【作者】 刘敏；

【导师】 崔刚；

【作者基本信息】 哈尔滨工业大学 ， 计算机科学与技术， 2015， 硕士

【摘要】 越来越多的车载系统GPS(全球定位系统)的设备改变了人们与网络交互的方式,也给我们带了了大量的代表人们位置记录的GPS轨迹信息。国内外的研究者们在用户的GPS信息的数据挖掘方向做了大量的研究,所得到的研究成果不仅在学术上有了很大的进步,也大大满足了人们的出行、生活等方面的需求。本文研究的主要内容是基于车联网GPS信息的数据挖掘。本课题的目的是基于多个车载用户的GPS轨迹信息挖掘在给定区域中的最重要地点和经典的旅行轨迹。这里重要的地点是指像北京的天安门广场这样的文化胜地和像购物广场和餐馆这样的经常有人到访的公共地点。这些信息可以帮助车载用户理解所处区域的周围的地点,也给出了出行路线推荐的研究空间。在本课题中,作者首先用基于树形结构的层次图(HG)对多个用户的出行轨迹建模。其次,在HG的基础上提出基于HITS算法的推理模型,此模型将一个人对一个地点的到访作为用户到那个地点的直接链接。这个模型对于地点的重要程度的推理主要基于以下三点:1)地点的重要程度不只考虑到访的用户数,还考虑用户的出行经验;2)用户的出行经验和地点的重要程度有相互强化的关系;3)地点的重要程度和人的旅行经验值只是相对的值,并且是区域相关的。最后,通过考虑多个地点的受欢迎程度和多个用户的旅行经验值本文挖掘了多个地点中经典的出行路径。课题最后提出了验证本文中提出的算法的方法并使用由182名车主在5年时间内收集的数据来验证课题内的理论。由验证结果可知,课题中的地点受欢迎度的挖掘在地点的代表性和排序上要比按地点的访问次数排名和按地点的访问频率排名这样的常用的数据挖掘算法表现的要好;除此之外,对于经典旅行序列的挖掘,课题中的方法也比单独的用根据地点的受欢迎度和根据用户经验值的要好。更多还原

【Abstract】 More and more GPS-enabled devices are playing a part in people’s life, which is changing the way people interacting with the internet and brings us an ocean of GPS trajectories representing people’s movement from a location to another. Researchers inland and abroad did a lot of research in the direction of the user’s GPS data mining, these research have made great progress not only in the academic, but also greatly satisfying the needs of the people’s travel and other aspects of life. The main contents of this paper are data mining based on car networking GPS information.The purpose of this project is data mining based on multiple users’ GPS track in a given area of the most important sites and classic travel sequence. The important sites here are the places refer to Beijing’s Tiananmen Square and cultural attractions where people often take a visit, such as shopping malls and restaurants. This information not only can help users understand the surrounding location, but also gives the travel routes recommending space to research.In this paper, the author first build a trajectory model for multiple users’ GPS log with tree-based multi-level graph(HG). Secondly, the author proposed a referring model based on HITS algorithm and HG. The three main ideas for this model of referring location interest are: 1) the importance of the location not only depends on the number of users visiting, also depends on the travel experience of users; 2) there are mutually reinforcement between users’ travel experience and location interest. 3) location interest and users’ travel experience are only relative value, and is the area related. Finally, mining classic travel path of multiple locations based on multiple locations’ interest and many users’ travel experience.Last but not least, the author proposed a method for the verification of the algorithm presented in the paper and a data set collected by 182 drivers in five years to verify the theoretical topic. The verification results shows that the referring model based on HITS algorithm in this paper gave a better performance than the baseline algorithms, which are rank by count and rank by frequency. Moreover, algorithm in this paper for mining travel experience and location interest is better than rank by count and rank by interest alone.更多还原