Research On Clustering Optimization Based On Knowledge Graph Technology

Due to the development of the Internet and the rapid growth of information in the 21 st century,the relationship between information has become more and more complex,and people have to deal with a large amount of data.Therefore,the 21 st century has entered the era of big data.In order to effectively retrieve the information in the chaotic ocean of data,a large number of data objects must be analyzed and classified first,so as to realize the quantitative processing of these data objects,and then use a certain fixed specification array or vectors to represent them.this representation will meet the needs of later data statistics,retrieval,recommendation and classification.A variety of data mining and analysis techniques is used to process these data.In the ocean of data,there are often many irregular data link sets,and the basic components of such sets usually have different attributes and characteristics.These parts are connected to each other to form a network.Therefore,this irregular data network is actually a non-Euclidean graph.In the real world,there are countless datasets with this structure.For example:social network,literature index network and protein molecular structure network.They are called knowledge graphs.In application of knowledge graph,clustering is a basic operation that divides the nodes or link objects that are similar in the Euclidean space,and its structure varies depending on the application.This article shows the clustering performance of the unsupervised algorithm.Because clustering usually lacks supervision,this article has more explored the classification performance of these methods.The classification performance of these models has more intuitive reference significance in their clustering capabilities.In these technologies,this article has conducted research work on two existing problems.First,in graph convolutional neural networks,due to imperfect data samples and some implicit information that needs to be mined,it is difficult for the network to extract these features and cannot achieve better clustering results.But extracting them by enhancing the training of neural networks will greatly increase the cost of training.Therefore,we hope to convert the implicit information into display information through a certain data preprocessing method,so as to optimize the clustering ability of neural networks in high-dimensional space.Based on this idea,this article introduces a data optimization method based on a non-Euclidean graph structure.This method will analyze the original data and aggregate adjacent features to explore whether the data processed by this method can optimize the model's performance.This method digs out the implicit data in the sample from the perspective of the adjacent characteristics of the graph and the feature distribution,and enhances the network's ability to extract them.In this article,the experiment also uses several popular data sets to prove the optimization of the method in the comparative experiments of implementing classification tasks.In addition,in the graph embedding technology,the learning frequency of the binary tree-shaped hierarchical network is unbalanced.When the training is overfitted,it may cause more classification errors.Therefore,we believe that the learning frequency of each layer of the Huffman tree cluster of the neural network needs to be adjusted appropriately.In response to this idea,we designed a hierarchical gradient adjustment mechanism based on the neural network.The mechanism aims to artificially and dynamically adjust the learning frequency of neurons in each layer of the tree neural network,so that the learning frequency distribution of each layer of the neural network is more reasonable to optimize the performance of the model.We also used several other data sets to verify the effectiveness of this method and summarize experience.This article will first introduce the application background,scope of application,mainstream analysis frameworks involved,technical principles and other aspects of different types of graph analysis methods.With these theoretical foundations,we will introduce the specific content of our work in this field,technical principle,experimental proof and summary.