Method for using large language model to automatically generate map annotations on basis of text

Abstract

A method for using a large language model to automatically generate map annotations on basis of a text relates to the field of visualization, and comprises: constructing a plurality of different types of map annotation templates on the basis of a deconstructed map annotation design space; using a large language model to extract geographic information content in a text to be annotated; and on the basis of the extracted geographic information content, matching a corresponding map annotation template so as to automatically generate an annotated map. A user can adjust annotation information and the annotation method by modifying the output of the large language model or parameters of the map annotation template matching the large language model, so as to complete customized map annotations. By using this method, corresponding map annotations can be automatically generated, and the map annotation result can be adjusted on the basis of the user input, improving the map annotation efficiency, lowering the barrier to use annotated maps, and laying a foundation for large-scale promotion and application of map annotations.

Description

A method for automatically generating map annotations based on text using a large language model Technical Field

[0001] The present invention belongs to the field of visualization, and in particular relates to a method for automatically generating map annotations based on text using a large language model. Background Art

[0002] Maps have always been an important medium for expressing geographic spatial information. Map creators use annotations on maps to convey more spatial information based on the cartographic map. However, creating relevant map annotations requires annotation design and code programming skills, which places high demands on map creators.

[0003] Existing map annotation tools fall into two categories. Simple online map annotation tools draw on image annotation methods, allowing users to add various shapes to the map for simple map annotation. However, these tools fall far short of meeting users' diverse annotation needs. Professional mapping tools, while capable of supporting a wide range of annotation needs, require users to possess highly specialized skills. The complex and tedious map annotation creation process makes these tools extremely difficult to use and limits their scope of application.

[0004] Summary of the Invention

[0005] In response to the defects in the existing technology, the purpose of the present invention is to provide a method for automatically generating map annotations based on text using a large language model. The large language model is used to extract the geographic information contained in the text, and this information is automatically mapped into the map annotation design space. Map annotations are automatically generated, which can improve the efficiency of map annotation and lower the threshold for using annotated maps. At the same time, by constructing multiple types of map annotation templates, it can meet the diverse map annotation needs of users and lay the foundation for the large-scale promotion and application of map annotation.

[0006] To achieve the above objectives, the present invention adopts a technical solution: a method for automatically generating map annotations based on text using a large language model, the method comprising the following steps:

[0007] S1. Construct a map annotation template based on the deconstructed map annotation design space;

[0008] S2. Extracting geographic information from the text to be annotated using a large language model;

[0009] S3. Automatically generate an annotated map based on the extracted geographic information content and a map annotation template.

[0010] Furthermore, the method also includes adjusting the annotation information and the annotation method by changing the output of the large language model to complete customized map annotation.

[0011] Furthermore, the method further includes adjusting the annotation information and the annotation method by changing the parameters of the large language model matching map annotation template to complete customized map annotation.

[0012] Furthermore, the map annotation design space in step S1 includes geographic element dependencies and annotation forms, wherein the geographic element dependencies include geographic elements existing in the map, and the annotation forms and data mapping on their attributes reflect the annotation method and annotation information.

[0013] Furthermore, in step S1, simple map annotations are combined to construct a map annotation template.

[0014] Furthermore, in step S1, the map annotation template adjusts the geographic element dependency or status information in the map annotation template according to the received input parameters.

[0015] Furthermore, the map annotation template in step S1 includes an element status map annotation template, a spatial relationship map annotation template and a trend information map annotation template. The element status map annotation template is used to annotate the element status of a single geographic element, the spatial relationship map annotation template is used to annotate the spatial relationship between two geographic elements, and the trend information map annotation template is used to annotate the trend information between two geographic elements.

[0016] Furthermore, in step S2, a large language model prompt word template is constructed to meet the large language model prompt word generation requirements of different extraction tasks.

[0017] Furthermore, in step S2, by embedding the text to be annotated, the sample text and the sample output into the prompt word template, the prompt words required for the task of extracting geographic information from the text content can be generated, that is, the geographic information content extracted from the text.

[0018] Furthermore, in step S3, the corresponding map annotation template is matched according to the geographic information content extracted from the letter, and the required input parameters are input into the corresponding map annotation template through the large language model, thereby automatically completing the map annotation.

[0019] The beneficial technical effects of the present invention are: using a method disclosed in the present invention to automatically generate map annotations based on text using a large language model, constructing a variety of different types of map annotation templates based on the deconstructed map annotation design space; using the large language model to extract geographic information content in the text to be annotated; matching the corresponding map annotation template based on the extracted geographic information content to automatically generate an annotated map. The user can adjust the annotation information and annotation method by changing the output of the large language model or by changing the parameters of the map annotation template matched by the large language model to complete customized map annotation. Using the method disclosed in the present invention, the corresponding map annotation can be automatically generated, and the map annotation results can be adjusted according to user input, thereby improving map annotation efficiency, lowering the threshold for using annotated maps, and laying the foundation for large-scale promotion and application of map annotation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] FIG1 is a flow chart of a method for automatically generating map annotations based on text using a large language model disclosed in Example 1 of the present invention;

