融合上下文信息的社会网络推荐系统

2016-01-18李慧,马小平,胡云等

智能系统学报 2015年2期

关键词：信息

网络出版地址：http://www.cnki.net/kcms/detail/23.1538.TP.20150302.1106.007.html

融合上下文信息的社会网络推荐系统

李慧1,2，马小平1，胡云2，施珺2

(1.中国矿业大学信电学院，江苏徐州 221008； 2. 淮海工学院计算机工程学院，江苏连云港 222005)

摘要：上下文环境和社会网络信息已经成为推荐系统所需的重要信息来源，在推荐系统中融入这些信息将进一步改进推荐系统的精度和用户满意度。为了提高用户对推荐系统的满意度，提出一种融入上下文信息与社交网络信息的个性化推荐系统CS。该算法应用随机决策树划分原始的用户-商品评分矩阵来进行上下文信息的处理，使得具有相似上下文信息的评分被分为一组。随后应用矩阵因式分解来预测用户对未评分项的预测。为了整合社交网络信息，在考虑上下文信息的环境下提出了一种融入社会网络关系的增强推荐模型，使用一种基于信任度的皮尔逊相关系数来衡量用户的相似度。在真实的实验数据集上进行验证，表明CS系统推荐较传统的基于上下文的和基于社会网络的推荐算法在性能上和推荐性能上有了很大的改善。

关键词：上下文；信息；社会网络；矩阵因式分解：推荐；协同过滤

DOI:10.3969/j.issn.1673-4785.201406017

中图分类号：TP391文献标志码：A

收稿日期：2014-06-11. 网络出版日期：2015-03-02.

基金项目：国家自然科学基金资助项目(61403156，61403155)；江苏省高校自然科学基金资助项目(13KJB520002,14KJB520005).

作者简介：

中文引用格式：李慧，马小平，胡云，等. 融合上下文信息的社会网络推荐系统[J]. 智能系统学报， 2015, 10(2): 293-300.

英文引用格式：LI Hui， MA Xiaoping， HU Yun， et al. Social network recommendaton system mixing contex information[J]. CAAI Transactions on Intelligent Systems, 2015, 10(2): 293-300.

Social network recommendaton system mixing contex information

LI Hui1,2， MA Xiaoping1， HU Yun2， SHI Jun2

(1. School of Information & Electrical Engineering, China University of Mining & Technology, Xuzhou 221008, China; 2. Department of Computer Science, Huaihai Institute of Technology, Lianyungang 222005, China)

Abstract：Contexts and social network information is valuable information for building an accurate recommender system. The merging of such information could further improve accuracy of the system and user satisfaction. This paper proposes the context and social (CS) network, which is novel context-aware recommender system incorporating elaborately processed social network information, in order to increase the user satisfaction on the recommendation system. The contextual information happens by applying random decision trees to partition the original user-item-rating matrix such that the ratings with similar contexts are together. The matrix factorization functionality is to predict missing preference of a user for an item using the partitioned matrix. An enhanced recommendation model aided by social relationships considering the context information is proposed. A trust-based Pearson Correlation Coefficient is proposed to measure user similarity. Real datasets based experiments showed that CS enhances its performance compared with traditional recommendation algorithms based on context and social networks.

Keywords：context; information; social network; matrix factorization; recommendation; collaborative filtering