Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (22): 132-141.DOI: 10.3778/j.issn.1002-8331.1908-0435

Previous Articles     Next Articles

Application Research of Multi-subdomain Random Forest in Context-Aware Recommendation

LI Ling, GU Xiaomei, LIU Zihao   

  1. 1.College of Computer and Information, Hohai University, Nanjing 211100, China
    2.School of Foreign Languages and Cultures, Nanjing Normal University, Nanjing 210097, China
    3.School of Computer Science, Jiangsu University of Science and Technology, Zhenjiang, Jiangsu 212003, China
  • Online:2020-11-15 Published:2020-11-13



  1. 1.河海大学 计算机与信息学院,南京 211100
    2.南京师范大学 外国语学院,南京 210097
    3.江苏科技大学 计算机学院,江苏 镇江 212003


Context-aware recommender system improves recommendation accuracy by adding context information, which is widely used in practical applications. However, the traditional context-aware recommendation method gives the same weight to the contextual factors, ignoring the differences of users’ preference items in different contexts, and the different influences of the contextual factors in the recommendation process. This paper presents a context-aware recommendation method based on multi-subdomain random forest algorithm. The method first improves the random forest algorithm by randomly selecting features from multiple feature subspaces which are classified by the importance of features. In addition, it uses the improved random forest algorithm to decompose and reduce the dimensions of context features of users, items and contexts. Finally, it uses the collaborative filtering recommendation algorithm to carry out personalized recommendations for cold chain logistics distribution. The simulation experiments of two data sets of LDOS-CoMoDa and Cycle Share show that the average absolute error of this method is reduced by nearly 10% compared with the traditional methods, which effectively improves the prediction accuracy of the recommender system and provides reference for the application of context-aware recommendation.

Key words: big data, random forest, context-aware, location-based service, recommender system


情境感知推荐系统通过增加情境信息来提高推荐精度,在实际应用中得到广泛的应用。然而,传统的情境感知推荐方法存在赋予情境因素相同权重,忽略了用户在不同情境下所偏好项目的不同,以及情境因素在推荐过程中所起的影响作用不同的问题。提出一种基于多子域随机森林算法的情境感知推荐方法。该方法对特征重要性按权值大小进行排序,将权值的取值区域分为多个大小相等的子区域,在这些子区域中随机选择特征,构造特征子空间来改进随机森林算法;通过改进的随机森林算法来分解并降低用户、项目和情境的特征维度;使用协同过滤推荐算法来进行冷链物流配载个性化推荐。对LDOS-CoMoDa和Cycle Share两个数据集进行仿真实验,结果表明该方法相比传统方法平均绝对误差减少近10%,有效地提高了推荐系统的预测精度,为情境感知推荐的应用提供借鉴。

关键词: 大数据, 随机森林, 情境感知, 基于位置服务, 推荐系统