User-centered big language model-driven text viewpoint mining method and system

Abstract

The invention discloses a user-centered big language model driven text viewpoint mining method and system, and the method comprises the steps: 1, generating a user portrait through a big language model LLM according to a to-be-detected text; 2, generating a context scene through a large language model (LLM) according to a to-be-detected text; 3, using a large language model LLM to play a user as a user agent, inputting a user portrait and a context scene into the user agent to simulate an interaction process of the user in the scene, and mining user viewpoints; 4, a large language model LLM is adopted to play a social media expert and a manager to serve as an expert agent and a management agent respectively, the management agent judges the viewpoint of the user agent, and the final viewpoint is output in combination with discussion results of the expert agent and the user agent. From the perspective of the user, the text viewpoint mining capability of the large language model is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of text opinion mining, and in particular relates to a user-centric large language model driven text opinion mining method and system. Background Art

[0002] In the digital age, social media has become an integral part of people's lives. According to a 2025 report released by the media monitoring company Meltwater, the number of active social media users worldwide has exceeded 5.04 billion, representing approximately 62.3% of the global population. With the massive amount of textual information on social media platforms, uncovering the underlying insights is crucial for public opinion monitoring, personalized user services, and the dissemination of social information.

[0003] Opinion mining is a core technology in natural language processing. Its goal is to extract and analyze users' subjective opinions and related elements, such as sentiment, stance, and sarcasm, from unstructured text. The current mainstream research paradigm is text-centric, essentially focusing on analyzing subjective elements in text content to construct a mapping system from raw data to opinions. However, from a social psychology perspective, user opinion formation is described as "an individual's tendency to evaluate a specific object." This current text-centric paradigm focuses solely on the object of evaluation and evaluation tendencies, neglecting the subject of opinion formation—the user. Humans communicate through symbols in social interactions, and the meaning of symbols is generated during this interaction. Users with different experiences, cultural backgrounds, and behavioral preferences may have very different views on the same sentence. Existing models lack the user's perspective when mining opinions, resulting in poor performance when analyzing text with similar features. Furthermore, text-centric models can only be applied to data in specific domains, and test cases must be highly relevant to the training data. Therefore, it is highly necessary to develop a universal, user-centric approach to text opinion mining. Summary of the Invention

[0004] One purpose of the present invention is to address the shortcomings of the existing technology and provide a user-centric large language model driven text opinion mining method. This method effectively improves the large language model's ability to mine text opinions from the user's perspective.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions: A user-centric large language model driven text opinion mining method includes the following steps: Step 1: Generate a user profile using the Large Language Model (LLM) based on the text to be tested. Step 2: Generate context scenarios using the large language model (LLM) based on the text to be detected. Step 3: Use the Large Language Model (LLM) to act as a user agent. Input the user profile and context into the user agent to simulate the user's interaction process in the scenario and mine the user's opinions. Step 4: Use the large language model (LLM) to act as the expert agent and the management agent. The management agent evaluates the opinions of the user agent and outputs the final opinion based on the discussion results of the expert agent and the user agent.

[0006] Furthermore, the implementation method of step 1 includes: Collect fine-tuning data for initializing the white-box LLM-type portrait attribute generator; Fine-tune the white-box LLM user profile attribute generator based on the data collected above; The white-box LLM profile attribute generator generates the user's key attributes based on the text to be tested and transmits them to the black-box LLM user profile generator to generate a user profile.

[0007] Furthermore, the Bayesian optimization method is used to optimize the collaboration process between the user profile attribute generator based on the white-box LLM and the user profile generator based on the black-box LLM.

