A Geographic Information Visualization Recommendation Method Based on Latent Space Encoding

Abstract

The present invention discloses a geographical information visualization recommendation method based on latent space coding, which is characterized by adopting an adaptive kernel density estimation algorithm and a chart sorting algorithm, and using latent space coding to automatically realize effective coding and visualization recommendation of given discrete geographical information data for geographical space data. Specifically, it includes steps such as geographical space data input, spatial feature extraction, pre-training of the Encoder coding structure, latent space coding extraction, latent space coding sorting, visualization presentation, and user interaction editing. Compared with the prior art, the present invention has the advantages of automatically visualizing and recommending given discrete geographical information data by using latent space coding, being able to effectively mine geographical space information features, configuring several template components to support user interaction, allowing users to independently modify the visualization coding, greatly simplifying the process for users to modify the geographical visualization recommendation results, reducing the visualization threshold for users, and improving the data analysis efficiency and visualization effect.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic recommendation systems, and in particular to a geographical information visualization recommendation method based on latent space encoding. Background Art

[0002] Geospatial data is a type of data with geographical location information. In the process of smart cities, profound changes are taking place in people's work and life styles, and many of people's activities are closely related to spatial information. How to use advanced information technology to process geospatial data to achieve maximum urban efficiency and the best quality of life is a key link in the urban intelligentization process. More and more devices can record and sense a large amount of geospatial data, such as the flow of people in a region, the driving trajectories of vehicles, and the distribution of shared bicycles. Understanding this data with location tags is very important for urban management, helps to timely solve some urban traffic and public resource scheduling problems, and provides strong support for making scientific and accurate decisions.

[0003] Visualizing geospatial data is one of the important ways to understand this data. It presents geographical information in a visual way to users, which can help users quickly master the information change trends and rules, and greatly reduces the difficulty of understanding the data. In the big data era, data-driven automatic visualization systems have become an urgent need. With the continuous development of technology, corresponding chart recommendation systems have emerged continuously to help non-data scientists discover sustainable insights. However, current visualization automatic recommendation systems only consider the numerical characteristics of data and ignore the location attributes of geospatial data. How to show the change trends and rules of information according to the distribution characteristics of geospatial data brings new challenges to the automatic recommendation of visualization charts.

[0004] The visualization recommendation methods of the prior art ignore the location attributes of geospatial data, and there are problems such as being unable to perform visualization recommendations on geospatial data and the visual effects being unintuitive. Summary of the Invention

[0005] The object of the invention is to design a geographical information visualization recommendation method based on latent space coding in view of the deficiencies of the prior art. By adopting an adaptive kernel density estimation algorithm and a chart sorting algorithm, the method can automatically realize effective coding and visualization recommendation of geographical space data for the given discrete geographical information data by using latent space coding, so that users can easily visualize the local geographical information data, recommend the most suitable chart type for different data sets, effectively help users intuitively perceive the information contained in the data, effectively mine the characteristics of geographical space information, and configure several template components to support user interaction. Users can freely modify the visualization coding, greatly simplifying the process of users modifying the geographical visualization recommendation results, reducing the visualization threshold of users, and preferably solving the problems that the existing visualization recommendation methods cannot perform visual recommendation on geographical space data and the visual effect is not intuitive. The method is simple, highly practical, has high analysis efficiency and good visualization effect.

[0006] The object of the present invention is achieved as follows: A geographical information visualization recommendation method based on latent space coding, which is characterized by adopting an adaptive kernel density estimation algorithm and a chart sorting algorithm, and automatically realizing effective coding and visualization recommendation of geographical space data for the given discrete geographical information data by using latent space coding. The method specifically includes the following steps:

[0007] 1) Users submit local geographical information data, and the geographical information data includes longitude data, latitude data, attribute data, etc.;

[0008] 2) Preprocess the relevant data, process the abnormal data in the data set, and preprocess the data so that the longitude data range is within [-180, 180], and the latitude data range is within [-90, 90];

