Intelligent Interactive Visualization Based On Deep Learning

In the era of big data,with the development and progress of technology and society,massive amounts of data have emerged.People are confused in front of the information overload.In the era of big data,with the development of society,massive data has emerged.People don't know what to do in front of them.They are afraid of missing the value of the data,while it is difficult to deal with the data.As a new research field,interactive visualization has been widely used in daily life.Where there is data,there is interactive visualization.The systems can explore the internal relationship of data,and then show them to users intuitively.Users can obtain the information more effectively through interaction with the system.But the traditional interactive visualization method is just like the manual workshop,which needs to design rules or heuristics manually,which can not adapt to the mass data of modern society.On the other hand,in recent years,deep learning has become more and more popular in various fields due to its stronger computing power and massive data.Deep learning has a strong fitting ability,which can fit the function between various data and tasks,which provides a new idea for solving the inherent problems of interactive visualization.At present,a small number of works have begun to apply deep learning to interactive visualization.Our work also discusses the two issues of point cloud selection and paper visualization,and the main contributions are as follows;1.The point cloud is a common data form of 3D visualization.Selection is the basic operation of its interaction.Most of the previous point cloud selection methods are based on some artificially designed features,which are difficult to select widely distributed point clouds.We propose LassoNet,a 3D point cloud selection method based on deep learning,to learn the mapping relationship of the point cloud,viewpoint,and lasso to the selection result.In the method,we use three-dimensional coordinate transformation,attention filtering,and farthest point sampling to enhance the perfor-mance of the method.We propose a dateset with 30000 lasso selection records.Finally,we prove the efficiency,effectiveness,and robustness of LassoNet in the 3D point cloud selection task through quantitative analysis and user evaluation.2.The scientific paper is the main carrier of recording scientific research results,which has important value for knowledge sharing and inheritance.In scientific research work,literature research is a very important basic work,so researchers can understand the development trend of the field and find related work.An academic search engine like Google Scholar can only solve the literature retrieval of certain words,but it is not convenient for high-level understanding and analysis.Although the visualization based on text and citation information can provide the relation between papers,it is still lacking in efficient paper representation.So we put forward PFV,a paper figures visualization technology.This technology improves the text feature extraction method of natural language processing and successfully applies them to the document data to accurately represent the document.Then,we use images,dimensionality reduction,and rearrangement technology to visualize the literature on an interface,which is convenient for users to retrieve the domain overview and related work.Finally,we use the same two methods to verify the effectiveness of paper representation and visualization in PFV.The successful application of deep learning in these problems verifies that it is still useful in the interactive visualization.We can continue to visualize data better along this road.