by real applications. The achievements of this project can offer a powerful and new framework for solving the qualitative problems, enrich the theory of evolutionary optimization. Therefore, the project is of considerable significance in theory and application.

定性指标优化问题没有明确定义的目标函数，该问题普遍存在，难以求解。尽管采用交互式进化优化方法可以解决该问题，但是，需要用户根据偏好直接参与候选解性能的评价，使得评价负担急剧增加，导致用户疲劳，从而限制了该方法在复杂定性指标优化问题中的应用。本项目研究基于人-机交互行为间接感知用户偏好的定性指标进化优化理论，并应用于电商个性化搜索中。通过研究，拟揭示用户人-机交互行为与候选解性能评价之间的内在联系，建立可定量计算的用户偏好感知模型，设计基于感知模型评价候选解的高性能进化优化方法，并用于解决电商个性化快速搜索问题。本项目是自动化、计算机、数学，以及管理等学科有机交叉、新颖且富有挑战性性的研究方向，有非常明确的行业需求；产生的成果能够丰富进化优化理论，提高复杂定性指标优化问题求解方法的性能，因此，具有重要的理论意义和实际应用价值。

融合用户偏好的交互式进化优化是解决定性指标优化问题的可行方法，本项目针对传统交互式进化优化基于用户显式评价，导致用户评价疲劳或厌倦，使得算法难以搜索到满意解的问题，基于立项申请中拟开展的研究内容，进行了为期4年的深入全面研究，提出了基于人-机交互行为感知用户偏好的定性指标进化优化理论，揭示了用户人-机交互行为与候选解性能评价之间的内在联系，建立了可定量计算的用户偏好模型，设计了多种基于感知模型的高性能进化优化方法，并用于解决多类电商个性化快速搜索。基于上述研究成果，获江苏省科学技术二等奖1项、授权发明专利1项、计算机软件著作权登记6项，在被SCI或EI等检索的国际国内重要学术刊物上发表论文15篇，其中，SCI源刊论文7篇（含顶级期刊论文1篇，JCR1、2区期刊论文4篇），培养博士研究生1名、硕士研究生4名，圆满完成了项目的预期目标。研究成果进一步丰富了定性指标问题进化优化理论，为交互式进化优化在个性化搜索中的应用奠定基础，有力促进了人工智能领域进化优化和机器学习的融合。