基于Min-Hash数据降维的伴随关系研究

Research of Accompany Relation Based on Min-Hash Data Dimensionality Reduction

文中提出一种新的计算伴随关系(同行、同停留)的方法,该方法基于手机信号数据、人脸行踪和车辆记录等多源数据。首先,利用ID-MAPPING技术将多源数据统一关联到同一ID,实现数据融合;接着,采用min-Hash算法进行数据降维,降低运算量和存储空间;最后,利用分块Hash映射将具有相同轨迹特征片段的用户映射进同一个桶,计算时空相似度,得到具有相似时空序列的人员列表,从而生成伴随关系。实测数据表明:该方法在提高伴随关系准确度的同时,运行效率比传统方法提升了12倍。

We propose A new method,which combines cellphone signal tracks,human face-recognition,vehicle driving tracks and etc,to mine accompany relation(homogeneous movement and homogeneous stay).Firstly,the method employs ID-MAPPING technique to build connection among multiple data sources with a unique ID and thus bring data sources into integration.Secondly,the method performs data dimension reduction with the min-Hash algorithm and thus reduces the amount of computation and storage.Last,the method utilizes block Hash-Mapping strategy to map persons sharing identical track characteristic segment into a same bucket.Within the bucket,we generate pairwise relations and calculate space-time similarity of each relation.The method finally returns a list of persons sharing similar spacetime series and brings them into accompany relations,An empirical test shows that the method not only improves the accuracy of capturing the relations but also increases the computation efficiency 12 times more than that of traditional computation.