多模态数据驱动的服务需求预测方法研究  

Research on Multimodal Data Driven Service Demand Prediction

作　　者：海燕 宋宗珀 刘志中 丰凯 HAI Yan;SONG Zongpo;LIU Zhizhong;FENG Kai(Department of Information Engineering,North China University of Water Resources and Electric Power,Zhengzhou 450046,China;Department of Computer and Control Engineering,Yantai University,Yantai 264005,China;Department of Computer Science and Technology,Henan Polytechnic University,Jiaozuo 464000,China)

机构地区：[1]华北水利水电大学信息工程学院,河南郑州450046 [2]烟台大学计算机与控制工程学院,山东烟台264005 [3]河南理工大学计算机科学与技术学院,河南焦作464000

出　　处：《郑州大学学报（理学版）》2023年第2期70-78,共9页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金项目(61872126,61772159);山东省自然科学基金重点项目(ZR2020KF019);河南省教育厅重点研发项目(20A520014)。

摘　　要：数据驱动的主动服务推荐已成为实现智能服务、提升用户体验的重要技术,如何精确预测用户的服务需求成为当前重要问题之一。针对这个问题,提出了一种基于软注意力与多模态机器学习(SAMML)的服务需求预测方法。该方法首先从文本数据和图像数据提取特征向量并实现多模态数据特征融合;然后应用软注意力机制处理融合后的特征数据,并将结果输入门控制循环单元(GRU)网络,从而更好地学习用户的服务兴趣;最后,基于用户特征与服务数据训练SAMML模型,并实现用户的服务需求精确预测。基于天池大数据众智平台提供的数据集进行了验证实验,SAMML模型的评估指标MAE、MSE、R 2分别比对比模型提高了0.0218、0.0263、0.0310。Data-driven active service recommendation has become an important technology for realizing intelligent services and improving user experience.How to accurately predict the service needs of users has become an important problem.To solve this problem,a service demand prediction method based on soft attention and multimodal machine learning(SAMML)was proposed.The method firstly extracted feature vectors from text and image data and realized feature fusion of multimodal data.Then,soft attention machine was applied to process the fused feature data,and the results were input into the gated recurrent unit network,so as to learn the user′s service interest better.Finally,the SAMML model was trained,based on user characteristics and service data,and the user′s service demand was accurately predicted.A large number of verification experiments were carried out based on the data set provided by Tianchi big data intelligence platform,and the SAMML model was 0.0218,0.0263 and 0.0310 higher than the comparison model on MAE,MSE and R 2,respectively.

关 键 词：软注意力机制 多模态机器学习 特征融合 门控循环单元 服务需求预测 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]