Electromagnetic spectrum game situation extraction method based on large language model

Abstract

The invention provides an electromagnetic spectrum game situation extraction method based on a large language model, and belongs to the technical field of crossing of electromagnetic spectrum management and artificial intelligence. The method comprises the following steps: constructing a service module based on a large language model by utilizing a Flask framework; constructing an input interface of a user instruction and an input interface of contents needing to be extracted, and generating a query statement according to the user instruction and the contents needing to be extracted; a query statement is sent in an HTTP request mode, and a character string reply containing situation information is returned from the service module; analyzing and extracting the situation data by using a natural language processing technology, and establishing a data dictionary; constructing an interface correction function for a user to intuitively check the data dictionary and modify data in the data dictionary; and drawing the electromagnetic spectrum game situation according to the data dictionary. According to the method, by introducing the large language model, efficient and accurate entity relationship extraction of electromagnetic spectrum game situation text description data is realized, and powerful support is provided for fine management of the electromagnetic spectrum.

Description

Technical Field

[0001] The present invention relates to the cross-technical field of electromagnetic spectrum management and artificial intelligence, and specifically to an electromagnetic spectrum game situation extraction method based on a large language model. Background Art

[0002] In electromagnetic spectrum management, accurately extracting and analyzing spectrum usage trends is of great significance for optimizing spectrum resource allocation, improving spectrum utilization efficiency, and ensuring frequency safety. However, traditional spectrum situation analysis methods often rely on manual interpretation and simple data processing, which makes it difficult to efficiently process large-scale and complex spectrum data. Especially when spectrum data exists in the form of text descriptions, how to automatically and accurately extract the entity relationships therein becomes an urgent problem to be solved.

[0003] At present, the use of artificial intelligence methods to extract required content has become the mainstream method. In the patent "Text Extraction Method, Device and Electronic Equipment" applied by Zhuo Shi Zhixing (Qingtian) Metaverse Technology Co., Ltd., with the authorization announcement number CN118522017B, for the image to be extracted, the specified text is realized through OCR recognition, segmentation, conversion model, verification model and mapping model. However, in electromagnetic spectrum management, compared with the expression of images, more parameter information such as frequency equipment type, power range, etc. needs to be extracted from text information.

[0004] In the patent "Text Extraction Method, Device, Electronic Device and Storage Medium" applied by China Mobile (Suzhou) Software Technology Co., Ltd. with the authorization announcement number CN118171648B, the text information to be identified is segmented, and the classification features of each segmentation are calculated, and finally the target text is generated based on all the segmentations. However, the implementation method of this method is too complicated, and the steps of directly extracting the required content are relatively cumbersome.

[0005] In the patent "A method, device, equipment and medium for visualizing electromagnetic situation numerical values" applied by the National University of Defense Technology of the People's Liberation Army of China and authorized with the announcement number CN117390849B, the electromagnetic situation request data is parsed, the parsed data is obtained, and the situation information grid is drawn based on the parsed data. However, this method requires that the data information provided must comply with the specification, and the situation description often uses the language text method, so this method has strict requirements on the way of inputting content and has low applicability.

[0006] In the article "Intelligent Assisted Argumentation Method Based on Large Language Model" published in Radio Engineering by Chang Xin et al., it is proposed to use large language model to extract key knowledge and scenarios from argumentation ideas and complete experimental data drawing. However, in the summary of the interaction process with the large language model, more human participation is required to complete the transformation from text to program. Summary of the invention

[0007] In view of this, the present invention proposes a method for extracting electromagnetic spectrum game situation based on a large language model. Aiming at the problem that more human participation is required in the process of constructing electromagnetic spectrum situation information using a large language model, the present invention constrains the standardization of the content extracted by the large language model through user instructions, generates a data dictionary with specified standards, and thus provides a good interactive basis for automated situation extraction and display. In addition, in order to prevent errors in the extraction of the large language model, the present invention designs a correction function for people in the loop to improve the accuracy of the extraction.

