Abstract: Current trajectory analysis mainly uses clustering method to mine common residence points from multi-user trajectories, calculate user similarity to find hot spots and extract public attributes of approximate people. It has no commercial value to calculate similarity for the same user, so it is seldom studied on single-user trajectory analysis. This paper presents a method of mining frequent patterns of individual user trajectories based on location semantics. The semantic trajectory is retrieved by inverse geocoding and preprocessed to obtain Top-[k] candidate frequent location itemsets. The frequent iteration calculation of long itemsets is transformed into regular operation of hierarchical sets by using intersection and merging methods of spatiotemporal sequences, and the frequent sequence supersets and subsets are obtained. This frequent pattern mining of semantic trajectories can actively identify and discover potential car pooling needs, and provide higher accuracy for location-based intelligent recommendation such as shared car pooling and HOV Lane travel. The results of simulation car pooling experiment prove the applicability and efficiency of this method.

Key words: semantic trajectory, frequent pattern, data mining, car pooling

摘要： 现有各种轨迹分析主要利用聚类方法从多用户轨迹中挖掘公共停留点、计算用户相似度以发现热点、提取近似人群的公共属性，对同一用户计算相似度也无商业价值，因此很少对单用户轨迹分析展开研究。提出了基于地点语义的个体用户轨迹频繁模式挖掘方法。先逆地理编码求得语义轨迹并进行预处理从而求取Top-[k]候选频繁地点项集，进而采用时空序列求交集和分治归并方法，将长项集的频繁迭代计算转化为分层集合正则运算，从而求出频繁序列超集和子集。这种语义轨迹频繁模式挖掘能主动识别和发掘潜在的拼车需求，为共享拼车、HOV车道出行等基于位置的智能推荐提供更高的精准度。仿真拼车实验结果证明了该方法的适用性和高效性。

关键词: 语义轨迹, 频繁模式, 数据挖掘, 拼车