用户行为驱动偏好代理模型辅助的交互式个性化进化搜索算法  

Preference surrogate-assisted interactive personalized evolutionarysearch algorithm based on user behaviors

作　　者：暴琳 齐亮[1] 吴杨 陈佳佳 BAO Lin;QI Liang;WU Yang;CHEN Jiajia(School of Automation,Jiangsu University of Science and Technology,Zhenjiang 212100,China)

机构地区：[1]江苏科技大学自动化学院,镇江212100

出　　处：《江苏科技大学学报（自然科学版）》2024年第2期65-72,共8页Journal of Jiangsu University of Science and Technology:Natural Science Edition

基　　金：国家自然科学基金项目(51875270);广东省数字信号与图像处理技术重点实验室开放基金项目(2021GDDSIPL-06)。

摘　　要：随着互联网用户数量迅猛增长,产生了大量用户生成内容,出现了信息过载现象.考虑用户生成数据,建立用户兴趣偏好模型,同时联合交互式进化计算,提出用户行为驱动偏好代理模型辅助的交互式个性化进化搜索算法,帮助用户从海量搜索空间中搜寻符合用户潜在需求和兴趣偏好的项目或内容.利用用户交互行为、评分数据和项目类别信息,构建基于受限玻尔兹曼机的用户偏好感知模型,抽取用户偏好特征;从进化优化的角度,设计基于用户偏好的代理模型及其进化策略,生成含用户偏好的新个体,并预测进化个体适应值,引导进化优化过程;根据新增用户生成内容和模型管理机制,动态更新各模型,及时跟踪用户偏好,顺利完成个性化进化搜索.通过大量真实世界数据集的实验,验证了所提算法处理动态个性化搜索和推荐任务的可行性及有效性.With the rapid growth of the number of users on internet,a lot of user-generated contents(UGCs)has been generated,and there has been information overload.This paper makes full use of UGCs to build a user interest preference model,and proposes a preference surrogate-assisted interactive personalized evolutionary search algorithm based on user behaviors.Combing the interactive evolutionary computing,it helps users search for the items that meet their potential needs and interest preferences from a massive search space.By using interaction behaviors,ratings and item category information,a user preference perception model based on restricted Boltzmann machine is constructed to extract the user preference features.From the perspective of evolutionary optimization,a surrogate model based on the user preference and its evolutionary strategies is designed to generate new individuals with the user preference,and predict the fitness value of new individuals to guide the evolutionary optimization process.Meanwhile,according to new UGCs and model management mechanism,these models are dynamically updated to timely track the user preference for the personalized evolutionary search.Through a large number of experiments in the real-world datasets,the feasibility and effectiveness of the proposed algorithm are verified in dynamic personalized search and recommendation tasks.

关 键 词：交互式进化计算 代理模型 用户生成内容 受限玻尔兹曼机 个性化搜索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]