基于隐马尔科夫模型的上下文感知活动计算

Context-Aware Activity Computing Based on Hidden Markov Model

为了推理移动用户在智能空间的活动,提出了基于隐马尔科夫模型的上下文感知活动计算.首先按照上下文的定义,采用元组方法表示移动用户和智能空间,然后根据活动理论基本构成元素和面向客体活动原理来描述用户活动和智能空间的状态变化,最后引用隐马尔科夫模型建立起用户活动与智能空间状态变化之间的联系,从而实现活动计算.该模型可以完整地描述活动分解为动作的过程,还可以根据每种活动的动作链标记用户活动数据,却不需要用户直接参与数据的标记.将该模型的动作状态数与上下文感知经验采样工具(ESM)的动作状态数进行比较,结果表明该模型的平均活动识别准确度比ESM高25%.

The context-aware activity computing based on hidden Markov model was proposed to predict the activity of mobile user in a smart space. According to the context definition, the mobile user and the smart space were expressed by the tuple method. The activity about mobile user and the state changes of smart space were then described in terms of the elements and the principle object oriented activity theory. Subsequently, the relation between user activity and the state changes of smart space was connected by hidden Markov model. Finally, the activity can be derived with the context information from smart space. The model can describe the process of activity breaking down actions. It also labels user activity data by using action chain of an activity without the user＇s participation. The experimental results show that the average accuracy in recognizing activity is increased by 25 %compared with the context-aware experience sampling tool from numbers of actions.