基于手机信令数据的用户位置预测方法研究

Research on User Location Prediction Method Based on Cell Phone Signaling Data

手机信令数据记录了用户的移动行为,为位置预测提供数据来源。为了发挥手机信令数据在位置预测中的优势,本文构建了强化位置间前后关联的LSTM预测方法。首先,本文提出一种循环迭代的数据清洗方法,可大幅减少冗余位置信息,为提取停留点作准备;其次,以基站小区为单位提取停留点,并结合背景地理信息获取用户含有语义特征的轨迹;然后采用矩阵降维的方法将稀疏的one-hot位置编码转化为位置嵌入向量,将位置间的语义关联隐含到位置嵌入向量中,与下游拥有预测任务的LSTM网络组成基于LSTM的位置预测模型。最后的实验证明该模型对手机信令数据具有良好的预测效果。

Cell phone signaling data records the user’s movement behavior and provides a data source for location prediction.In order to exploit the advantages of cell phone signaling data in location prediction to the full,in this paper,a LSTM location prediction method was established which can strength the relation between former and latter location. Firstly,a cyclic iterative data cleaning method was proposed,which could greatly reduce redundant location information and make preparations for the extraction of stop points. Secondly,the base station cell as the unit was took to extract the stop point;and then the background geographic information was combined to obtain semantic track. Then,the matrix dimensionality reduction method was used to convert the sparse one-hot position codes into position embedding vectors,and the semantic association between the positions was implicitly embedded in the position embedding vectors. The above contents,being combined with the downstream LSTM network which own prediction task,a LSTM-based position prediction model was constituted. The final experiments show that the model has a satisfactory prediction effect on mobile phone signaling data.