A Study On Hierarchical Clustering Of Micro-learning Units Based On Topic Feature Centers

With the rapid development of modern information technology,the era of Web2.0 has started a truly meaningful journey of Internet online learning in the history of mankind.The change of information operation carrier and transmission form lead to the fragmentation of learning content,while the fast-paced life and multi-task processing lead to the fragmentation of learning time.In addition,the shortened cycle of information update has led to the arrival of the era of lifelong learning.Driven by the big data background of fragmented learning and lifelong learning,constructivism has given birth to a new learning paradigm----microlearning.Loose links and dynamic reorganization of learning units of microlearning are conducive to expanding learners' divergent thinking.Micro-learning resources can be provided on demand,learning time is controllable and flexible,learning interest is easy to maintain,learning efficiency is higher and other advantages,so micro-learning this new learning paradigm has been rapidly developed.Web2.0 changes the "one-way" architecture of the early web,allowing the public to send and share information to the network.Micro learning takes advantage of this function and opens the door to the era of "micro".With the rise of micro-learning,more and more micro-learning platforms are continuously established and developed.As one of the representative micro-learning platforms,Massive Open Online Courses(MOOCs)have gradually become a trend.However,the publishers of micro-learning platforms are not the only one.The release of a large number of learning resources lead to multiple sources of learning content,namely repetition problem.Learners waste most of their time on resource selection,resulting in low learning efficiency.Therefore,the rational organization and management of micro-learning resources is the basic core direction of microlearning research.Micro-learning resources are composed of loosely and dynamically connected micro-learning units,which can be divided into text,picture and video.Among them,video-style resources are popular among the public,stimulating the fun of learning,and existing in a large number of micro-learning platforms.In this paper,video-type micro-learning unit is taken as the research object.By analyzing the characteristics of video subtitle transcripts as micro-learning units.Cluster related technology is adopted to reasonably organize micro-learning resources.The main work of this study is as follows:(1)by analyzing micro-learning,this paper focuses on the micro-learning resources in the form of video in the MOOC platform.In this paper,each microlearning video is regarded as a micro-learning unit,and text subtitle transcript of each micro-learning video are analyzed.In addition,the advantages and disadvantages of commonly used text clustering methods are compared and analyzed,and the hierarchical clustering technology and AP clustering technology are mainly studied.(2)by analyzing the existing hierarchical structure and aggregation characteristics of micro-learning units,this paper creatively adopts the fusion hierarchical clustering technology to find high-quality representative points in view of the incapability of agglomeration hierarchical clustering algorithm to deal with large-scale data and the irreducibility and incompleteness of the algorithm's termination conditions.The main idea of this algorithm is: first used of the AP algorithm to get the class represent the points and the corresponding clustering clusters,and then combined with fast peak density algorithm to optimize the AP to find high quality class represent the points,finally used hierarchical agglomerative clustering technology to build relationships within the cluster tree structure and the relationships between cluster tree structure,to complete hierarchical clustering based on topic feature centers discovery.(3)by analyzing the relevant algorithms used in text clustering technology and combining the characteristics of micro-learning units,a model of integrated hierarchical clustering system for text topics of micro-learning units is constructed,and specific implementation are carried out through experiments.It is mainly divided into three stages,which are pre-processing stage of micro-learning unit text,text model representation stage and text clustering stage.The pre-processing stage is divided into text word segmentation,noise removal words and word stem extraction.In the text model presentation stage,LDA model is used to model the micro-learning unit topic.In the text clustering stage,JS distance is used as text similarity and the clustering method designed in this paper is implemented.In this paper,k-means,HAC clustering algorithm and the algorithm in this paper are respectively used for comparative experiments on the subtitle transcripts data set of the micro-learning unit.The experiment proves that the fusion hierarchical clustering method based on high-quality representative points has better clustering results for the micro-learning units,and clustering division results are more accurate.