AI设计下的文本视觉问答技术  

Text-based Visual Question Answering with AI Design

作　　者：晋赞霞 覃京燕[1] 殷绪成[1,2] JIN Zan-xia;QIN Jing-yan;YIN Xu-cheng(University of Science and Technology Beijing,Beijing 100083,China;Shunde Graduate School,University of Science and Technology Beijing,Foshan 528399,China)

机构地区：[1]北京科技大学,北京100083 [2]北京科技大学顺德研究生院,佛山528399

出　　处：《包装工程》2021年第6期7-12,共6页Packaging Engineering

基　　金：长江学者奖励项目(FRF-TP-18-010C1);国家重大专项课题(2018YFB0704301);北科大顺德研究生项目(BK19AE011)。

摘　　要：目的分析基于AI设计的文本视觉问答模型的有效性,旨在利用AI设计更好地指导当前AI模型的构建,提升模型效果和用户体验。方法以传统文本视觉问答框架为基础,结合AI设计改进当前模型。具体包括加强基于场景设计原则的关系挖掘,根据不同理解层次需求的答案关键词预测,并对模型被投入应用所将面临的问题的分析。结果基于AI设计完善模型可进一步提升模型效果;同时,通过AI设计对不同年龄认知差异的建模可指导回复生成,提升整体用户体验。结论通过理论分析和实验对比,可以得出AI设计是AI技术投入到应用的一个重要步骤。基于AI设计对模型进行重构,可提高当前模型的效果,解决AI技术落地中将面临的用户体验问题,满足不同人群的需求。It analyzes the effectiveness of the text visual question answering model based on AI design,aiming to better guide the construction of current artificial intelligence models with AI design,and improve model performance and user experience.It is based on the traditional text visual question answering framework,and the current model can be improved by combining AI design.Specifically,it includes strengthening relationship mining based on the principles of scenario design,predicting answer keywords according to the needs of different levels of understanding,and analyzing the problems that the model will face when it is put into application.Modifying the model based on AI design can further improve the performance of the model,and modeling the cognitive differences of different ages through AI design to guide response generation can improve the overall user experience.Through theoretical analysis and experimental comparison,it can be concluded that AI design is an important step in the application of AI technology.Reconstructing the model based on AI design can improve the performance of the current model,solve the user experience problems that will be faced in the implementation of AI technology,and meet the needs of different groups of people.

关 键 词：AI设计 AI 文本视觉问答 认知差异 

分 类 号：TB472[一般工业技术—工业设计]