面向常识的时间推理

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 Oriented Temporal Reasoning

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.