[0008] Furthermore, the Bayesian hint optimization method specifically includes: 1) Randomly initialize n groups of soft tokens , concatenate each group of soft tokens with the embedding of the user portrait attribute generator in turn to obtain n groups of user portrait attribute prompts; 2) The user profile generator generates a user profile based on the user profile attribute prompts and the text to be detected; 3) The large language model plays the role based on the generated user profile and mines the opinions of the text to be tested from the user's perspective, and calculates the accuracy and F1-score of the opinion mining; 4) Through Gaussian process The mapping function distribution is modeled, and the Gaussian process is specifically expressed as follows: Given a set of observation data , the Gaussian process hypothesis function The joint distribution of any finite set of points follows a multivariate normal distribution; for a new point , given the observation data , The posterior distribution of is:

[0009] in, It's a new point and the covariance vector between the observation points, yes The covariance matrix of ;function is the covariance function, is the observation vector, yes Identity matrix; therefore, ; 5) Determine the next potential optimal soft token through the expected improvement algorithm , which is expected to produce a better F1 score in the next iteration, the next soft token Determined by the following formula:

[0010] in is the probability density function, yes The value of After multiple rounds of iterations, the soft token with the highest F1 score is selected as the prefix of the white-box LLM to generate a user profile prompt, which is used as the input of the black-box LLM.

[0011] Furthermore, the method for generating the context scenario in step 2 is: Input instructions and prompts into the large language model. The instructions are to generate a scenario based on the text, and the prompts specifically describe the four aspects of the generated scenario: background, task description, event description, and event process.

[0012] Furthermore, the process of the user agent simulating the user and generating thinking opinions is expressed as: Initialize the user agent: initialize according to the user portrait generated in step 1, , It represents the base model of the user agent, and u represents the user portrait; Place the user agent in the context scene and think about the text to be detected; transfer the context scene and the text to be detected generated based on step 2 to the new environment E, which is represented by , t is the text to be detected, S is the context scene generated in step 2, and the user agent makes a decision based on the environment , C represents the thinking process of the user agent.

[0013] Furthermore, step 4 specifically includes: The management agent evaluates whether the user agent's thinking process and conclusions are reasonable, and directly outputs the results if they are reasonable; If the management agent believes that the user agent's thinking results and conclusions deviate from the text to be tested, the user agent's thinking results will be stored in the thinking pool and provided to the expert agent as a reference, and the expert agent will make a secondary evaluation; After the expert agent discusses with the user agent, the expert agent presents its own opinions and thinking process, and then the manager agent selects one from the user's opinion and the expert's opinion as the final opinion.

[0014] The present invention also provides a system for implementing the above-mentioned user-centric large language model driven text opinion mining method, comprising: The user portrait generation module generates a user portrait based on the text to be detected through the large language model (LLM); The context scene generation module is used to generate the context scene based on the text to be detected through the large language model LLM; The user opinion mining module uses the Large Language Model (LLM) to act as a user agent, inputs user profiles and context scenarios into the user agent, simulates the user's interaction process in the scenario, and mines user opinions; The final opinion generation module is used to use the large language model (LLM) to play the role of expert agent and management agent. The management agent judges the opinions of the user agent and outputs the final opinion based on the discussion results of the expert agent and the user agent.

[0015] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0016] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1) First, we leverage the LLM's rich knowledge and powerful reasoning capabilities to construct user portraits and scenarios behind the text. We then use the LLM to role-play users and uncover their perspectives on specific objects. This approach effectively enhances the model's ability to mine textual perspectives from the user's perspective, and by understanding textual perspectives from the user's perspective, it fills the gap in perspective mining research that lacks user elements. 2) In the user profile generation stage, this paper designs a black-box LLM collaborative strategy. By optimizing and aligning the black-box and white-box models through Bayesian hints, the paper combines the respective advantages of the two models to generate high-quality user profiles, avoiding the low-quality and repetitive role played by the LLM and solving the uncontrollable problem of the black-box model. In addition, this paper fills the gap in opinion mining research that lacks user factors. 3) The user-centric large language model-driven text opinion mining framework proposed in this invention performs well on datasets of multiple tasks, surpassing existing methods. At the same time, the present invention supports multiple subtasks and is a universal opinion mining architecture. Moreover, the framework proposed in this invention is a universal text opinion mining framework that does not rely on data from a specific domain and can identify various user opinions. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a method for text opinion mining driven by a user-centric large language model according to an embodiment of the present invention; Figure 2 Schematic diagram of the user portrait generation process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0021] The present invention will be further described below with reference to specific examples, but they are not intended to limit the present invention.

