闵潞, 王根生, 黄学坚.融合语义相似度的矩阵分解推荐算法[J].河南理工大学学报（自然科学版）,2020,39(4):112-117.

MIN L , WANG G S, HUANG X J.Matrix factorization recommendation algorithm by fusing semantic similarity[J].Journal of Henan Polytechnic University(Natural Science) ,2020,39(4):112-117.

融合语义相似度的矩阵分解推荐算法

闵潞1, 王根生1,2,3, 黄学坚2

1.江西财经大学 人文学院，江西 南昌 330013;2.江西财经大学 计算机实践教学中心，江西 南昌 330013;3.江西财经大学 国际经贸学 院，江西 南昌 330013

摘要:针对矩阵分解推荐算法没有考虑推荐对象本身内涵特征的问题，提出一种融合对象语义相似度的矩阵分解推荐算法。首先利用知识图谱分布式表示学习算法将推荐对象所属领域的语义数据嵌入到一个低维语义向量空间；其次计算对象间语义相似度，把该语义相似度融入矩阵分解的目标优化函数中，从语义视角弥补矩阵分解推荐算法没有考虑推荐对象本身内涵特征的不足。结果表明，该改进算法相比于传统矩阵分解推荐算法具有更高的准确率、召回率和覆盖率。

关键词:推荐算法;矩阵分解;知识图谱;分布式表示学习;语义相似度

doi:10.16186/j.cnki.1673-9787.2020.4.16

基金项目:国家自然科学基金资助项目（71461012）；江西省科技项目（GJJ181550 ）；教育部科技发展中心产学研创新基金资助项目 （2018A01012 ）；深圳市哲学社会科学规划项目（SZ2019D050）

收稿日期:2019/11/16

修回日期:2020/01/06

出版日期:2020/07/15

Matrix factorization recommendation algorithm by fusing semantic similarity

MIN Lu1, WANG Gensheng1,2,3, Huang Xuejian2

1.School of Humanities， Jiangxi University of Finance and Economics，Nanchang  330013， Jiangxi，China;2.Computer Practice Teaching Center，Jiangxi University of Finance and Economics， Jiangxi，Nanchang  330013，Jiangxi，China;3.School of International Trade and Economics， Jiangxi University of Finance and Economics，Jiangxi，Nanchang  330013，Jiangxi，China

Abstract:In order to solve the problem that the matrix factorization recommendation algorithm does not consider the characteristics of the recommended objects， a matrix factorization recommendation algorithm based on items semantic similarity was proposed. Firstly， the knowledge map distributed representation learning algorithm was used to embed the semantic data of the recommendation object domain into a low - dimensional semantic space. Then， the semantic similarity between the objects was calculated， which was integrated into the objective optimization function of matrix factorization. From the semantic perspective， it made up for the shortcomings that the recommendation algorithm of matrix factorization did not consider the characteristics of the recommended objects. The results showed that the improved algorithm had higher accuracy，recall and coverage than the traditional matrix factorization recommendation algorithm.

Key words:recommendation algorithm;matrix factorization;knowledge map;distributed representation learning;semantic similarity

融合语义相似度的矩阵分解推荐算法_闵潞.pdf