基于特征迁移的在线教育导师推荐方法

在推荐系统中,数据稀疏和数据冷启动问题一直是待解决的重要难题。针对推荐系统中用户数量过少、评价数据稀疏、模型启动困难等问题,以及针对在线教育领域存在特征关联的特性,提出了一种全新的基于特征迁移的导师推荐方法。为了迁移出更多有用的信息,该方法基于有限的领域特征,在目标领域和训练领域之间建立了一个基于特征相似度的桥梁。首先,获取训练领域和目标领域的推荐矩阵。然后,向量化用户和项目的特征空间,计算目标领域和训练领域之间特征的相似度。最后,构建特征迁移模型对目标领域进行迁移,得出目标推荐矩阵。研究结果表明,提出的方法能够很好地解决在线教育导师推荐领域中数据冷启动以及数据稀疏问题,与传统的推荐方法相比取得了很好的效果。

A Feature Transfer-Based E-Learning Tutor Recommendation Method

In recommendation system research, the problem of data sparseness and cold start has always been an important problem to be solved. In this paper, faced with the challenges that the number of users in e-learning systems was too relatively small, the evaluation data was sparse, and the model was difficult to start, a novel feature transfer-based recommendation method was proposed in feature-associated e-learning systems. In order to transfer more useful information, the method was designed to establish a bridge based on feature similarity between the target domain and the training domain via limited domain characteristics. First, the method obtained a recommendation matrix for the training and target domains. Then, the feature space of the user and the item was vectorized, and the similarity of features between the target domain and the training domain was calculated. Finally, the feature transfer model was constructed to migrate the target domain and the target recommendation matrix was proposed. The experimental results show that the proposed method can solve the cold start and data sparse problems in e-learning tutor recommendation task, and has achieved good results compared with the traditional recommendation methods.