Abstract:

The traditional collaborative filtering recommendation algorithm has a common problem of data sparsity; heterogeneous information network models applied in the field of recommendation algorithm usually identify the similarity relationship of objects symmetrically, which has limitations in the actual problem processing. To solve these problems, a fuzzy recommendation algorithm for asymmetric heterogeneous information networks is proposed. Firstly, the algorithm uses the advantage of fuzzy set theory in dealing with the degree of user preference. Then, according to the rich semantic information of meta path in heterogeneous information network, the user association from different angles is obtained. Then, the asymmetric coefficient of object relationship is introduced in the similarity calculation, and the calculation results of different feature element paths are weighted. In order to improve the accuracy of similarity relationship between users, the score prediction is realized by matrix decomposition method. Experimental results show that the algorithm effectively solves the problem of data sparsity and improves the recommendation accuracy.

Key words: fuzzy set, heterogeneous information network, meta-path, asymmetry coefficient

摘要：

传统的协同过滤推荐算法存在普遍的数据稀疏性问题；应用于推荐算法领域的异构信息网络模型对对象的相似关系认定通常是对称的，这种对称关系的认定在实际问题的处理中存在局限性。为解决上述问题，提出一种非对称异构信息网络的模糊推荐算法。该算法利用模糊集理论在处理用户喜好程度方面的优势，从模糊的信息种获取用户的准确偏好，根据异构信息网络中元路径的丰富语义信息，获取不同角度的用户关联，在相似度计算中引入对象关系的非对称系数，对不同特征元路径的计算结果进行加权，以此提高用户之间相似关系的准确度，通过矩阵分解的方法实现评分预测。实验结果表明，该算法有效解决了数据稀疏性问题，提升了推荐精度。

关键词: 模糊集, 异构信息网络, 元路径, 非对称系数