[0008] The technical solution adopted by the present invention is:

[0009] A method for extracting electromagnetic spectrum game situation based on a large language model comprises the following steps:

[0010] Step 1: Use the Flask framework to build a service module based on a large language model;

[0011] Step 2: Build an input interface for user instructions and an input interface for the content to be extracted, and generate query statements based on the user instructions and the content to be extracted;

[0012] Step 3: Send a query statement in the form of an HTTP request, and the service module returns a string reply containing situation information;

[0013] Step 4: Use natural language processing technology to parse and extract situation data and build a data dictionary;

[0014] Step 5: Build an interface-based correction function to allow users to visually check the data dictionary and modify the data in the data dictionary;

[0015] Step 6: Draw the electromagnetic spectrum game situation based on the data dictionary.

[0016] Furthermore, the specific method of step 1 is:

[0017] Use the Flask framework to build an API interface to receive HTTP requests and pass query statements.

[0018] Furthermore, the specific method of step 2 is:

[0019] Step 2-1: Construct a Prompt text box for inputting user instructions and a Content text box for inputting the content to be extracted; construct a text adjustment module for dynamically adjusting the size of the text box;

[0020] Step 2-2: Build an event binding module, associate keyboard events with the text adjustment module, and trigger real-time adjustment of the width and height of the text box;

[0021] Step 2-3: Build a submit button to concatenate the user instructions entered into the Prompt text box and the content to be extracted in the Content text box to form a query statement;

[0022] Step 2-4: Enter the user command into the Prompt text box, enter the content to be extracted into the Content text box, click the Submit button, and concatenate the user command and the content to be extracted into a query statement.

[0023] Furthermore, the specific method of step 3 is:

[0024] Step 3-1: Start the service module based on the large language model;

[0025] Step 3-2: Send the query statement to the service module via HTTP POST request;

[0026] Step 3-3: Input the text of the query statement into the large language model;

[0027] Step 3-4: Generate a string containing detailed information about the electromagnetic spectrum game situation after processing by the large language model;

[0028] Step 3-5: Return a string response containing detailed information about the electromagnetic spectrum game situation via HTTP;

[0029] Step 3-6: A string reply containing status information is returned from the service module.

[0030] Furthermore, the specific method of step 4 is:

[0031] Step 4-1: Define a regular expression for extracting data dictionary;

[0032] Step 4-2: Use the standard library re module specifically used for processing regular expressions to find matching items from the string reply containing situation information;

[0033] Step 4-3: Extract the wrapped dictionary string;

[0034] Step 4-4: Remove the blank characters before and after the dictionary string;

[0035] Step 4-5: Delete line breaks, carriage returns, and spaces;

[0036] Step 4-6: Use the JSON lightweight data exchange format module to convert the dictionary string into a Python dictionary object.

[0037] Furthermore, the specific method of step 5 is:

[0038] Step 5-1: Build a pop-up window module, which includes an interface for displaying nested dictionaries and supports generating input boxes with different indentations according to the nesting level;

[0039] Step 5-2: Generate corresponding input boxes based on the key-value pairs of the dictionary structure. When encountering a nested dictionary, recursively generate a new input box, and display the indentation structure on the interface to indicate the nesting level. Indentation or layering is used on the interface to indicate different levels of the nested dictionary, so that users can intuitively distinguish the nested structure.

[0040] Step 5-3: Match the data structure edited by the user on the interface with the original dictionary structure, store the modified nested structure hierarchically into the data dictionary and save it.

[0041] Furthermore, the specific method of step 6 is:

[0042] Step 6-1: Draw the electromagnetic situation scenario range;

[0043] Step 6-2: Draw the position and shape of each entity;

[0044] Step 6-3: Draw the capabilities of the equipment carried by each entity and label them.

[0045] The beneficial effects of the present invention are:

[0046] 1. By introducing a large language model, the present invention realizes efficient and accurate entity relationship extraction of electromagnetic spectrum game situation text description data, providing strong support for the refined management of the electromagnetic spectrum.

[0047] 2. The present invention not only improves the efficiency of spectrum management, but also enhances the accuracy and reliability of data processing, and has important practical value and application prospects.

[0048] 3. The present invention realizes the display function of the electromagnetic spectrum game situation based on the data dictionary. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention.