[0009] 3) Establish a geographical information data set by using the method of artificial construction and annotation, and pre-train an Encoder model. The Encoder model includes a feature extraction part, Batch Normalization processing, and Sigmoid and Leaky ReLU activation functions;

[0010] 4) Perform adaptive kernel density estimation on the preprocessed geographical information data, which can more precisely express the distribution of geographical space information in the spatial details;

[0011] 5) Input the geographical information data into the pre-trained Encoder model to realize the coding of the geographical information data, obtain the corresponding latent space coding, and implicitly represent the characteristics of the geographical information data through the latent space coding;

[0012] 6) Use the sorting model to sort the latent space encodings so that different chart types can obtain corresponding visualization rankings. The sorting results are mapped to the geographic visualization system to achieve automatic visualization presentation.

[0013] 7) Users can interactively edit the visualization results according to their own needs and edit the corresponding visualization display by interactively inputting customized parameters.

[0014] The preprocessing includes: outlier processing. Since the features are eventually converted into density surfaces, it is necessary to check missing and abnormal data. For the missing data, constant filling or interpolation filling is performed based on the understanding of the meaning of the missing variables to ensure the continuity and integrity of the data indicators. For abnormal data, it is eliminated to ensure that the data indicators are within a reasonable latitude and longitude range.

[0015] The geographic information visualization recommendation method based on latent space coding extends the kernel density estimation method into an adaptive kernel density estimation method. According to the adaptive kernel density estimation algorithm, the kernel density estimation result can be obtained more accurately, thereby better converting the discretely distributed geographic point set into a density surface.

[0016] The adaptive kernel density estimation algorithm is combined with the quadtree segmentation method. The algorithm fully combines the characteristics of geographic spatial data. First, the original data area is divided into four equal parts, and then the number of points and side lengths in the segmented area are recorded. After the judgment conditions are executed for all areas, we can get the segmentation results and adaptively calculate the bandwidth, and then find a relative balance between the scattered point dense area and the sparse area to achieve adaptive kernel density estimation.

[0017] The geographic information visualization recommendation method based on latent space coding uses a dataset-visualization pair to pre-train an Encoder model, which can perform feature encoding on the adaptive kernel density estimation result.

[0018] The geographic information visualization recommendation method based on latent space coding is characterized by automatically recommending visualization results of data in combination with latent space coding. As a deep hidden feature, latent space coding can better mine the correlation between geographic space data and contribute to the decoupled learning of the model.

[0019] The described geographical information visualization recommendation method based on latent space encoding is characterized in that after obtaining the latent space encoding, a sorting model is used to screen candidate chart types to obtain the visualization chart type that best suits the geospatial data. A scatter plot is a visualization chart that presents all data in the form of points on a map layer and is commonly used to analyze the correlation between point sets. A bubble chart is usually used to display and compare the attribute relationships between data, and the correlation between dimensions of geospatial data is explored by comparing the positions and sizes of bubbles. A heat map is suitable for viewing the overall situation of geospatial data and can smoothly display continuously distributed geospatial information data. A hexagonal binning chart can better reduce the visual difficulties brought by scatter plots when dealing with large-scale geospatial datasets.

[0020] The present invention details the recommendation results of chart types in a visual form, allowing users to have an intuitive perception of the dataset they provide. In the interactive part, users can edit the corresponding visual display by interactively inputting custom parameters.

[0021] The present invention has the following beneficial technical effects and significant technological progress compared with the prior art:

[0022] 1) In the outlier processing of the present invention, filling or elimination is flexibly carried out according to the characteristics of geospatial data to ensure the integrity and availability of geospatial data indicators.

[0023] 2) By extending the kernel density estimation method to an adaptive kernel density estimation method, the present invention can obtain more accurate results of kernel density estimation, and then better transform the discrete distribution of geospatial point sets into a density surface.

[0024] 3) The present invention automatically encodes different geospatial data, makes recommendations based on the latent space encoding results, and uses different visualization charts for display, enabling users to quickly and accurately construct geovisualization charts and discover the hidden laws and variation relationships in geospatial data.

