面向常识的时间推理  

Commonsense Oriented Temporal Reasoning

作　　者：徐晋晖[1] 毛希平[1] 刘桂霞[2] 石纯一[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084 [2]吉林大学计算机科学系,长春130002

出　　处：《软件学报》2000年第6期809-815,共7页Journal of Software

基　　金：国家自然科学基金!(Nos.6 97730 2 6 ;6 97330 2 0 );教育部高等学校重点实验室访问学者基金!(吉林大学"符号计算与知识工程教育部重点

摘　　要：常识和时间推理是人工智能研究的两个主要课题 .Allen等人提出的时间推理缺少时间点、时区和时距的统一表示 ;过分考虑计算 ,缺少规则推理 ;求解算法难以应用于多 Agent环境并且没有考虑常识不一致性 .该文提出一种时间信息表示网络 ,分析了约束之间的推导规则 ,给出了常识时间问题的多 Agent合作满足弱路径一致性的求解方法 .该文的工作改进了 Meiri,Wetprasit和 Sattar等人的工作 ,为时间推理结合常识特性和适应多 Agent合作环境提供了可行的方案 .Commonsense and temporal reasoning are two main research topics in artificial intelligence. Temporal reasoning proposed by Allen and others lacks a unified representative network about time point, interval and duration, overemphasizes computation while lacking rule inference, is difficult to apply in multiagent setting, and does not consider inconsistency in common sense. In this paper, the authors present a unified representative network about time information, analyze the inference rules among constraints, and propose the multiagent cooperative method satisfying weak path consistency to solve commonsense temporal task. The work of this paper improves the work of Meiri as well as that of Wetprasit and Sattar, provides viable scheme for temporal reasoning to combine commonsense property and to fit the cooperative multiagent setting.

关 键 词：常识 时间推理 时间点 时区 时距 定量约束 定性约束 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]