﻿ 基于社交网络的上下文感知推荐算法 Context-Aware Recommendation Algorithm Based on Social Network

Software Engineering and Applications
Vol.04 No.05(2015), Article ID:16212,13 pages
10.12677/SEA.2015.45014

Context-Aware Recommendation Algorithm Based on Social Network

Lei Chen, Gui Li, Zhengyu Li, Ziyang Han, Ping Sun

Faculty of Information & Control Engineering, Shenyang Jianzhu University, Shenyang Liaoning

Email: cl090303009@163.com

Received: Oct. 2nd, 2015; accepted: Oct. 16th, 2015; published: Oct. 26th, 2015

ABSTRACT

Context and social network information is very valuable for building accurate recommendation system. However, traditional recommendation systems could not combine different types of such information effectively to further improve the quality of recommendation. Therefore, we propose the context-aware recommendation algorithm based on social network SCRA (Social Network Based Context-Aware Recommendation Algorithm). For different types of context, we partition the rating matrix of initial user item by introducing random decision tree. In the leaf node of the tree, matrix factorization is used. Besides, we incorporate social network information by introducing Pearson Correlation Coefficient which contains context information to measure the similarity of users. To predict the rating of users for an item, we solve the objective function. Real datasets based experiments show that SCRA is better than the traditional recommendation algorithm in terms of precision.

Keywords:Recommendation System, Context-Aware, Social Network, Matrix Factorization

Email: cl090303009@163.com

1. 引言

(a) (b) (c)

Figure 1. Context-aware recommendation; (a) User-item-rating matrix; (b) Context-aware user-item-rating matrix; (c) Social network

2. 相关工作

2.1. 矩阵分解

(1)

(2)

(3)

(3) 式可应用随机梯度下降法(SGD)来求解，通过迭代更新用户特征矩阵和项目特征矩阵[4] 。

2.2. 上下文感知推荐系统

2.3. 社交推荐

3. 基于社交网络的上下文感知推荐算法

3.1. 基本概念

3.2. 上下文感知推荐

3.2.1. 随机决策树的构造

3.2.2. 用户评分预测

(4)

Figure 2. Random decision trees (one tree)

(5)

(6)

3.3. 基于社交网络的改进算法

3.3.1. 用户相似度计算

(7)

(8)

3.3.2. 基于上下文的用户相似度计算

PCC的一个优点是它考虑到有些用户会对大部分项目给予很高的评分(例如，在五等Likert scale中的4或5)，而另一些挑剔的用户通常会给予低评分(例如，在五等Likert scale中的2或3)。然而，这个经典的相似性度量方法只利用了评分的值，未考虑任何上下文信息，而上下文是另一类对相似度计算很有价值的信息。为了进一步提高用户相似度计算的精度，提出了一种融入上下文信息的皮尔森相关系数：

(9)

c代表上下文信息，项目的权重通过下式计算：

(10)

(11)

3.3.3. 目标函数的生成

(12)

(13)

(14)

4. 实验结果及评价

4.1. 实验方法

4.1.1. 实验数据集

4.1.2. 实验对比

RPMF是随机分割矩阵分解的简称，是一种利用随机分割技术基于树结构构造的上下文协同过滤模型。具体来说，它将具有相似上下文的评分分配到决策树的同一结点上。然后，在每个结点上应用矩阵分解来预测评分。多个结点和决策树上的预测结合起来生成最终的推荐。尽管该模型所提出的算法也用到了基于树的方法，但和本文提出的算法相比，它们仍有着明显的不同：1) RPMF通过处理和上下文信息相关的隐语义的值间接处理上下文信息，而本文提出的SCRA直接处理上下文信息；2) RPMF在树的每个结点都应用矩阵分解，而本文提出的SCRA仅在树的叶子结点应用矩阵分解。另一个重要的不同之处在于RPMF未考虑任何的社交网络信息。

SoReg是一种基于社交网络信息的推荐模型。在基础的矩阵分解模型的基础上加入了一个社交正则化项。它有两种变式：1) 基于平均值的正则化，它限制用户的偏好值和用户好友偏好平均值之间的差异；2) 基于个体的正则化则限制用户偏好和他每位好友的偏好之间的差异。实验中，我们只比较本文的算法和更为精确的基于个体的变式。

