MFWT:一种推荐学术论文的混合模型

MFWT: a Hybrid Model for Academic Paper Recommender

为了改善概率矩阵分解模型进行学术论文推荐时存在的数据稀疏性和冷启动问题,提出了一种混合推荐模型——主题矩阵分解模型.通过提出的作者-会议-时间-主题模型和传统的潜在狄利克雷分布主题模型分别构建用户和论文的主题特征,并通过这2类特征分别增强概率矩阵分解模型的用户潜在因子特征向量和项目潜在因子特征向量.实验结果表明,该模型较好地解决了概率矩阵分解模型的数据稀疏性问题和冷启动问题,有效提升了学术论文的推荐效果.

The inherent data sparsity and cold start problems in probabilistic matrix factorization（ PMF）limit the effect of academic paper recommender. To remedy the shortcomings and enhance the recommender effect,a new hybrid recommender model named as matrix factorization with topic（ MFWT） was proposed. The model constructs topic characteristics of both users and papers using the author-conferencetopic over time（ ACTOT） model and the traditional latent dirichlet allocation topic model respectively,enhancing the corresponding user and paper latent factor characteristic vectors of PMF model. Experiments show that the model well overcomes the data sparsity problem and the cold start problem of PMF and increases the effect of academic paper recommender.