地球信息科学学报 ›› 2016, Vol. 18 ›› Issue (2): 160-166.doi: 10.3724/SP.J.1047. 2016.00160

• • 上一篇    下一篇

空间级联模式下的个性推荐模拟

李奥勇1,2(), 许珺1,**()   

  1. 1. 中国科学院地理科学与资源研究所 资源与环境信息系统国家重点实验室,北京 100101
    2. 中国科学院大学资源与环境学院,北京 100049
  • 收稿日期:2015-04-10 修回日期:2015-07-20 出版日期:2016-02-10 发布日期:2016-02-04
  • 通讯作者: 许珺 E-mail:liay.14s@igsnrr.ac.cn;xujun@lreis.ac.cn
  • 作者简介:

    作者简介:李奥勇(1992-),男,硕士生,研究方向为数据挖掘、推荐系统研发。E-mail: liay.14s@igsnrr.ac.cn

  • 基金资助:
    国家自然科学基金项目(41371380、41171296);中国科学院地理科学与资源研究所培育项目(TSYJS03)

Spatial Cascaded Model for Personalized Recommender System

LI Aoyong1,2(), XU Jun1,*()   

  1. 1. State Key laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
    2. College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2015-04-10 Revised:2015-07-20 Online:2016-02-10 Published:2016-02-04
  • Contact: XU Jun E-mail:liay.14s@igsnrr.ac.cn;xujun@lreis.ac.cn

摘要:

自个性化推荐系统出现以来,逐渐成熟并成功应用于多种互联网商品推荐,成为解决信息过载问题的有效手段。目前,各种移动终端可实时接入网络并获取用户位置,使得考虑位置的推荐进入人们的视野,但是现有的应用主要关注单一目标下的用户选择,很少考虑用户位置移动时后续活动对当前选择的影响。本文通过对连续多个选择建模,在传统推荐算法的基础上,将未来活动的影响及空间关系的影响引入传统个性推荐算法,提出空间级联模式下的推荐模型。通过实验将传统的推荐算法与空间级联模型算法作对比,综合考虑2种推荐结果的用户偏好度及空间距离变化,提出距离-偏好损益指标,同时基于百度API实现可视化。实验所得的距离-偏好损益指标和可视化结果显示,在综合考虑用户偏好和空间关系方面,空间级联模式的个性化推荐模型可得到更加合理的推荐结果。

关键词: 推荐系统, 空间级联, 个性化, 距离-偏好损益指标

Abstract:

Recommendation system has become mature and been successfully applied in many fields since its emergence. Due to the popularization of different types of mobile terminals, spatial information is brought into the recommendation systems. However, the existing researches mainly focus on spatial locations and rarely consider spatial relations. Meanwhile, the existing recommendation algorithms usually consider only the user’s history behaviors but not the influence of future behaviors on current recommendations. According to the activity chain theory, future activities have an impact on the current behavior as well as the past activities did. If a user has two steps of information retrieval, and the second step is based on the result of the first step, he would choose an item from the result which is convenient for him to make the second choice, and thus he can get the best choices at both steps of retrieval. That is to say, the current selection would be affected by the next retrieval of information. In this paper, we model the spatial distribution of the travel targets by considering one’s future intentions as well as the past data, and propose a spatial cascaded model for personalized recommender system. The model is built for situations with a series of continuous choices in the spatial space based on the traditional recommendation algorithm and the influence of future activities. The influence of spatial relation is introduced into the traditional recommendation algorithm as a distance decay function. In order to prove the feasibility of spatial cascaded recommender system, a restaurant recommender system is developed based on the proposed model. Taking into account of user’s preference and distance, a cost-benefit index was proposed to evaluate the result. The result shows that when considering further activities and spatial relations in recommendation, the system can produce a more reasonable result.

Key words: recommender system, spatial cascade model, personalization, cost-benefit index