[0050] Figure 2 It is the input interface for user instructions and content to be extracted.

[0051] Figure 3 It is the interface after inputting user instructions and content to be extracted.

[0052] Figure 4 This is a nested structure diagram of electromagnetic spectrum game situation information.

[0053] Figure 5 The storage completion prompt interface is displayed.

[0054] Figure 6 This is a map of the electromagnetic spectrum game situation. DETAILED DESCRIPTION

[0055] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0056] A method for extracting electromagnetic spectrum game situation based on a large language model, the overall process is as follows: Figure 1 The specific steps include:

[0057] Step 1: Use the Flask framework to build a service module based on a large language model; the specific method is:

[0058] Build an API interface. Use the Flask framework to build an API interface that can receive HTTP requests and pass query statements. The query statements are natural language text generated by the user through the input interface.

[0059] Introduce the Flask framework through the following statement:

[0060] from flask import Flask,request,jsonify

[0061] Create a Flask application using the following statement:

[0062] app = Flask(__name__)

[0063] Create a chat interface using the following statement:

[0064] @app.route(' / chat',methods=['POST'])

[0065] The chat interface can utilize HTTP POST method to transmit the extracted content.

[0066] Use defchat() to define the extraction function, use request.json.get('message',") to get the question content entered by the user from the request body and store it in user_input.

[0067] Step 2: Build the input interface for user instructions and content to be extracted and the query statement generation function;

[0068] Step 2-1: Construct the user command Prompt text box and the Content text box to be extracted, which are used to accept the string entered by the user and dynamically adjust the size of the input text box. Use the tkinter module to design the input interface. First, create a "Prompt" input box. Use tk.Label to create a label that displays the text "Prompt:", and place it at row 0 and column 0, set the margins padx and pady to 10, and align the text to the left (sticky = 'w'). Use tk.Text to create a multi-line text input box with an initial width of 40 and an initial height of 1 (only one line is displayed) for users to enter the "Prompt" content. The input box is placed at row 0 and column 1, and the margins are set. Bind the input box <keyrelease>Event, when the event is triggered, call the adjust_textbox_size method to dynamically adjust the size of the text box to fit the input content. Then, create the "Content" input box. Similar to "Prompt", use tk.Label to create a label "Content:" and place it at row 1, column 0. Use tk.Text to create another multi-line text input box for entering the relevant content of "Content" and place it at row 1, column 1. The width and height settings are the same as those of the "Prompt" input box. Similarly bind <keyrelease>Event, dynamically adjust the size of the input box.

[0069] Step 2-2: Build an event binding module, associate keyboard events with the text adjustment module, and trigger real-time adjustment of the width and height of the text box. It is used to change the width and height of the text box in real time according to the content entered by the user. Every time the user enters or deletes content in the text box, this method will be triggered. First, the text box component that triggers the event is obtained, and then the content in the text box is extracted and divided into a list by line. Next, the length of each line of text is calculated, and within the set maximum width (40 characters), the length of the longest line is selected as the new width of the text box. At the same time, the height of the text box is adjusted according to the total number of lines of text, but for the sake of the beautiful interface, the maximum height is limited to 10 lines. Finally, the textbox.config() method is used to update the width and height configuration of the text box to ensure that the text box can always display the content entered by the user without excessive blank area or content overflow.

[0070] Step 2-3: Construct a submit button to concatenate the user instructions entered into the Prompt text box and the content to be extracted from the Content text box to form a query statement. Create a submit button. Use tk.Button to create a button that displays the text "Submit". When the button is clicked, the show_popup method is called to process the user's submission operation. The button is placed in the 2nd row and 1st column, right-aligned (sticky = 'e'), and padding is set to 20 to keep the button and the upper component at an appropriate distance. The input interface for user instructions and content to be extracted is as follows: Figure 2 shown.

