基于LDA-HMM的知识流动模式发现研究

Research on Knowledge Flow Discovery Based on LDA and HMM

【目的/意义】知识流动模式能够动态反映知识形态在知识流动中发生的变化,体现学科系统中不同知识体系的发展规律,因此从科学文献角度出发研究知识流动模式的发现具有重要意义。【方法/过程】以图书情报学领域为例,设计一种组合方法,首先采用LDA(latent Dirichlet allocation,隐狄里克雷分布)主题模型进行领域内文献的主题聚类,然后提取引用和被引用数据构成主题知识流入、流出特征,并作为HMM(Hidden Markov Model,隐马尔可夫模型)训练模型的观测值序列,识别不同的隐藏状态,进一步对模型的拟合效果进行评估,最后通过聚类分析将隐藏状态序列与知识流动模式一一对应,揭示图书情报学领域存在不同知识流动模式的差异性。【结果/结论】实验结果显示,不同类型的知识流动模式具有不同的表现形式,反映了领域内部主题研究的演变历程,为理解和认识科学发展趋势具有一定价值。

【Purpose/significance】Knowledge flow pattern will dynamically reflect the changes of knowledge forms, and embody the development of different knowledge systems in a subject. So it is quite meaningful to study the knowledge flow pattern from the perspective of scientific literature.【Method/process】A combination method is designed for Library and Information Science(LIS) as an example in this paper. First, the LDA(Latent Dirichlet Allocation) model is used to gather the topic distribution of the documents in the field. And then the reference and referenced data are used as the features of knowledge inflow and outflow, which is input as the observation sequences into the HMM(Hidden Markov Model) training.After identifying different hidden states through HMM, the fitting effect of the model is further evaluated. Finally, the hidden state sequence is corresponded to the knowledge flow pattern by cluster analysis, which reveals the differences among different knowledge flow patterns in the field of LIS.【Results/conclusion】The results show various types of knowledge flow patterns have display different forms, and the evolution of some topics research inside the field of LIS is reflected, having certain value on understanding and recognizing the trends of the development of sciences.