面向活动类型挖掘和主题提取的人类行为模式研究

The human behavior patterns mined from trajectory data depict the relationship between human behavior, social and geographical environment, and help us to reveal the law of human behavior. The previous researches failed to effectively express the fundamental relationship in human activity patterns, because they neglected individual activity types, and overly simplified the relationship among individuals, spatiotemporal activities and behavior patterns. To address these problems, this project intends to adopt social media trajectory as data sources, develop modeling methods of human behavior patterns by identifying potential activity types and extracting human activity topics. Based on the knowledge graph theory and convolutional neural network, the human activity type discrimination method combining text content and trajectory position is designed to realize the accurate identification of human activity types in the trajectory. Using spatiotemporal activity flows instead of spatiotemporal activity hotspots and sparse simplification theory, a human activity topic model combining activity types and spatiotemporal locations is established to achieve accurate expressing of human behavioral patterns. In addition, the project will conduct research on the evaluation of human activity topic model considering of spatiotemporal heterogeneity. Those research results of this project are expected to theoretically reveal the essential relationship between individual spatiotemporal activities and human behavior patterns, provide methods for identifying human activities and for extracting activity patterns, and contribute to improving of understanding, prediction and intervening of human behavior.

从轨迹数据中挖掘隐含的人类行为模式，可以刻画人类行为与社会、地理环境之间的关系，揭示人类行为规律。当前研究忽视了个体活动类型信息，过度简化了个体、时空活动、行为模式间的关系，导致难以有效表达人类活动模式中的本质关系。本课题拟以社交媒体轨迹数据为数据源，通过识别潜在的活动类型、抽取人类活动主题，对人类行为模式的精细化建模展开研究。基于知识图谱和卷积神经网络，设计兼顾文本内容与轨迹位置的人类活动类型判别方法，实现轨迹中人类活动类型的精确判别；采用用户轨迹的时空活动流代替时空活动区域，基于稀疏化简的思路，建立结合活动类型和时空位置的人类活动主题提取模型，实现人类行为规律的准确表达；并在顾及时空异质性的基础上，对人类活动主题的评价进行研究。通过本课题研究，可望在理论上揭示个体时空活动与人类行为模式间的本质关系，建立人类活动内容识别、行为模式抽象的新方法，在应用上为理解、预测和干预人类行为提供支撑。

从轨迹数据中挖掘隐含的人类行为模式，可以刻画人类行为与社会、地理环境之间的关系，揭示人类行为规律。本课题以社交媒体轨迹数据为数据源，通过识别潜在的活动类型、抽取人类活动主题，对人类行为模式的精细化建模展开研究。基于知识图谱和卷积神经网络，设计兼顾文本内容与轨迹位置的人类活动类型判别方法，实现轨迹中人类活动类型的精确判别；采用用户轨迹的时空活动流代替时空活动区域，基于稀疏化简的思路，建立结合活动类型和时空位置的人类活动主题提取模型，实现人类行为规律的准确表达；并在顾及时空异质性的基础上，对人类活动主题的评价进行了研究。本课题主要成果在于提出了三种人类活动识别模型，一种人类活动模式提取模型，发布了一个可用于训练和评估人类活动类型识别模型的数据集。本课题的研究成果在理论上揭示了个体时空活动与人类行为模式间的本质关系，建立了人类活动内容识别、行为模式抽象的新方法，在应用上为理解、预测和干预人类行为提供支撑。