电子学报 ›› 2020, Vol. 48 ›› Issue (11): 2178-2185.DOI: 10.3969/j.issn.0372-2112.2020.11.012

• 学术论文 • 上一篇    下一篇

基于稀疏轨迹数据的出租车载客区域推荐

廖祝华, 张健, 刘毅志, 肖浩, 赵肄江, 刘建勋   

  1. 湖南科技大学计算机科学与工程学院, 湖南湘潭 411201
  • 收稿日期:2019-12-04 修回日期:2020-05-26 出版日期:2020-11-25 发布日期:2020-11-25
  • 通讯作者: 刘毅志
  • 作者简介:廖祝华 男,1977年9月生,湖南株洲人.副教授,分别于2004年和2012年在中科院研究生院和中科院计算所获硕士和博士学位,主要研究方向为数据挖掘、网络数据获取和计算机网络;张健 男,1994年8月生,湖南岳阳人.2019年于湖南科技大学获硕士学位,主要研究方向为轨迹数据挖掘和大数据分析技术.
  • 基金资助:
    国家科学自然基金(No.61370227,No.41871320);湖南省自然科学基金(No.2017JJ2081,No.2018JJ4052);湖南省教育厅重点项目(No.17A070);湖南省教育厅一般项目(No.19C0755)

Taxi Pick-Up Area Recommendation Based on Sparse Trajectory Data

LIAO Zhu-hua, ZHANG Jian, LIU Yi-zhi, XIAO Hao, ZHAO Yi-jiang, LIU Jian-xun   

  1. College of Computer Science and Engineering, Hunan University of Science & Technology, Xiangtan, Hunan 411201, China
  • Received:2019-12-04 Revised:2020-05-26 Online:2020-11-25 Published:2020-11-25

摘要: 基于短期出租车轨迹数据的载客区域推荐能极大减少系统开销,提高推荐效率,但常伴随着数据稀疏性的问题.针对该问题,本文提出了一种融合地理信息的隐语义模型-GeoLFM.该模型通过将出租车司机所处的客观地理环境信息,融合到司机-载客区域矩阵分解的过程中,从而弥补数据稀疏性带来的不足.同时,根据出租车实时的空间位置信息,为身处不同地点的出租车推荐不同的载客区域.实验证明,本文提出的方法与常用方法相比,推荐结果与真实的出租车司机载客情况间的平均绝对误差和均方根误差都得到大幅降低,较好的提升了推荐效果.

关键词: 轨迹挖掘, 载客推荐, 数据稀疏性, 隐语义模型, 地理信息

Abstract: Taxi pick-up areas recommendation based on the short-term taxi trajectory data can greatly reduce the system overhead and improve the efficiency of the recommendation,but it often has the problems of data sparseness.For this reason,a Latent Factor Model integrated with the geographic information,called GeoLFM,is put forward.This model makes up the faultiness of data sparseness by integrating the geographic information relating to drivers into the Matrix decomposition,which records the visiting relationship between drivers and pick-up areas.Meanwhile,different pick-up areas can be recommended for the taxis in various locations according to the real-time spatial context of the taxis.Experimental results show that,with the comparison between our proposed method and others,the Mean Absolute Error (MAE) and the Root Mean Square Error (RMSE) between the recommended results and the actual value are significantly reduced,which indicates the recommendation effect is better improved.

Key words: trajectories mining, pick-up recommendation, data sparsity, latent factor model, geographic information

中图分类号: