社交网络中融合社交关系和语义信息的推荐算法

协同过滤方法广泛应用于推荐,但是数据稀疏成为模型提供高质量推荐的一大障碍.为了解决此问题,文中提出融合社交关系和语义信息的推荐算法,提高协同过滤方法的推荐性能,有机融合稀疏的用户行为记录、项目的社交信息和项目的语义信息.应用矩阵分解技术把行为矩阵和项目社交关系映射到一个低维的特征空间,提供项目社交关系信息分解的显式解释,分析关系信息对用户行为偏好产生的影响.同时,使用社会化因子正则的级联去噪自编码器模型学习项目语义特征,改进传统深度学习模型.在真实腾讯微博和Twitter数据集上的实验表明,文中方法有效提高召回率、准确率和推荐效率.

Recommendation Algorithm with Social Relations and Content Information in Social Networks

Collaborative filtering is a widely adopted approach in recommendation. However, sparse data remain the main obstacle to provide high quality recommendations. To address this issue, a method is proposed to improve the performance of collaborative filtering recommendations by integrating sparse action records data generated by users, the social information among items and the content information of these items. Matrix factorization technique is adopted to map the user action matrix and item social relations into the low-dimensional latent feature space to provide an explicit interpretation of factorization on item social relations and analyze the influence of social relations of item on user action preferences. Meanwhile, to learn more effective features from the item content, a social factor regularized stacked denoising autoencoder model is utilized and it is an extension of conventional deep learning model. Experimental results on the Tencent blog and Twitter datasets show that the proposed model outperforms several traditional methods in terms of recall and average precision, and it improves the recommendation efficiency effectively.