作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2019, Vol. 45 ›› Issue (9): 1-7. doi: 10.19678/j.issn.1000-3428.0052883

• 先进计算与数据处理 • 上一篇    下一篇

基于ST-LSTM网络的位置预测模型

许芳芳1, 杨俊杰2, 刘宏志1   

  1. 1. 北京大学 软件与微电子学院, 北京 102600;
    2. 国网中兴有限公司, 北京 100761
  • 收稿日期:2018-10-16 修回日期:2018-12-16 出版日期:2019-09-15 发布日期:2019-09-03
  • 作者简介:许芳芳(1992-),女,硕士研究生,主研方向为数据挖掘、深度学习;杨俊杰,硕士;刘宏志,副教授、博士。
  • 基金资助:
    国家重点研发计划(2017YFB1002000)。

Location Prediction Model Based on ST-LSTM Network

XU Fangfang1, YANG Junjie2, LIU Hongzhi1   

  1. 1. School of Software and Microelectronic, Peking University, Beijing 102600, China;
    2. State Grid Zhongxing Co., Ltd., Beijing 100761, China
  • Received:2018-10-16 Revised:2018-12-16 Online:2019-09-15 Published:2019-09-03
  • Supported by:
    This work is supported by the National Science and Technology Support Program (No.2015BAA05B01) and the National Key R&D Program of China (No.2017YFC0210203).

摘要: 针对现有位置预测研究多数忽略时间和空间之间关联性的问题,提出一种基于时空特性的长短期记忆模型(ST-LSTM)。基于LSTM网络添加单独处理用户移动行为时空信息的时空门,并考虑用户签到的时间及空间因素,从而使模型具有时空特性。在ST-LSTM网络中引入个人修正因子,对每类用户的输出结果进行修正,在确保基本特性的基础上突出个性化,更好地学习每类用户的行为轨迹特征,同时在保证ST-LSTM网络特性的前提下给出2种ST-LSTM网络的简化变体模型。在公开数据集上的测试结果表明,与主流位置预测方法相比,该预测模型精确率、召回率、F1值都有明显提升。

关键词: 位置预测, 长短期记忆模型, 时空信息, 个性化, 行为轨迹

Abstract: Existing research on location prediction usually neglects the correlation between time and space.To address the problem,this paper proposes a Long and Short Term Memory model based on Spatial and Temporal features(ST-LSTM).Based on the LSTM network,a spatial-temporal gate that independently processes the spatial-temporal information of the user's move is added,and the spatial and temporal factors of the user's sign-in are added,so that the model has spatial-temporal characteristics.The personal correction factor is introduced in the ST-LSTM network to correct the output of each type of users,highlighting the individuality on the basis of ensuring the basic characteristics,and better learning the behavioral trajectory characteristics of each type of users.At the same time,a simplified variant model of two ST-LSTM networks is proposed under the premise of ensuring the characteristics of the ST-LSTM network.Results of testing on the public dataset show that compared with the mainstream location prediction methods,the prediction model has better performance in terms of accuracy,recall rate,and F1 value.

Key words: location prediction, Long Short Term Memory(LSTM) model, spatial-temporal information, individuality, behavioral trajectory

中图分类号: