基于协同过滤的海量学习数据混合推荐算法设计  

Design of a Massive Learning Data Hybrid Recommendation Algorithm Based on Collaborative Filtering

作　　者：杨续昌 胡居虎 蒋宇宏 YANG Xuchang;HU Juhu;JIANG Yuhong(College of Management,Anhui Science and Technology University,Bengbu 233030,China)

机构地区：[1]安徽科技学院管理学院,安徽蚌埠233030

出　　处：《新乡学院学报》2025年第3期47-52,共6页Journal of Xinxiang University

基　　金：安徽省教育厅重点教研项目(2022jyxm356,2023jyjxggyjY233);安徽科技学院校级教研项目(Xj2022101)。

摘　　要：海量的学习数据带来了丰富的信息资源,但时由于数据量大,用户与学习数据之间的交互不足,使得关系向量无法准确反映实体之间的真实关系,进而干扰语义相似度计算,导致不同Top-k的近邻数下的推荐效果不佳。因此,设计基于协同过滤的海量学习数据混合推荐算法。首先,利用基于物品的协同过滤推荐算法构建学习者—学习数据评分矩阵,在此基础上采用余弦相似度方式确定学习数据之间的相似性,通过调整相似性权重避免评分数据不足导致的偏差。利用改进的TransHR模型将学习数据知识图谱实体间的复杂关系转换到关系空间中,得到准确的学习数据的向量化表示,通过计算学习数据向量之间的欧式距离度量学习数据之间相似度,量化分析直接关系与间接关系对相似度计算的影响,确定学习数据之间的语义相似度。最后,将其与基于用户评分的相似性作线性加权融合后,计算学习者对学习数据的预测评分,并生成推荐列表,实现海量学习数据推荐。实验结果表明:该算法可实现生产管理课程数据的精准推荐,提高实验组学生的生产管理课程学习效果;在Top-k的近邻数为80条件下,融合比例系数为5∶5,推荐列表长度为3时,学习数据推荐效果更突出。The massive amount of learning data not only brings abundant information resources,but also due to the large amount of data,there is insufficient interaction between users and learning data,which makes the relationship vector unable to accurately reflect the true relationship between entities,thereby interfering with semantic similarity calculation and resulting in poor recommendation performance under different Top-k neighbor numbers.Therefore,this article designs a massive learning data hybrid recommendation algorithm based on collaborative filtering.Firstly,a learner learning data scoring matrix is constructed using an item based collaborative filtering recommendation algorithm.Based on this,cosine similarity is used to determine the similarity between learning data,and the similarity weights are adjusted to avoid bias caused by insufficient scoring data.Next,the improved TransHR model is used to transform the complex relationships between entities in the learning data knowledge graph into a relationship space,obtaining accurate vectorized representations of the learning data.The Euclidean distance between the learning data vectors is calculated to measure the similarity between the learning data,and the impact of direct and indirect relationships on similarity calculation is quantitatively analyzed to determine the semantic similarity between the learning data.Finally,it is linearly weighted and fused with similarity based on user ratings to calculate learners'predicted ratings of the learning data and generate a recommendation list,achieving massive learning data recommendation.The experimental results show that the algorithm can achieve accurate recommendation of English CET-6 listening learning data and improve the English learning effectiveness of the experimental group students.When the number of neighbors in Top-k is 80,the fusion ratio coefficient is 5:5,and the recommendation list length is 3,the learning data recommendation effect is more prominent.

关 键 词：协同过滤 余弦相似度 改进TransHR模型 欧式距离 语义相似度 线性加权融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]