[0021] FIG2 is a schematic diagram of a map annotation effect obtained by using different annotation forms for different geographic elements using a method for automatically generating map annotations based on text using a large language model disclosed in Example 1 of the present invention;

[0022] FIG3 is a schematic diagram of a process for constructing a spatial relationship map annotation template disclosed in the first embodiment of the present invention;

[0023] FIG4 is a schematic diagram of annotating a map base map using three types of map annotation templates disclosed in the first embodiment of the present invention;

[0024] FIG5 is a flowchart of a user performing customized map annotation according to the first embodiment of the present invention;

[0025] FIG6 is a labeled map generated for news on the Russia-Ukraine conflict using a method disclosed in Example 1 of the present invention for automatically generating map annotations based on text using a large language model. DETAILED DESCRIPTION

[0026] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0027] Example 1

[0028] As shown in FIG1 , an embodiment of the present invention provides a method for automatically generating map annotations based on text using a large language model. The method includes the following steps:

[0029] S1. Construct a map annotation template based on the deconstructed map annotation design space.

[0030] As shown in Figure 2, the map annotation design space includes two important components, namely geographic element dependency and annotation form. Geographic element dependency includes the objects of map annotation, that is, the element content that actually exists in the map; while the annotation form and the data mapping on its attributes reflect the annotation method and the information content of the annotation. Among them, geographic elements include three basic element types, namely points, lines and surfaces; basic geographic elements can be further composed of more complex geographic elements. Figure 2 shows a schematic diagram of the map annotation effect obtained when different annotation forms are used when the geographic element dependency is a single point, a single line, a single surface and two point-shaped geographic elements.

[0031] On the other hand, annotation styles also include three basic types: point, line, and surface. Basic annotation styles can be further combined into more complex annotation objects. Different annotation styles can support mapping data information to different attribute channels.

[0032] Based on the deconstructed map annotation design space, simple map annotations can be combined to form more complex annotation forms, so as to construct map annotation templates to meet the richer and more complex annotation requirements in actual tasks. The composite map annotation template has clear annotation semantics and is pre-set with corresponding annotation design details. The annotation design details are mainly reflected in the specific values ​​of the visual channel, such as the actual color on the color channel and the specific selected primitives on the primitive channel. These templates provide the necessary parameter interfaces, allowing annotations to be applied in different scenarios to meet the corresponding annotation requirements. In an embodiment of the present invention, the constructed map annotation template includes an element state map annotation template, a spatial relationship map annotation template, and a trend information map annotation template. The element state map annotation template is used to annotate the element state of a single geographic element, the spatial relationship map annotation template is used to annotate the spatial relationship between two geographic elements, and the trend information map annotation template is used to annotate the trend information between two geographic elements. The geographic element dependency or status information in the map annotation template is adjusted according to the received input parameters.

[0033] The element state map annotation template contains a geographic element dependency and a state message, adjusting all parameters except the geographic element dependency based on the received input. For point-like geographic elements, the element content placed at the corresponding point-like geographic element position is adaptively modified based on the input state attributes to construct the map annotation. For linear and area-like geographic elements, the stroke and fill attributes of the relevant geographic elements are modified based on the input state attributes to complete the map annotation.

[0034] Spatial relationship map annotation templates include two geographic element dependencies as well as spatial distance and orientation relationships. This type of template is based on linear annotation arrows and provides text annotations along the arrows for corresponding distance and spatial orientation relationships. Different types of dependent geographic elements will affect the location and method of drawing arrows. Point-like geographic elements will be directly used as the starting or ending point of the arrow annotation; linear geographic elements will select a point on the line as the starting or ending point of the arrow annotation; for surface elements, the template will select the center of gravity of the enclosed area as the starting or ending point of the arrow annotation.

[0035] The spatial relationship map annotation template requires two point geographic element dependencies and four geographic information parameters. By passing the corresponding parameters, the combined template can contain simpler templates. As shown in Figure 3, the template semantics is to annotate the spatial relationship between two point geographic elements. It requires two point geographic element dependencies and four geographic information parameters, including the labels of the two point geographic elements, and the distance and orientation relationship between them. The spatial relationship map annotation template consists of three relatively simple sub-templates, which have two annotation semantics: highlighting a single point geographic element and drawing a labeled arrow between two point geographic elements. The parameter input required by the sub-template is completely provided by its parent template, which is specifically shown on the arrow in the figure. The sub-template is composed of more basic templates, as shown in the template on the far right of Figure 3. Through a similar method, the combination of geographic elements and annotation forms is completed, thereby constructing an annotation template with more complex semantics and structure.

[0036] The Trend Information Map Annotation Template contains information about both the geographic element dependencies and the trend type. Similar to the Spatial Relationship Map Annotation Template, the Trend Information Map Annotation Template also uses linear arrows as its basis. To distinguish it from the Spatial Relationship Map Annotation Template, the Trend Information Map Annotation Template uses a different arrow style and sets the arrow color based on the trend type information. The arrows for trend annotations are positioned in the same way as those for spatial relationship annotations. Depending on the type of geographic element they rely on, an appropriate anchor point is selected as the arrow's start or end point.

