Abstract: Aiming at the problem that it is difficult for the store to analyze the customer’s shopping process, a method for deep shopping data analysis is proposed. By using the phase information that reader collects from passive RFID tags to calculate the relative speed of items, then the shopping process is analyzed according to the relative movement speed. Considering the dense distribution characteristics of shopping malls, an improved I-kNN algorithm is proposed, and then I-kNN and HAC are used for in-depth analysis of the speed data to identify products of interest, hot items, hot areas and related commodities. Finally, the prototype of the proposed system is established by using the existing commercial equipment, and has been tested in the actual environment. Theoretical analysis and experimental results show that the algorithm is feasible in the analysis of shopping data, the accuracy rate is more than 78%. Meanwhile, the search data quantity and the computational complexity are reduced.

Key words: Radio Frequency Identification（RFID）, phase, velocity, improved k-nearest neighbor, dynamic time warping, hierarchical agglomerative clustering

摘要： 针对实体店很难对顾客整个购物过程进行深度分析的问题，提出了一种深度购物数据分析方法。该算法通过使用阅读器收集无源RFID标签的相位信息，以此间接计算商品的相对移动速度，再根据相对移动速度对购物过程进行分析。在考虑到超市商品的密集分布特点后，提出改进的I-kNN算法，采用I-kNN算法和HAC算法进行深入的速度数据分析，识别出感兴趣商品、热销商品、热点区域以及相关商品。随后利用现有的商用设备，对所提出的系统建立了原型，在实际环境中进行测试。理论分析与实验结果表明，该算法在购物数据分析上是可行的，准确率在78%以上，且能够减少需要分析的数据量，降低计算复杂度。

关键词: 时频识别（RFID）, 相位, 速度, 改进k邻近算法, 动态时间规划, 层次聚类算法