三维有偏权值张量分解在授课推荐上的应用研究  

A Three-Dimensional Partial Weight Tensor Model for Teaching Recommendation

作　　者：姚敦红 李石君[2] 胡亚慧[2,3] 

机构地区：[1]怀化学院计算机科学与工程学院,湖南怀化418000 [2]武汉大学计算机学院,武汉430072 [3]空军预警学院四系,武汉430010

出　　处：《电子科技大学学报》2017年第5期747-754,共8页Journal of University of Electronic Science and Technology of China

基　　金：国家自然科学基金(61272109);湖南省教育厅科学研究项目(15C1086)

摘　　要：为解决现今学校授课安排无推荐依据这一实际问题,首先给出了一系列形式化方法用于规约教师的专业基础、课程难度及教学评价;定义了一种加权函数计算出每组专业基础、课程难度和教学评价的综合有偏权值;构建了一种基于"教师-课程-评价-权值"四元关系的三维有偏权值张量模型,张量元素使用综合有偏权值。在此基础上,设计了一种基于Tucker分解的算法,对张量进行高阶奇异值分解(HOSVD)得到降维后的近似张量,按课程分类实现了Top_N授课推荐。实验结果表明,当迭代阈值达到一个合理值时,该方法能实现精准授课推荐,可作为一种新的智能化授课推荐方法应用于各类学校。To address the problem that the teaching arrangements are not on the basis of recommendation in current school, a series of formalized methods are used to specify teachers＇ specialty foundation, course difficulty, and teaching evaluation first. Then, a kind of weighted function is defined to calculate the comprehensive partial weight for each group of teachers＇ professional foundation, course difficulty, and teaching evaluation. Next, the three-dimensional tensor model with partial weight is built on the 4-tuples relation of teacher-course- evaluation-weight and the comprehensive weight is endowed to the tensor elements. Finally, on the basis of above, a new kind of decomposition algorithm based on Tucker Decomposition is designed to obtain the approximate tensor of dimensionality reduction with the higher-order singular value decomposition （HOSVD）, achieving the Top-N recommendation of teaching arrangements. Experiment results show that our proposed method can realize precise teaching arrangements recommendations when the iterative threshold value reaches a reasonable value, which can be used as a new intelligent recommendation method applied to the teaching arrangements in all kinds of schools.

关 键 词：数据规约 授课推荐 张量分解 三维有偏权值张量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]