[0071] Step 2-4: After clicking the submit button, the user instruction entered in the Prompt text box and the content to be extracted in the Content text box are concatenated to form a query statement. The user instruction is to generate data in the format of a Python dictionary based on the content, without any explanations or annotation information, in the format of {"red":{"platform_red_1":{"camp":"red","location":[0,0],"unit":"km","equipment":{"radar":{"range":5,"unit":"km"},"communication":{"range":2,"unit":"km"}}}}}. In addition to constraining the content in the user instruction, the format is also constrained. The content to be extracted is that in the scenario, there are red and blue camps. The red camp includes Platform Red 1 and Platform Red 2. Platform Red 1 is located at (0km, 0km), and Platform Red 2 is located at (1km, 0km). Platform Red 1 is equipped with a radar and a wireless communication device, and Platform Red 2 is equipped with a wireless communication device. The power range of the red radar is 5km, and the effective range of the wireless communication device is 2km. There is also Platform Blue 1, located at (5km, 5km). Platform Blue 1 is equipped with a radar countermeasure device and a communication countermeasure device. The power range of the radar countermeasure device is 3km, and the position range of the communication countermeasure device is 6km. Both are in string form and are linked together using the ":" symbol to form the query statement "Generate data in the format of a Python dictionary based on the content, without any explanations or annotation information, in the format of {"red":{"platform_red_1":{"camp":"red","location":[0,0],"unit":"km","equipment":{"radar":{"range":5,"unit":"km"},"communication":{"range":2,"unit":"km"}}}}}: In the scenario, there are red and blue camps. The red camp includes Platform Red 1 and Platform Red 2. Platform Red 1 is located at (0km, 0km), and Platform Red 2 is located at (1km, 0km). Platform Red 1 is equipped with a radar and a wireless communication device, and Platform Red 2 is equipped with a wireless communication device. The power range of the red radar is 5km, and the effective range of the wireless communication device is 2km. There is also Platform Blue 1, located at (5km, 5km). Platform Blue 1 is equipped with a radar countermeasure device and a communication countermeasure device. The power range of the radar countermeasure device is 3km, and the position range of the communication countermeasure device is 6km."

[0072] The interface after the input of the user instruction and the content to be extracted is as Figure 3 shown.

[0073] Step 3: Send a query statement in the form of an HTTP request, and the service module returns a string reply containing situation information;

[0074] Step 3-1: Start the service module based on the large language model. Run the main function to start the Flask application that hosts the large language model if__name__ == "__main__": app.run(host = '0.0.0.0', port = 5000). Because it is used on the local machine, the IP address is '0.0.0.0' and the port number is 5000.

[0075] Step 3-2: Send the query statement to the service module via HTTP POST request. Construct request data headers = {'Content-Type':'application / json'} and payload = {"message":message}, where message is the query statement, and send the message to the Flask server by sending a POST request, i.e. requests.post(URL, headers = headers, data = json.dumps(payload)), where URL is the address URL of the Flask server = http: / / 127.0.0.1:5000 / chat.

[0076] Step 3-3: Input the natural language text into the large language model. This embodiment adopts the open source large language model GLM-4-9B released by Zhipu Qingyan. In the open source code released by Zhipu Qingyan, the original contents of building historical dialogues, updating historical dialogues, loading word segmenters, loading models and custom stop classes are retained, and its quantitative reasoning mode is set to 4bit mode, which corresponds to the quantization_config=BitsAndBytesConfig(load_in_4bit=True) instruction. In addition, in order to prevent the historical information of the large language model GLM-4-9B from being compared with this extraction and causing deviations, the history variable of the updated historical record will be set to empty after each answer is completed.

[0077] Step 3-4: After processing by the large language model, a string containing detailed information on the electromagnetic spectrum game situation is generated.

[0078] Set generated_text = "" as the space to store generated text, and put the large language model GLM-4-9B generated text streamer into generated_text one by one:

[0079] for new_token in streamer:

[0080] ifnew_token:

[0081] generated_text+=new_token

[0082] Step 3-5: Return a string reply containing detailed information about the electromagnetic spectrum game situation via HTTP. jsonify is a tool function provided by Flask. It uses jsonify to convert a Python dictionary or list into an HTTP response in JSON format, and removes blank characters (such as spaces, line breaks, etc.) at the beginning and end of the text generated by the large language model through the .strip() method. In addition, add a status code of 200, indicating that the request is successful, that is, returnjsonify({"response":generated_text.strip()}),200.