4.1.3. 评价指标

Table 1. Statistics of the Douban dataset

(15)

4.2. 结果评价

Figure 3. Impact of parameter α; Experimental parameters (λ = 0.1, α = 0.01, latent factor dimensionality = 10, iterations = 20)

Figure 4. Impact of number of decision trees (λ = 0.1, α = 0.01, latent factor dimensionality = 10, iterations = 20)

Figure 5. Impact of quantity of contextual information (λ = 0.1, α = 0.01, latent factor dimensionality = 10, iterations = 20)

Table 2. Performance comparison on the Douban dataset

(a) SCRA (b) SoReg(c) RPMF

Figure 6. Impact of latent factor dimensionality (iterations = 10)

(a) SCRA (b) SoReg(c) RPMF

Figure 7. Impact of iterations

5. 结论

Context-Aware Recommendation Algorithm Based on Social Network[J]. 软件工程与应用, 2015, 04(05): 101-113. http://dx.doi.org/10.12677/SEA.2015.45014

1. 1. Zhong, E.H., Fan, W. and Yang, Q. (2012) Contextual collaborative filtering via hierarchical matrix factorization. Proceedings of the SIAM International Conference on Data Mining, Anaheim, 26-28 April 2012, 744-755. http://dx.doi.org/10.1137/1.9781611972825.64

2. 2. Liu, F.K. and Lee, H.J. (2010) Use of social network information to enhance collaborative filtering performance. Expert Systems with Applications, 37, 4772-4778. http://dx.doi.org/10.1016/j.eswa.2009.12.061

3. 3. Ma, H., Zhou, D., Liu, C., Lyu, M.R. and King, I. (2011) Recommender systems with social regularization. Proceedings of the 4th ACM International Conference on Web Search and Data Mining, Hong Kong, 9-12 February 2011, 287- 296.

4. 4. Jiang, M., Cui, P., Liu, R., Yang, Q., Wang, F., Zhu, W. and Yang, S. (2012) Social contextual recommendation. Proceedings of the 21st ACM International Conference on Information and Knowledge Management, Maui, 29 October-2 November 2012, 45-54. http://dx.doi.org/10.1145/2396761.2396771

5. 5. 王立才, 孟祥武, 张玉洁 (2012) 上下文感知推荐系统. 软件学报, 1, 1-20.

6. 6. 李蕊, 李仁发 (2007) 上下文感知计算及系统框架综述. 计算机研究与发展, 44, 269-276.

7. 7. Adomavicius, G. and Tuzhilin, A. (2011) Context-aware recommender systems. In: Ricci, F., Rokach, L., Shapira, B. and Kantor, P.B., Eds., Recommender systems handbook, Springer, New York, 217-253. http://dx.doi.org/10.1007/978-0-387-85820-3_7

8. 8. Karatzoglou, A., Amatriain, X., Baltrunas, L. and Oliver, N. (2010) Multiverse recommendation: N-dimensional tensor factorization for context-aware collaborative filtering. Proceedings of the 4th ACM Conference on Recommender Systems, Barcelona, 26-30 September 2010, 79-86. http://dx.doi.org/10.1145/1864708.1864727

9. 9. Liu, X. and Aberer, K. (2013) SoCo: A social network aided con-text-aware recommender system. Proceedings of the 22nd International Conference on World Wide Web, 13-17 May 2013, Rio de Janeiro, 781-802.

10. 10. Koren, Y., Bell, R. and Volinsky, C. (2009) Matrix factorization techniques for recommender systems. Computer, 42, 30-37. http://dx.doi.org/10.1109/MC.2009.263

11. 11. Buchin, M., Dodge, S. and Speckmann, B. (2012) Context-aware similarity of trajectories. In: Hutchison, D., Kanade, T. and Kanade, T., Eds., Geographic Information Science, Volume 7478 of Lecture Notes in Computer Science, Springer, Berlin, 43-56.