[0022] like Figure 1 As shown, the embodiment of the present invention discloses a user-centric large language model driven text opinion mining method, comprising the following steps: Step 1: Generate a user profile based on the text to be tested using a large language model. The generated user profile will serve as the benchmark for user role-playing. See Figure 2The core steps of this step include: collecting fine-tuning data for initializing the white-box LLM portrait attribute generator; fine-tuning the white-box LLM-based user portrait attribute generator; using Bayesian optimization to optimize the collaboration process between the white-box LLM-based user portrait attribute generator and the black-box LLM-based user portrait generator; generating a user profile corresponding to a specific input sample based on the optimized collaboration process.

[0023] In this embodiment, to collect fine-tuning data for initializing the user profile attribute generator, this embodiment uses GPT-4o to generate outputs for several randomly selected training samples in the format of <Instruction, Input, Output>. This process is repeated 50 times, and 10 high-quality <Instruction, Input, Output> samples are manually inspected and selected. These samples capture the key attributes describing the user and can cover different aspects of the user, such as values or spiritual beliefs and interests. This embodiment uses these 10 <Instruction, Input, Output> as seed examples and uses GPT-4o to generate 500 <Instruction, Input, Output> samples. The above sample data is used to fine-tune the white-box user profile attribute generator. This embodiment uses an efficient low-rank adaptation (LoRa) parameter fine-tuning method to fine-tune the white-box LLM profile attribute generator, enabling the white-box LLM to generate user profile attributes. This method is a conventional parameter fine-tuning method and will not be described in detail here.

[0024] In this embodiment, to ensure collaboration between the white-box LLM-based user profile attribute generator and the black-box LLM-based user profile generator, a Bayesian hint optimization method is designed. The user profile attribute generator optimizes a set of soft tokens connected to the text embedding. The optimization goal is to maximize the probability that the generated user profile reflects the true opinion. The Bayesian hint optimization method designed in this embodiment consists of five steps: 1) Randomly initialize n groups of soft tokens , concatenate each group of soft tokens with the embedding of the user portrait attribute generator in turn to obtain n groups of user portrait attribute prompts; 2) The user profile generator generates a user profile based on the user profile attribute prompts and the text to be detected; 3) The large language model plays the role of the generated user profile and mines the opinions of the text from the user's perspective. Based on the true labels of the training data, the accuracy rate and F1-score of the opinion mining are calculated. The accuracy rate and F1-score calculation formulas are: Accuracy = ; F1-score= ; Where, TP (True Positive): the number of samples that are actually positive and predicted to be positive; TN (True Negative): The number of samples that are actually negative and predicted to be negative; FP (False Positive): The number of samples that are actually negative but predicted to be positive; FN (False Negative): The number of samples that are actually positive but predicted to be negative; 4) Through Gaussian process The mapping function distribution is modeled, and the Gaussian process is specifically expressed as follows: Given a set of observation data , Gaussian process hypothesis function The joint distribution of any finite set of points follows a multivariate normal distribution. Specifically, for a new point , given the observation data , The posterior distribution of is:

[0025] in, It's a new point and the covariance vector between the observation points, yes The covariance matrix of .function is the covariance function, is the observation vector, yes The identity matrix. Therefore, ; 5) Determine the next potential optimal soft token through the expected improvement algorithm , which is expected to produce a better F1 score in the next iteration. The next soft token Determined by the following formula:

[0026] in, is the probability density function, yes value.

[0027] In this example, the white-box LLM selects the soft token with the highest F1 score after multiple iterations as the prefix of the white-box LLM to generate a user profile prompt. The black-box LLM then describes the user profile that best matches the text to be tested based on the prompt.

