Research And Application Of Graph Clustering Algorithm Based On Lateral Inhibition

“Everything has changed in an instant”.Since the end of last century,the rapid increase of the science and technology,especially the development of Internet technology has improved the labor productivity a qualitative leap only in a few decades.Advanced science and technology bring us abundant information,so that we can work communicate more conveniently,more efficiently and entertain with more options.At the same time of bidding farewell to the traditional life mode which has lasted for thousands of years,we are also facing a new problem of how to deal with the huge amount of information.As an important means of data processing,clustering has been paid more attention in recent years.We call it a graph clustering problem when samples can be organized into a graph.We can find that the essence attribute and internal relations between nodes which represent samples through clustering.The information will help researchers to understand sample characteristics deeply,which is helpful to find differences between communities.Therefore,graph clustering algorithm is often applied to many other data mining algorithms for data preprocessing.In this paper,we introduce a graph clustering model based on lateral inhibition.The core idea is to construct eigenvector space from similarity matrix by using spectral clustering algorithm for dimension reduction of the sample.Then establish a self-organizing feature mapping to complete the "twice clustering problem" of spectral clustering.Select the method of lateral inhibition as the adjustment strategy of competition layer node weights,so as to achieve the self-classification and reasonable clustering.After the establishment of the model,we used protein sequence data to verify the superior of lateral inhibition and analyzed the model structure and parameter selection.After determining the validity of the model,we made clustering analysis on compulsory courses and students in college of computer science and technology,Jilin University with this algorithm.Compared with the subjective anticipation,the clustering result was basically consistent,and we also analyzed curriculum assessment,teaching methods,examine methods,student learning ways and some other aspects according to of information had not been properly estimated.The reasonable clustering of sample set can help us to recognize many essential attributes and internal relations that have not been found in the sample.Special clustering algorithm also lies in the fact that it may not have the so-called "optimal",because the different clustering algorithm and different focus can find various sample characteristics,even the hiding characteristics.So on one hand we should seek the suitable clustering algorithm for sample set,on the other hand we should consider the inevitability of the clustering results combining the objective results and subjective anticipation.