面向购物行为的感知技术研究与应用

发布时间：2019-01-29 20:59

【摘要】：顾客的购物时浏览商品的时空数据对于商家进行商业决策具有重要的参考价值。在电子商务中,商家通过记录顾客购物时的浏览记录,可以分析和推测顾客的偏好和习惯。利用这些数据,商家可以充分挖掘顾客的潜在需求,为顾客提供个性化的推荐服务,同时也便于商家及时调整营销策略与手段,极大地减小了商业风险。然而在线下实体购物场景中,还没有一个有效的方法来识别和收集顾客的购物信息。因此,研究与应用线下购物过程中的信息感知技术具有重要的应用前景。本文利用商场已有的无线基础设施,分析无线信号传输特性,设计了并实现了一套线下购物信息收集识别方案,实现了一个原型系统。本课题的主要创新点和工作内容如下:改进了现有基于CSI信道状态信息的位置指纹生成方案。该方案充分利用无线传输过程中的CSI信道状态信息的多径成分,同时考虑了信号的空间特性和频率特性,在缩减数据维度的同时保证了位置指纹的准确性。提出了一种基于动态位置指纹时间序列相似性的室内定位方法。与现有通过提前采集静态无线信号位置指纹信息来实现定位的方法不同,该方法采集目标与参考点同一时刻的无线信号特征信息,利用新的位置指纹生成方案生成CSI动态位置指纹时间序列。通过DTW(Dynamic Time Warping)来评估位置指纹时间序列的相似性来实现定位。实验证明该方法在复杂的商场室内环境中,能够保持较高的精度和准确性。提出了一个顾客线下购物浏览位置信息识别与收集原型系统。该系统利用本文提出的室内定位方法,在人流量不高的情况下,识别出顾客购物时的基本运动特性,如行走、驻足等;同时该方案关联分析多个监测点的数据,描绘出顾客购物时的大致运动轨迹。在人流量较大的环境下,采用k-mediods聚类算法对无线覆盖范围内的信号进行分析,识别出顾客聚集的热点区域。本文进行了一系列实验来验证以上算法及方案,设计并实现了一个顾客购物信息识别与收集原型系统。实验结果表明,该系统可以在定位上实现0.9米的精度,在运动状态及热点区域识别上达到了85%的准确率。
[Abstract]:It has important reference value for merchants to browse the time and space data of goods when they are shopping. In electronic commerce, merchants can analyze and speculate on customers' preferences and habits by recording the browsing records. Using these data, merchants can fully tap the potential needs of customers and provide personalized recommendation services for customers. At the same time, it is convenient for businesses to adjust marketing strategies and means in time, which greatly reduces the business risk. However, there is no effective method to identify and collect customer shopping information in off-line physical shopping scenarios. Therefore, the research and application of offline shopping process of information perception technology has an important application prospects. Based on the existing wireless infrastructure of shopping mall, this paper analyzes the characteristics of wireless signal transmission, designs and implements a set of offline shopping information collection and identification scheme, and implements a prototype system. The main innovations and work of this thesis are as follows: the existing location fingerprint generation scheme based on CSI channel state information is improved. The scheme makes full use of the multipath components of the CSI channel state information in wireless transmission, and takes into account the spatial and frequency characteristics of the signal, thus reducing the data dimension and ensuring the accuracy of the location fingerprint. An indoor location method based on time series similarity of dynamic location fingerprint is proposed. This method is different from the existing methods, which collect the position fingerprint information of the static wireless signal in advance, and collect the characteristic information of the wireless signal at the same time between the target and the reference point. The CSI dynamic position fingerprint time series is generated by a new position fingerprint generation scheme. The similarity of time series of location fingerprint is evaluated by DTW (Dynamic Time Warping) to realize localization. Experiments show that the method can maintain high accuracy and accuracy in complex indoor environment. In this paper, a prototype system of customer online shopping location recognition and collection is proposed. The system uses the indoor positioning method proposed in this paper, in the case of low flow of people, identify the basic characteristics of customer shopping, such as walking, stopping, etc. At the same time, the scheme analyzes the data of multiple monitoring points and describes the general trajectory of shopping. The k-mediods clustering algorithm is used to analyze the signals in the wireless coverage area under the environment of large population flow, and the hot spots of customer clustering are identified. In this paper, a series of experiments are carried out to verify the above algorithms and schemes, and a prototype system of customer shopping information identification and collection is designed and implemented. The experimental results show that the system can achieve the accuracy of 0.9 meters in the positioning, and achieve the accuracy of 85% in the recognition of the motion state and the hot spot area.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN92

【相似文献】