[0028] Step 2: Generate a context field using a large language model based on the text to be detected; In this step, contextual scenarios are generated by inputting instructions and prompts into the large language model. The instructions are to generate a scenario based on the text. The prompt specifically describes four aspects of scenario generation: background, task description, event description, and event process. Specifically, the background and task descriptions provide general information for constructing the scenario, while the event description and event process provide information about "what events occurred in the scenario that may have led the user to post this text." Generating a contextual scenario can be specifically represented as: providing the text to be detected t, the instructions i regarding the task, and the prompt p for the large language model. The instructions i and prompt p are manually written.

[0029] In this embodiment, instruction i is represented as: You are a social media expert, please generate scenarios based on context. Hint p is expressed as: -Background: The foundational setting of the scene, encompassing time, place, environment, and any static elements that establish the scene'sframework, mood, and potential for events. -Task Description: The specific objective or guideline for generating the scene, directing how to analyze text elements (eg, characters, setting, time) to construct a coherent and contextually relevant scene. -Event Description: A brief overview of the main event or coreconflict occurring in the scene, summarizing its essence and its connection to the user's text. Focuses on "what happened" and its relevance to the user's post. -Event Process: The detailed sequence of the event, including its cause, development, climax, and resolution. Reveals the event's full scope and its impact on the scene. The obtained scene S can be expressed as , Represented as a large model called.

[0030] Step 3: Use the Large Language Model (LLM) to act as the user agent. Input the user profile and context into the user agent to simulate the user's interaction process in the scenario and mine the user's opinions. In this step, the Large Language Model (LLM) serves as the agent's reasoning engine, simulating the user through the user profile to obtain an LLM-based user agent. The LLM-based user agent is required to consider the viewpoints in the text based on the user profile and context information. Specifically, instructions and prompts are input to the large language model. The instruction is "Act as the user based on the user profile and contextual scenario, and consider what the viewpoint of this text is?" The prompt details the user profile and contextual scenario generated in steps one and two. The large language model will provide results and thinking processes based on the above. The process of the LLM-based user agent simulating the user and generating thinking content can be specifically expressed as follows: 1) Initialize the user agent: initialize according to the user portrait generated in step 1, , It represents the base model of the user agent, and u represents the user portrait; 2) Place the user agent in a contextual scene and think based on the text to be detected; transfer the contextual scene and the text to be detected generated in step 2 to a new environment E, which is represented by , the user agent thinks based on the environment , C represents the thinking process of the user agent.

[0031] Step 4: Use the Large Language Model (LLM) to act as the expert agent and the management agent. The management agent evaluates the opinions of the user agent and outputs the final opinion based on the discussion results of the expert agent and the user agent. In order to enhance the robustness of opinion mining, this embodiment further introduces two LLM-based agents, an expert agent and a management agent. In this step, the large language model LLM serves as the agent's reasoning engine. The management is simulated through the management portrait to obtain the LLM-based management agent, and the expert is simulated through the expert portrait to obtain the expert agent based on the LLM. The expert agent, management agent, and user agent will interact spontaneously. Specifically, the management agent evaluates whether the user agent's thinking process and conclusion can be used as the final judgment. If the management agent passes the evaluation, the opinion will be output; if not, the user agent's thinking results will be stored in the thinking pool, and then the thinking process will be shared with the expert agent. After the expert agent discusses with the user agent, the expert agent will put forward its own opinions and thinking process; then, the management agent selects one from the user or expert's opinions as the final output. The interaction process is expressed as: ① The user agent obtains the thinking results and process C; ② The management agent determines whether the thinking results and process are reasonable and whether they can be used as the final output. If so, it directly outputs and ends the process. If not, it proceeds to step ③; ③ The management agent provides the thinking results and process of the user agent to the expert agent. The expert agent integrates the information and proposes its own thinking process and results.

[0032] ④ The expert agent completes a round of interaction with the user agent, readjusts the thinking process and results, and submits them to the management agent.

