基于协同过滤的知识推荐模型研究

Research on Knowledge Recommendation Model Based on Collaborative Filtering

随着在线教育的普及,各类学习平台产生了海量的知识资源和用户数据。基于这些辅助数据和协同过滤算法,提出了一种混合的知识推荐模型,用于解决传统推荐系统中的冷启动和稀疏性问题,从而提高在线学习的效率。该模型首先利用用户的注册信息和对知识的隐式反馈构建初步的个人概貌,其次根据用户对知识的显式评分和其他反馈完善个人概貌,最后根据概貌和知识间的相似度进行推荐。实验表明,该模型使用不同邻居数量测试的MAE(Mean Absolute Error,平均绝对误差)均值约为0.7755,低于修正余弦算法的0.7901,且没有明显的噪点,同时MAE的标准差约为0.0724,低于皮尔逊算法的0.0837,相较于传统的修正余弦算法和皮尔逊算法,其在知识推荐上能兼顾良好的准确性和稳定性。

With the popularization of online education,various learning platforms have generated a massive amount of knowledge resources and user data.Based on these auxiliary data and collaborative filtering algorithms,this paper proposes a hybrid knowledge recommendation model to solve the cold start and sparsity problems in traditional recommendation systems,thereby improving the efficiency of online learning.Firstly,in this model,the user's registration information and implicit feedback on knowledge are used to construct a preliminary personal profile.Secondly,the user's explicit rating of knowledge and other feedback are used to improve the personal profile.Finally,recommendations are made based on the similarity between the profile and knowledge.The experiment shows that the average MAE(Mean Absolute Error)of the model tested with different number of neighbors is about 0.7755,which is lower than the 0.7901 of the modified cosine algorithm and there is no significant noise.At the same time,the standard deviation of MAE is about 0.0724,which is lower than the 0.0837 of the Pearson algorithm.Compared with traditional modified cosine algorithm and Pearson algorithm,the proposed model can well balance accuracy and stability in knowledge recommendation.