[0025] 4) The present invention provides a system that can realize the full-automatic processing of preprocessing of geospatial data, spatial feature mining, chart recommendation, and visualization, which is convenient to use and is beneficial to improving the visual experience and interaction effect of users.

[0026] 5) Based on the adaptive kernel density estimation algorithm and chart sorting algorithm, the present invention can automatically realize the effective encoding and visualization of geospatial data, solve the problems that existing visualization recommendation methods cannot perform visual recommendations on geospatial data and the visual effects are not intuitive, and has high analysis efficiency and good visualization effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1Schematic diagram of the process of the present invention;

[0028] Figure 2 Visualization effect diagram of the adaptive kernel density estimation in Embodiment 1;

[0029] Figure 3 Visualization effect diagram of the geographical visualization intelligent recommendation system in Embodiment 1. Detailed implementation manners

[0030] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and detailed implementation manners and specific operation processes are given, but the protection scope of the present invention is not limited to the following embodiments.

[0031] Embodiment 1

[0032] Refer to Figure 1 The geographical information visualization recommendation of the present invention is carried out according to the following steps:

[0033] Step 1: Input of geographical space data

[0034] The user submits local geographical information data, and the geographical information data includes longitude data, latitude data, attribute data, etc.

[0035] Step 2: Data preprocessing

[0036] Perform preprocessing on the relevant data, process the abnormal data in the data set, and preprocess the data so that the longitude data range is between [-180, 180], and the latitude data range is between [-90, 90]. The preprocessing includes:

[0037] 1) Outlier processing: Since the features will ultimately be converted into density surfaces, it is necessary to check for missing and abnormal data. For the missing part of the data, according to the understanding of the meaning of the missing variable, constant filling or interpolation filling methods are used to ensure the continuity and integrity of the data indicators. For the abnormal data, elimination processing is performed to ensure that the data indicators are within a reasonable longitude and latitude range.

[0038] 2) Establish a geographical information data set by using the method of artificial construction and annotation, and pre-train an Encoder model. The Encoder model includes a feature extraction part, Batch Normalization processing, and Sigmoid and Leaky ReLU activation functions.

[0039] Step 3: Extract spatial features by using the adaptive kernel density estimation algorithm

[0040] Performing adaptive kernel density estimation on the preprocessed geographical information data can more precisely represent the distribution of geographical spatial information in the spatial details; the adaptive kernel density estimation algorithm combines the quadtree segmentation method. This algorithm fully combines the characteristics of geographical spatial data. First, the original data area is equally divided into four parts, and then the number of points and the side length in the divided area are recorded. After all areas have completed the judgment conditions, we can obtain the segmentation result and adaptively calculate the bandwidth, thereby finding a relative balance between the dense area and the sparse area of the scatter points and achieving adaptive kernel density estimation.

[0041] Step Four: Pre-train the Encoder encoding structure

[0042] Use the pre-trained Encoder model to perform feature encoding on the results of adaptive kernel density estimation to obtain the latent space encoding of the geographical spatial data.

[0043] Step Five: Extract the latent space encoding for spatial features

[0044] Combine the latent space encoding to automatically recommend the visualization results of the data. As deep hidden features, the latent space encoding can better mine the correlation relationships between geographical spatial data and contribute to the decoupled learning of the model.

[0045] Step Six: Use the latent space encoding for chart sorting

[0046] Use the obtained latent space encoding to screen the candidate chart types using a sorting model to obtain the visualization chart type that best suits the geographical spatial data.

[0047] Step Seven: Automatic visualization

[0048] Map the sorting result to the chart, that is, map it to the geographical visualization system to achieve automatic visualization presentation.

[0049] Step Eight: Interactive editing

[0050] Users perform interactive editing on the visualization results according to their own needs and edit the corresponding visualization display by interactively inputting custom parameters.

[0051] Refer to Figure 2 , Figure a shows the fixed bandwidth radius calculated according to the rule of thumb; Figure b shows the variable bandwidth radius calculated according to the adaptive kernel density; Figure c shows the density map generated based on the radius in Figure a; Figure d shows the density map generated based on the radius in Figure b; it can be clearly seen from the comparison results of the fixed bandwidth kernel density estimation method in this embodiment that when the adaptive kernel density estimation method is adopted, the present invention obtains more accurate kernel density estimation results.