[0033] ⑤ The management agent combines the thinking results of the two people, obtains the final judgment result and outputs it.

[0034] Note: Users, experts, and management agents are all completed by GPT-4o as the base model.

[0035] In order to illustrate the effectiveness of this embodiment, a test set Dtest is selected to obtain the opinion mining results of each text to be detected in the test set. For stance detection, this example selects P-stance and Sem16 as test sets, for hate detection, HateXplain is selected as the test set, and for sentiment analysis, rest16 and laptop14 are selected as test sets. The baseline models selected in the testing phase of this example are divided into two categories: one is based on GhatGPT and its variants, including ChatGPT, ChatGPT+CoT, and KASD-ChatGPT; the other is based on fine-tuning, including FACTUAL, RCT, APARN, and CEIB. In addition, PerspectiveAPI is also used as a baseline model for comparison in hate detection. This is a hate detection tool launched by Google. Tables 1-3 show the comparison results between the method of this embodiment and other model methods in three different opinion mining tasks: stance detection, hate detection, and sentiment analysis. It can be concluded from Tables 1-3 that: a. This example achieves state-of-the-art results on both datasets. Compared to the next-best approach, the macro-F1 score improves by approximately 6%, 2%, and 3%, respectively. Taking the Sem16 dataset as an example, this example achieves significant breakthroughs on both real-world topics and abstract topics (such as climate change). b. For the hate speech detection task, this example conducts experiments on two subtasks: detecting whether it is hate speech (two-category classification) and classifying the type of hate speech (three-category classification). This example achieves the best results in both subtasks. Compared with the baseline using GPT-4o directly, this example improves the weighted F1 score by approximately 2% in the two-category task and by 22% in the three-category task, demonstrating its effectiveness in hate speech detection. c. This example achieves significant improvements in sentiment analysis tasks. The macro-F1 scores on the Rest16 and Lap14 datasets increased by 6% and 1%, respectively. This further demonstrates the effectiveness of the user-centric sentiment analysis perspective. The final experimental results show that the user-centric text opinion mining framework driven by the large language model proposed in this embodiment has achieved the best results on different datasets of multiple tasks. Most of the baseline models compared above are limited by task-specific design and can only perform 1 or 2 tasks. In contrast, the user-centric opinion mining framework of this example can be easily deployed in different tasks. It is worth noting that although the direct use of the powerful GPT-4o and manual prompt engineering can adapt to the above three tasks, its performance is poor, which proves the effectiveness of text analysis for opinion mining tasks from a user-centric perspective.

[0036]

[0037] Table 1 shows that the method of this embodiment achieves the best results on the two stance detection datasets, P-stance and Sem16.

[0038]

[0039] Table 2 shows that the method of this embodiment achieved the best results on the two tasks of the HateXPlain hate detection dataset.

[0040]

[0041] Table 3 shows that the method of this embodiment achieves the best results on the two aspect-level sentiment analysis datasets, Rest14 and laptop14.

[0042] An embodiment of the present invention further provides a system for implementing the above-mentioned user-centric large language model driven text opinion mining method, comprising: The user portrait generation module generates a user portrait based on the text to be detected through the large language model (LLM); The context scene generation module is used to generate the context scene based on the text to be detected through the large language model LLM; The user opinion mining module uses the Large Language Model (LLM) to act as a user agent, inputs user profiles and context scenarios into the user agent, simulates the user's interaction process in the scenario, and mines user opinions; The final opinion generation module is used to use the large language model (LLM) to play the role of expert agent and management agent. The management agent judges the opinions of the user agent and outputs the final opinion based on the discussion results of the expert agent and the user agent.

[0043] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0044] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0045] The above are only preferred embodiments of the present invention and do not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of the present invention specification should be included in the protection scope of the present invention.