[0083] Step 3-6: A string reply containing status information is returned from the service module.

[0084] Monitor the reply message response of requests.post(URL, headers = headers, data = json.dumps(payload)). When the status code of the reply message response is 200, that is, response.status_code == 200, parse to obtain a string reply containing status information, that is, response_data = response.json().

[0085] Step 4: Use natural language processing technology to parse and extract situation data and build a data dictionary;

[0086] Step 4-1: Define regular expressions. Experiments show that the above prompt will cause the large language model to save the data dictionary in specific text characters, so a regular expression is designed to use DOTALL in the re module to match multiple lines and extract them, that is, pattern = r""'python(.*?)"'".

[0087] Step 4-2: Use the standard library re module, which is specifically used to process regular expressions, to find matching items. Search for matching items in response_data, re.search(pattern,response_data,re.DOTALL), and save the results in the match variable.

[0088] Step 4-3: Extract the wrapped dictionary string. Execute match.group(1) and save the result in dict_json_str.

[0089] Step 4-4: Remove the whitespace characters before and after the dictionary string. Further execute dict_json_str.strip(), and overwrite and save it.

[0090] Step 4-5: Delete newline characters, carriage return characters, and spaces. Further execute dict_json_str = dict_json_str.replace("\n","").replace("\r","").replace("",""), and overwrite and save it.

[0091] Step 4-6: Use the JSON lightweight data interchange format module to convert the dictionary string into a Python dictionary object. That is, execute json.loads(dict_json_str), and save the dictionary object into dict_data. The extracted data dictionary is: {'blue':{'platform_blue_1':{'camp':'blue','location':[5,5],'unit':'km','equipment':{'radar_countermeaure':{'range':3,'unit':'km'},'communication_countermeaure':{'range':3,'unit':'km'}}}},'red':{'platform_red_1':{'camp':'red','location':[0,0],'unit':'km','equipment':{'radar':{'range':5,'unit':'km'},'communication':{'range':2,'unit':'km'}}},'platform_red_2':{'camp':'red','location':[1,0],'unit':'km','equipment':{'communication':{'range':2,'unit':'km'}}}}}

[0092] Step 5: Build an interface correction function that allows users to visually check the data dictionary and modify the data in the data dictionary;

[0093] Step 5-1: Build a pop-up window module that includes an interface for displaying nested dictionaries and supports generating input boxes with different indents according to the nesting levels.

[0094] Step 5-2: Generate corresponding input boxes based on the key-value pairs of the dictionary structure. When encountering a nested dictionary, recursively generate a new input box, and display the indentation structure on the interface to indicate the nesting level. Indentation or layering is used to indicate different levels of nested dictionaries on the interface so that users can intuitively distinguish the nested structure, such as Figure 4 shown.

[0095] Step 5-3: Match the data structure edited by the user on the interface with the original dictionary structure, store the modified nested structure hierarchically in the data dictionary and save it, such as Figure 5 As shown;

[0096] Step 6: Draw the electromagnetic spectrum game situation based on the data dictionary, such as Figure 6 shown.

[0097] Step 6-1: Draw the electromagnetic situation scenario range.

[0098] Step 6-2: Draw the position and shape of each entity.

[0099] Step 6-3: Draw the capabilities of the equipment carried by each entity and label them.

[0100] In summary, the present invention cleverly combines the natural language processing capabilities of a large language model, the lightweight Web service technology of the Flask framework, and the precise reply language processing technology. By constructing a user-friendly input interface, the present invention can receive and splice user instructions and the content to be extracted, and then generate a complete query statement. The Flask framework is used to build a service based on a large language model, receive and process these statements in the form of HTTP requests, and return a string reply containing situation information. It is worth noting that dictionary data is embedded in the reply, and through regular expressions and JSON parsing technology, these dictionary data can be automatically extracted and converted into Python dictionary objects, which is convenient for subsequent data processing and analysis. In addition, the present invention also provides an interface correction function, allowing users to intuitively view and modify the extracted data, ensuring the accuracy and reliability of the data. The present invention realizes the extraction of key parameters of the electromagnetic spectrum game situation from the text, and saves it as dictionary data, which can be used for the realization of functions such as electromagnetic spectrum game situation drawing. Through the present invention, the extraction process of the electromagnetic spectrum game situation is highly automated and intelligent, providing strong support for the refined management and decision-making of the electromagnetic spectrum.< / keyrelease> < / keyrelease>

