1. 上海交通大学 网络空间安全学院， 上海 200240； 2. 携程旅游网络技术有限公司 平台商务部， 上海 200335； 3. 上海市信息安全综合管理技术研究重点实验室， 上海 220240
出版日期:2018-07-28发布日期:2018-07-28通讯作者:李生红，男，教授，博士生导师，E-mail:shli@sjtu.edu.cn.基金资助:国家重点研发计划资助项目(2016YFB0801003)Preference Degree Based Personalized Recommendation Algorithm
HUANG Jinchao，ZHANG Jiawei，CHEN Ning，CHEN Yihong1. School of Cyber Security, Shanghai Jiao Tong University, Shanghai 200240, China;2. Platform Business Department, Ctrip Travel Network Technology Co., Ltd., Shanghai 200335, China; 3. Shanghai Key Laboratory of Integrated Administration Technologies for Information Security, Shanghai 220240, China
Online:2018-07-28Published:2018-07-28摘要/Abstract
摘要： 随着在线旅游业酒店数量的日益增多，用户点评信息稀疏问题愈加严重，这不仅导致推荐准确度大幅下降，而且使传统推荐算法的计算负荷随之增加，难以满足实时性要求.基于此，从挖掘用户历史信息与待推荐物品之间潜在相关性的角度出发，对基于内容的推荐算法进行改进，提出了一种基于偏好度特征构造的个性化推荐算法.该算法通过计算偏好分来构造偏好度特征，并借助机器学习领域的分类算法得以实现.将该算法应用于线上旅游业的个性化子房型推荐，通过对真实数据集的实验与分析，验证了所提出个性化推荐算法的简便与有效性，且较传统推荐算法更具实时性和通用性.
关键词: 基于内容的推荐, 潜在相关性, 偏好度构造, 子房型推荐
Abstract: Faced with increasing number of hotels in online tourism, the problem of sparse data is becoming more and more serious. On one hand, it leads to a significant decrease in recommendation accuracy; on the other hand, the computational load of traditional recommendation algorithm is increased, which is difficult to meet the real-time requirement. So, this paper firstly proposed a preference degree based personalized recommendation algorithm which mined the potential correlation between user historical data and recommend items. The novel algorithm utilized users’ historical data to calculate preference degree and then construct new features, and its realization is based on classification algorithm. Besides, the new method is applied to make personalized recommendations in online tourism. Results from real data sets showed that the proposed preference degree based personalized recommendation algorithm is effective and universal.
Key words: content-based recommendation, potential correlation, construction of preference degree, room recommendation
PDF全文下载地址:
点我下载PDF
