多粒度时空事件序列相似度算法研究

现有研究集中于不带有时间空间信息或带有固定时间空间信息的活动序列相似度计算，没有从不同层次来度量用户行为序列的相似性，为了实现对用户行为多粒度多视角的动态认知，提出一种基于序列比对算法Needleman-Wunsch的多粒度时空序列比对算法（multi-granular spatiotemporal sequences alignment，MGSSA），扩展了NW算法的得分函数以结合时间、空间信息，通过粒度调控实现了从不同的粒度来计算时空事件序列的相似度.实验证明，多粒度时空序列比对算法MGSSA是有效且可行的. 

Study on Similarity Algorithm of Multi-Granularity Spatio Temporal Event Sequences

In the intelligent environment, by analyzing the similarity of the user''s daily activity sequences, it is possible to group users with similar behaviors, detect abnormal behaviors of users and query other behavior sequences that are similar to a given sequence, so as to personalize the user finely. That can provide users with a perfect personalized information services. The existing research focuses on the activity similarity calculation without spatiotemporal information or with fixed spatiotemporal information, and does not measure the similarity of user behavior sequences from different levels. In order to realize the dynamic cognition of multi-granularity and multi-view of user behavior, a multi-granularity spatiotemporal sequence algorithm (MGSSA) based on Needleman-Wunsch algorithm was proposed. It was arranged to extend the score function of NW algorithm to combine the temporal information and spatial information, and to realize the similarity of spatiotemporal event sequences from different granularities through granular control. Finally, some experiments were carried out. The results show that the multi-granular spatiotemporal sequence alignment algorithm is effective and feasible.