Claims

1. A method for extracting electromagnetic spectrum game situation based on a large language model, characterized in that: The following steps are involved: Step 1: Use the Flask framework to build a service module based on a large language model; Step 2: Build an input interface for user instructions and an input interface for the content to be extracted, and generate query statements based on the user instructions and the content to be extracted; Step 3: Send a query statement in the form of an HTTP request, and the service module returns a string reply containing situation information; Step 4: Use natural language processing technology to parse and extract situation data and build a data dictionary; Step 5: Build an interface-based correction function to allow users to visually check the data dictionary and modify the data in the data dictionary; Step 6: Draw the electromagnetic spectrum game situation based on the data dictionary.

2. According to the method for extracting electromagnetic spectrum game situation based on large language model in claim 1, it is characterized in that: The specific method of step 1 is: Use the Flask framework to build an API interface to receive HTTP requests and pass query statements.

3. According to the method for extracting electromagnetic spectrum game situation based on large language model in claim 1, it is characterized in that: The specific method of step 2 is: Step 2-1: Construct a Prompt text box for inputting user instructions and a Content text box for inputting the content to be extracted; construct a text adjustment module for dynamically adjusting the size of the text box; Step 2-2: Build an event binding module, associate keyboard events with the text adjustment module, and trigger real-time adjustment of the width and height of the text box; Step 2-3: Build a submit button to concatenate the user instructions entered into the Prompt text box and the content to be extracted in the Content text box to form a query statement; Step 2-4: Enter the user command into the Prompt text box, enter the content to be extracted into the Content text box, click the Submit button, and concatenate the user command and the content to be extracted into a query statement.

4. According to the method for extracting electromagnetic spectrum game situation based on large language model in claim 1, it is characterized in that: The specific method of step 3 is: Step 3-1: Start the service module based on the large language model; Step 3-2: Send the query statement to the service module via HTTP POST request; Step 3-3: Input the text of the query statement into the large language model; Step 3-4: Generate a string containing detailed information about the electromagnetic spectrum game situation after processing by the large language model; Step 3-5: Return a string response containing detailed information about the electromagnetic spectrum game situation via HTTP; Step 3-6: A string reply containing status information is returned from the service module.

5. The electromagnetic spectrum game situation extraction method based on a large language model according to claim 1 is characterized in that: The specific method of step 4 is: Step 4-1: Define a regular expression for extracting data dictionary; Step 4-2: Use the standard library re module specifically used for processing regular expressions to find matching items from the string reply containing situation information; Step 4-3: Extract the wrapped dictionary string; Step 4-4: Remove the blank characters before and after the dictionary string; Step 4-5: Delete line breaks, carriage returns, and spaces; Step 4-6: Use the JSON lightweight data exchange format module to convert the dictionary string into a Python dictionary object.

6. The electromagnetic spectrum game situation extraction method based on a large language model according to claim 1 is characterized in that: The specific method of step 5 is: Step 5-1: Build a pop-up window module, which includes an interface for displaying nested dictionaries and supports generating input boxes with different indentations according to the nesting level; Step 5-2: Generate corresponding input boxes based on the key-value pairs of the dictionary structure. When encountering a nested dictionary, recursively generate a new input box, and display the indentation structure on the interface to indicate the nesting level. Indentation or layering is used on the interface to indicate different levels of the nested dictionary, so that users can intuitively distinguish the nested structure. Step 5-3: Match the data structure edited by the user on the interface with the original dictionary structure, store the modified nested structure hierarchically into the data dictionary and save it.

7. The electromagnetic spectrum game situation extraction method based on a large language model according to claim 1 is characterized in that: The specific method of step 6 is: Step 6-1: Draw the electromagnetic situation scenario range; Step 6-2: Draw the position and shape of each entity; Step 6-3: Draw the capabilities of the equipment carried by each entity and label them.

Images