情报科学 ›› 2021, Vol. 39 ›› Issue (11): 167-172.

• 业务研究 • 上一篇    下一篇

基于马尔可夫模型的图书馆用户聚类分群方法研究

  

  • 出版日期:2021-11-01 发布日期:2021-11-15

  • Online:2021-11-01 Published:2021-11-15

摘要: 【目的/意义】针对图书馆用户群体聚类分群不稳定且错误率较高的问题,提出基于马尔可夫模型的图书馆
用户聚类分群方法,提升图书馆用户聚类分群精准度。【方法/过程】采用一阶马尔可夫混合模型构建用户动作序列
模型,通过模型产生用户行为聚类,体现用户动作的动态性,采用自适应自然梯度算法,依据用户行为分离状态自
适应调整自身步长,优化模型参数学习中模型自动选择问题,实现最佳图书馆用户聚类分群。【结果/结论】通过实
验结果能够证明,实际聚类数量小于L值时,提出方法能够实现参数学习过程中模型的自动选择。提出方法的分群
数量最多,能够划分出最大的取值区间,聚类错误率最低为0.22%,聚类性能比较稳定,分群结果更加精准,达到了
设计的预期。【创新/局限】采用一阶马尔可夫混合模型实现了图书馆用户聚类分群。后续将进一步研究可考虑用
户序列间关联的高阶马尔可夫分量模型,以提高分群算法的准确性和稳定性。

Abstract: 【Purpose/significance】Aiming at the instability and high error rate of library user clustering,a library user clustering meth?
od based on Markov model is proposed to improve the accuracy of library user clustering.【Method/process】The first-order Markov hy? brid model is used to construct the user action sequence model.The user behavior clustering is generated through the model to reflect the dynamics of user actions.The adaptive natural gradient algorithm is used to adaptively adjust its own step size according to the sep? aration state of user behavior,optimize the automatic model selection problem in model parameter learning,and realize the optimal li? brary user clustering.【Result/conclusion】The experimental results show that the proposed method can automatically select the model in the process of parameter learning when the actual number of clusters is less than the value.The proposed method has the largest number of clusters,can divide the largest value interval,and the lowest clustering error rate is 0.22%.The clustering performance is rela? tively stable,and the clustering results are more accurate,which meets the design expectation.【Innovation/limitation】The first-order Markov hybrid model is used to realize the clustering of library users.In the future,we will further study the high-order Markov compo? nent model that can consider the correlation between user sequences,so as to improve the accuracy and stability of clustering algorithm.