基于注意力机制与改进TF-IDF的推荐算法

针对传统推荐系统主要依赖用户对物品的评分数据而无法学习到用户和项目的深层次特征的问题，提出基于注意力机制与改进TF-IDF的推荐算法（AMITI）。通过将双层注意力机制引入并行的神经网络推荐模型，提高模型对重要特征的挖掘能力。基于用户评分及项目类别改进TF-IDF，依据项目类别权重将推荐结果分类以构建不同类型的项目组并完成推荐。实验结果表明，AMITI算法能提高对文本中重要内容的关注度以及项目分配的注意力权重，有效提升推荐精度并在实现项目组推荐后改善推荐效果。

Recommendation Algorithm Based on Attention Mechanism and Improved TF-IDF

Traditional recommendation systems rely heavily on item rating data of users, and fail to learn the deep-seated features of users and items. To address the problem, a recommendation algorithm based on attention mechanism and improved TF-IDF(AMITI) is proposed. A two-level attention mechanism is introduced into the parallel neural network recommendation model to raise its ability of mining key features. Then TF-IDF is improved based on rating data and item category. The recommendation results are classified according to the weight of each item category to build different types of item groups, completing the recommendation process. Experimental results show that the proposed algorithm improves the attention acquired by the important parts in a text, and can allocate different weights to different items. It significantly improves the recommendation accuracy and thus performance by implementing item group recommendation.