[0052] Refer to Figure 3, as can be seen from the visualization effect diagram of the geographical visualization intelligent recommendation system in this embodiment, Module A can support interaction, Module B displays the visualization chart with the best recommendation results, and Module C displays the candidate visualization charts obtained according to the sorting model.

[0053] This embodiment effectively mines the geographical space information features and configures several template components to support user interaction. Users can freely modify the visualization encoding, greatly simplifying the process for users to modify the geographical visualization recommendation results, reducing the visualization threshold for users, and improving the data analysis efficiency.

[0054] The above has described in detail the preferred specific embodiments of the present invention, and is not intended to limit this patent. Equivalent implementations without departing from the spirit and scope of the inventive concept of the present invention shall be included within the scope of the claims of this patent. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the inventive concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field based on the inventive concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art shall be within the protection scope determined by the claims.

Claims

1. A geographical information visualization recommendation method based on latent space encoding, characterized in that An adaptive kernel density estimation algorithm and a chart sorting algorithm are adopted, and the given discrete geographic information data is automatically and effectively encoded and visually recommended for geographic space data by using latent space encoding. The specific steps are as follows: Step 1: The user submits local geographic information data including longitude, latitude, and attributes; Step 2: Preprocess the local geographic information data submitted by the user so that the longitude data range is [-180, 180] and the latitude data range is [-90, 90]; Step 3: Establish a geographic information data set by using the method of manual construction and annotation, and pre-train an Encoder model. The Encoder model includes a feature extraction part, Batch Normalization processing, and Sigmoid and LeakyReLU activation functions; Step 4: Perform adaptive kernel density estimation on the preprocessed geographic information data so that it can more finely express the distribution of geographic space information in the spatial details; Step 5: Input the geographic information data into the pre-trained Encoder model to encode the geographic information data and obtain the corresponding latent space encoding, and implicitly represent the features of the geographic information data through the latent space encoding; Step 6: Use a sorting model to sort the obtained latent space encoding so that different chart types obtain corresponding visual rankings, and map the sorting results to the geographic visualization system to achieve automatic visual presentation; Step 7: The user performs interactive editing on the visual result according to their own needs, and edits the corresponding visual display by interactively inputting custom parameters; The adaptive kernel density estimation algorithm divides the original data area into four equal parts by using a quadtree, records the number of points and the side length in the divided area, and after all areas have executed the judgment conditions, obtains the division result and adaptively calculates the bandwidth, finding a relative balance between the dense area and the sparse area of the scatter points to achieve adaptive kernel density estimation.

2. The method for visualizing and recommending geographic information based on latent space encoding according to claim 1, wherein The preprocessing of the local geographic information data submitted by the user includes: checking for missing and abnormal data. For the missing part of the data, according to the understanding of the meaning of the missing variable, it is filled with constants or interpolated to ensure the continuity and integrity of the data indicators; for abnormal data, it is removed to ensure that the data indicators are within a reasonable longitude and latitude range.

3. The method for visualizing and recommending geographic information based on latent space encoding according to claim 1, wherein The adaptive kernel density estimation algorithm is extended from the kernel density estimation method and can transform a set of discrete geographic points into a density surface.

4. The method for visualizing and recommending geographical information based on latent space encoding according to claim 1, wherein The pre-training of the geographic information data set pre-trains an Encoder model by using the data set-visualization, which can perform feature encoding on the results of adaptive kernel density estimation.

5. The geographic information visualization recommendation method based on latent space encoding according to claim 1, wherein The latent space encoding is a deep hidden feature, which can better mine the correlation relationship between geographic space data and contribute to the decoupled learning of the model.

6. The method for visualizing and recommending geographical information based on latent space encoding according to claim 1, wherein The sorting model screens the candidate chart types to obtain the visual chart type that best conforms to the geographic space data.

Images