[0037] S2. Use a large language model to extract geographic information from text.

[0038] Large language models can support tasks requiring small or even zero samples, making them suitable for extracting rich and complex geographic information from text. In an embodiment of the present invention, a large language model prompt word template is constructed to meet the requirements for generating large language model prompt words for different extraction tasks. By embedding the text to be annotated, sample text, and sample output into the prompt word template, the prompt words required for extracting geographic information from text content can be generated. By parsing the results returned by the large language model, the desired geographic information content can be obtained.

[0039] S3. Automatically generate an annotated map based on the extracted geographic information content and a map annotation template.

[0040] As shown in Figure 4, after obtaining the geographic information from the text, the corresponding map annotation template is matched based on the geographic elements and geographic information data contained in the information. In other words, a map basemap is constructed based on the geographic information data. Subsequently, the relevant parameter information, including the dependent geographic elements and geographic information data, is passed to the map annotation template parameters through the large language model, and the corresponding map annotation content is constructed according to the template. After combining the annotations corresponding to all the extracted geographic information, the final annotated map is obtained.

[0041] As shown in FIG5 , users can adjust the annotation information and annotation method by changing the output of the large language model or by changing the parameters of the map annotation template that matches the large language model to complete a customized annotation design.

[0042] As shown in FIG6 , a method for automatically generating map annotations based on text using a large language model disclosed in an embodiment of the present invention is used to generate a set of annotated maps based on news about the Russia-Ukraine conflict, where (0) is the base map and (1-11) are the generated annotated maps.

[0043] It can be seen from the above embodiments that the present invention discloses a method for automatically generating map annotations based on text using a large language model, explores and deconstructs the design space of map annotations, and designs a set of standard grammars for generating map annotations that can meet a wide range of map annotation tasks to support the annotation requirements of different geographic information; and by introducing an automated method, simplifies the creative process of users generating corresponding map annotations, thereby lowering the threshold for creating annotated maps.

[0044] The method described in the present invention is not limited to the embodiments described in the specific implementation manner. Those skilled in the art may derive other implementation manners based on the technical solution of the present invention, which also fall within the scope of the technical innovation of the present invention.

Claims

1. A method for automatically generating map annotations based on text using a large language model, the method comprises the following steps: S1. Construct a map annotation template according to the deconstructed map annotation design space; S2. Use the large language model to extract the geographical information content in the text to be annotated; S3. Based on the extracted geographical information content, automatically generate an annotated map based on the map annotation template.

2. The method for automatically generating map annotations based on text using a large language model as claimed in claim 1, characterized in that: The method further includes adjusting the annotation information and annotation method by changing the output of the large language model to complete customized map annotation.

3. The method for automatically generating map annotations based on text using a large language model as claimed in claim 1, characterized in that: The method further includes adjusting the annotation information and annotation method by changing the parameters of the large language model for matching the map annotation template to complete customized map annotation.

4. The method for automatically generating map annotations based on text using a large language model as claimed in claim 1, characterized in that: In step S1, the map annotation design space includes geographical element dependencies and annotation forms. The geographical element dependencies include the geographical elements existing in the map, and the data mapping on the annotation form and its attributes reflects the annotation method and annotation information.

5. The method for automatically generating map annotations based on text using a large language model as claimed in claim 1, characterized in that: In step S1, simple map annotations are combined to construct a map annotation template.

6. The method for automatically generating map annotations based on text using a large language model as claimed in claim 1, characterized in that: In step S1, the map annotation template adjusts the geographical element dependencies or status information in the map annotation template according to the received input parameters.

7. The method for automatically generating map annotations based on text using a large language model as claimed in claim 1, characterized in that: In step S1, the map annotation template includes an element status map annotation template, a spatial relationship map annotation template, and a trend information map annotation template. The element status map annotation template is used to annotate the element status of a single geographical element, the spatial relationship map annotation template is used to annotate the spatial relationship between two geographical elements, and the trend information map annotation template is used to annotate the trend information between two geographical elements.

8. The method for automatically generating map annotations based on text using a large language model as claimed in claim 1, characterized in that: In step S2, a large language model prompt template is constructed to meet the generation requirements of large language model prompts for different extraction tasks.

9. The method for automatically generating map annotations based on text using a large language model as claimed in claim 8, characterized in that: In step S2, by embedding the text to be annotated, sample text, and sample output into the prompt template, the prompt words required for the extraction task corresponding to the geographical information in the text content can be generated, that is, the geographical information content in the text can be extracted.

10. A method for automatically generating map annotations based on text using a large language model as described in claim 1, characterized in that: In step S3, according to the geographical information content extracted from the letter, the corresponding map annotation template is matched, and the required input parameters are input into the corresponding map annotation template through the large language model, so as to automatically complete the map annotation.

Images