基于改进LDA-FCM的UserCF知识推荐研究

Research on UserCF Knowledge Recommendation Based on Improved LDA-FCM

摘要:

针对传统知识推荐算法存在的语义缺失和精准性低问题，本文提出一种基于改进LDA-FCM的知识推荐算法。首先获取用户知识文档，采用主题优化的LDA模型挖掘用户知识主题。继而通过FCM算法将用户聚类，缩小相似度计算的遍历范围，并采用JS散度代替欧氏距离，实现FCM对象到用户的转换。最后基于UserCF算法构建用户对知识的兴趣指数并进行TOP-N推荐。爬取中国知网500篇期刊论文实测发现：与传统UserCF算法相比，改进算法的准确率、召回率和F1值分别提高了28.17%、59.62%、53.88%。

Abstract:

The improved LDA topic model and FCM user clustering algorithm are integrated into UserCF to solve the problem of semantic missing and low precision of current knowledge recommendation algorithm. An adaptive function is constructed to achieve an objective solution of the number of topics and the knowledge topic of interest to the user is mined through LDA. FCM algorithm is used to cluster the user topics, and the users are divided into clusters with similar interests, thereby narrowing the traversal range of the user similarity calculation. The JS divergence is used instead of the Euclidean distance to convert FCM object to user. Finally, based on the UserCF algorithm, the user's interest index on knowledge is constructed, and the user is recommended by TOP-N. The comparison results show that the precision, recall and F1 value of the proposed algorithm are increased by 28.17%, 59.62%, 53.88%.