基于多上下文信息的协同过滤推荐算法

随着电子商务和互联网的发展,数据信息呈爆炸式增长,协同过滤算法作为一种简单而高效的推荐算法,能在一定程度上有效地解决信息爆炸问题。但是传统协同过滤算法仅通过单一评分来挖掘相似用户,推荐效果并不占优势。为了提高个性化推荐的质量,如何充分利用用户(物品)的文本、图片、标签等上下文信息以使数据价值最大化是当前推荐系统亟待解决的问题。对此,提出了一种融合多种类型上下文信息的协同过滤算法。以用户商品交互信息为二部图,根据不同类型上下文的特点构建不同的相似度网络,设计目标函数在多种上下文信息网络的约束下联合矩阵分解,并学得用户商品的表示学习。在多个数据集上进行了充分实验,结果表明,融合多种类型上下文信息的协同过滤算法不仅能有效提高推荐的准确度,而且能在一定程度上解决数据稀疏性问题。

Collaborative Filtering Recommendation Algorithm Based on Multi-context Information

With the development of e-commerce and the Internet,as well as the explosive growth of data information,collaborative filtering algorithm as a simple and efficient recommendation algorithm can effectively alleviate the problem of information explosion.However,the traditional collaborative filtering algorithm only uses a single rating to mine similar users,and the recommendation effect is not dominant.In order to improve the quality of personalized recommendations,how to make full use of the user(items)text,pictures,labels and other information to maximize the value of data is an urgent problem to be solved by the current recommendation system.Therefore,user-product interaction information is used as a bipartite graph,and different simila-rity networks are constructed according to the characteristics of different contexts.The design objective function is combined with matrix decomposition under the constraints of various information networks and user or item embedding can be gotten.Extensive experiments are conducted on multiple data sets,and the results show that the collaborative filtering algorithm by fusion of multiple types of information can effectively improve the accuracy of recommendations and alleviate the problem of data sparsity.