Research On Visual Abstraction And Analysis Of Large Scale Network Data Via Representational Learning

As a ubiquitous data structure,network is always used to encode the relationship of entities in a variety of fields,such as the social relations between people in sociology,the interaction between proteins in biology and the transactions between financial companies,and so on.With the advent of big data' era,the collection and storage ability of network data is increasing day by day.The in-depth and detailed analysis of large-scale network data can effectively understand and explore human behavior patterns.However,the scale of network data is becoming larger and larger,and the structure is becoming more and more complex,which brings some difficulties to the research of large-scale network,especially in the fields of network analysis,data mining and visual analysis.For example,the graph mining algorithm usually has a high computational complexity,and the graph visualization method is also inherently limited by the complexity of the algorithm used,the screen space,the visual clutter,and the human perception capability when reading data.Therefore,it is especially urgent to improve the demand of large-scale network data cognitive level and analysis ability through high-quality simplified expression of large-scale network data.Different from the network representation technology,the graph representation technology can effectively extract the network structure and other characteristic information,and help the user to obtain a more differentiated network representation.The sampling technique is an efficient method of data reduction,which is widely used to simplify various large-scale networks.As an interdisciplinary field,visual analysis can not only integrate the theories and methods of data mining,model analysis and other disciplines,but also realize exploratory interface and interaction design according to users' prior knowledge or specific requirements,so as to help users obtain and evaluate large-scale network data intuitively.The purpose of this paper is to study the simplified representation of large-scale network data by using graph representation learning,graph sampling technology and data visual analysis method.The main innovation points of this paper are as follows:(1)Aiming at the feature extraction of large-scale network data,a visual analysis of the graph representation algorithm driven by natural language processing technology is carried out.The analogy between elements in large-scale networks and natural language processing(NLP)terms is designed.According to the context association of network data,a large-scale corpus is constructed,and the original network structure in cyberspace is transformed into vector representation with context structure characteristics in vector space by using representation learning model.The similarity between vectors is calculated and t-SNE is used to project the vectors in two dimensions.With the help of rich visual coding methods,users' visual perception of the original network feature information is enhanced.Through visual analysis cases and expert feedback,it is proved that the chart representation algorithm has obvious advantages in efficiently extracting the original network structure and other feature information.(2)Aiming at the problem of scale reduction of large-scale network data,a visual analysis of a multi-object sampling algorithm with enhanced large-scale structure enhancement is presented.The graph sign learning model is used to learn the original network,the vector representation of the original network structure is established,and the two-dimensional space with structural similarity characteristics is obtained by projection.based on this,a multi-objective sampling model of adaptive blue noise sampling is designed,so that users can update the sampling points according to the needs while retaining the similarity features,so as to meet the needs of users.Finally,the visual analysis system is designed by integrating the above techniques,and the effectiveness of the sampling algorithm is verified by a practical case.To sum up,in view of the features of large-scale network data such as large scale and complex correlation,our study uses the graph representation learning method to achieve the abstraction and visual analysis of large-scale network data.It not only shows the large-scale network data and its structural properties from various angles and all aspects with the help of rich visual design,but also provides a large number of interactive approaches to support users to explore and discover the hidden structural patterns and feature information in large-scale network data interactively,which is of great significance to the graph data mining and application.