摘要
基于用户及其历史行为间关系的社会化推荐引起了广泛的关注,而现在的推荐模型几乎都忽略了社会关系的异质性和多样性。针对该不足,提出个人和社会隐含因子(ISLF)联合模型来进行社会化推荐的方法。该方法结合协同过滤和社交网络建模方法来进行社会化推荐,利用最新的混合隶属随机分块模型为每个用户提取相应的社会因子向量;并采用一种优化的期望最大化算法(EM算法)来优化ISLF模型,以便进行最快的期望计算;最后,基于真实的数据集豆瓣网来进行实验。结果表明该方法比现存的社会化推荐方法提供的推荐更准确,质量更高,效果更好。
The social recommendation based on users and relationship of their historical behaviours arouses wide attention, but almost all of the existing recommendation models ignore the heterogeneity and diversity of social relationship. In response to this deficiency, we proposed a social recommendation method with the joint model of individual and social latent factor （ ISLF ）. The method combines collaborative filtering and social network modelling approach, utilises the latest mixture membership stochastic block model to extract vectors of social factors for each user. Moreover, it employs an optimised expectation maximisation algorithm （EM） to optimise ISLF model so as to carry out the fast expectation computation. Finally, we made the experiments based on real dataset-DouBan, results show that the new method proposed can provide more accurate recommendations with higher quality and better effect than current social recommendation methods.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第9期98-101,205,共5页
Computer Applications and Software
关键词
推荐系统
社会化关系
社会隐含因子
协同过滤
社交网络
Recommendation system
Socialisation relationship
Latent factor
Collaborative filtering
Social network
