基于用户轨迹数据的群体聚集模式分析系统的设计与实现

发布时间：2018-12-19 09:33

【摘要】：随着信息科技的发展,网络正在深入地改变人们生活方式,衣食住行等生活基本需要无一不被网络渗透,社交、教育、娱乐方式也因网络的兴起有了巨大的变化。在此背景下,本系统着手研究网络日志中用户的时空数据,提取位置信息并按照时间划分构建用户时空数据,在大量单个用户时空数据的基础上,挖掘多个用户的数据在时空中的特征,分析出符合聚集模式的用户时空数据特征,发现由多个用户在持续的一段时间内参与的群体事件。本系统提出基于用户轨迹数据分析群体聚集模式,从开始时间、结束时间、参与人数、参与用户、聚集状态等方面描述聚集信息。基于Map Reduce并行处理计算模型和Spark大数据处理引擎设计和实现系统聚类算子和发现算法。采用了时间分片技术和线性插值建模用户轨迹,构建了规范、完整的用户轨迹数据。设计了应用于单个时间片的用户位置数据的DBSCAN聚类算法,得到满足密度要求的包含用户位置数据和基本信息的簇。设计和实现了适用于并行计算环境的满足规模、密度、持续时间和几何稳定等特性的人群发现算法。设计和实现了在已发现人群基础上发现用户聚集信息的聚集验证算法。在获取聚集结果的基础上,设计了展示平台,以Web的形式展示结果,通过接入GIS服务提供商的插件,实现了简洁易用友好的界面,Redis作为实时呈现数据库,利用其快速读写特性,支持聚集结果的实时呈现,MySQL作为历史存储数据库,实现历史分析结果的存储。最后,本文设计了该系统的测试方法,包括功能测试和性能测试,在测试过程中优化程序,达到工业使用条件,目前系统已经处于正常运转中。
[Abstract]:With the development of information technology, the network is deeply changing people's way of life. The basic needs of life, such as clothing, food, shelter and transportation, are permeated by the network. Under this background, this system starts to study the user's space-time data in the network log, extracts the location information and constructs the user's space-time data according to the time division, on the basis of a large number of single user's space-time data. Mining the features of multiple users' data in time and space, analyzing the spatial and temporal characteristics of user data in accordance with the aggregation pattern, and finding the group events in which multiple users participate in a continuous period of time. In this system, a cluster aggregation model based on user trajectory data is proposed to describe the aggregation information from the start time, the end time, the number of participants, the participating users, the aggregation status, and so on. The system clustering operator and discovery algorithm are designed and implemented based on Map Reduce parallel processing model and Spark big data processing engine. Time slicing technique and linear interpolation are used to model the user trajectory, and the canonical and complete user trajectory data are constructed. The DBSCAN clustering algorithm applied to the user location data of a single time slice is designed to obtain the clusters containing user location data and basic information that meet the requirements of density. A crowd discovery algorithm suitable for parallel computing environment is designed and implemented, which can satisfy the characteristics of scale, density, duration and geometric stability. This paper designs and implements an aggregation verification algorithm for discovering user aggregation information on the basis of discovered crowd. On the basis of obtaining the aggregate result, the display platform is designed to display the results in the form of Web. By accessing the plug-in of the GIS service provider, the simple and user-friendly interface is realized. Redis is used as the real-time presentation database. By using its rapid reading and writing features, the real-time presentation of aggregate results is supported. MySQL is used as a historical storage database to store the historical analysis results. Finally, this paper designs the test method of the system, including function test and performance test, optimizes the program in the test process, and achieves the industrial use condition. At present, the system is already in normal operation.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.52

【相似文献】