Claims

1. A user-centric large language model driven text opinion mining method, characterized by: The following steps are involved: Step 1: Generate a user profile using the Large Language Model (LLM) based on the text to be tested. Step 2: Generate context scenarios using the large language model (LLM) based on the text to be detected. Step 3: Use the Large Language Model (LLM) to act as a user agent. Input the user profile and context into the user agent to simulate the user's interaction process in the scenario and mine the user's opinions. Step 4: Use the large language model (LLM) to act as the expert agent and the management agent. The management agent evaluates the opinions of the user agent and outputs the final opinion based on the discussion results of the expert agent and the user agent.

2. The user-centric large language model driven text opinion mining method according to claim 1, characterized in that: Step 1 implementation method includes: Collect fine-tuning data for initializing the white-box LLM-type portrait attribute generator; Fine-tune the white-box LLM user profile attribute generator based on the data collected above; The white-box LLM profile attribute generator generates the user's key attributes based on the text to be tested and transmits them to the black-box LLM user profile generator to generate a user profile.

3. The user-centric large language model driven text opinion mining method according to claim 2, characterized in that: The Bayesian optimization method is used to optimize the collaboration process between the user profile attribute generator based on the white-box LLM and the user profile generator based on the black-box LLM.

4. The user-centric large language model driven text opinion mining method according to claim 3, characterized in that: The Bayesian hint optimization method specifically includes: 1) Randomly initialize n groups of soft tokens , concatenate each group of soft tokens with the embedding of the user portrait attribute generator in turn to obtain n groups of user portrait attribute prompts; 2) The user profile generator generates a user profile based on the user profile attribute prompts and the text to be detected; 3) The large language model plays the role based on the generated user profile and mines the opinions of the text to be tested from the user's perspective, and calculates the accuracy and F1-score of the opinion mining; 4) Through Gaussian process The mapping function distribution is modeled, and the Gaussian process is specifically expressed as follows: Given a set of observation data , the Gaussian process hypothesis function The joint distribution of any finite set of points follows a multivariate normal distribution; for a new point , given the observation data , The posterior distribution of is: in, It's a new point and the covariance vector between the observation points, yes The covariance matrix of ;function is the covariance function, is the observation vector, yes Identity matrix; therefore, ; 5) Determine the next potential optimal soft token through the expected improvement algorithm , which is expected to produce a better F1 score in the next iteration, the next soft token Determined by the following formula: in is the probability density function, yes The value of After multiple rounds of iterations, the soft token with the highest F1 score is selected as the prefix of the white-box LLM to generate a user profile prompt, which is used as the input of the black-box LLM.

5. The user-centric large language model driven text opinion mining method according to claim 1, characterized in that: The method for generating the context scenario in step 2 is: Input instructions and prompts into the large language model. The instructions are to generate a scenario based on the text, and the prompts specifically describe the four aspects of the generated scenario: background, task description, event description, and event process.

6. The user-centric large language model driven text opinion mining method according to claim 1, characterized in that: The process of user agent simulating users and generating thinking opinions can be expressed as: Initialize the user agent: initialize according to the user portrait generated in step 1, , It represents the base model of the user agent, and u represents the user portrait; Place the user agent in the context scene and think about the text to be detected; transfer the context scene and the text to be detected generated based on step 2 to the new environment E, which is represented by , t is the text to be detected, S is the context scene generated in step 2, and the user agent makes a decision based on the environment , C represents the thinking process of the user agent.

7. The user-centric large language model driven text opinion mining method according to claim 1, characterized in that: Step 4 specifically includes: The management agent evaluates whether the user agent's thinking process and conclusions are reasonable, and directly outputs the results if they are reasonable; If the management agent believes that the user agent's thinking results and conclusions deviate from the text to be tested, the user agent's thinking results will be stored in the thinking pool and provided to the expert agent as a reference, and the expert agent will make a secondary evaluation; After the expert agent discusses with the user agent, the expert agent presents its own opinions and thinking process, and then the manager agent selects one from the user's opinion and the expert's opinion as the final opinion.

8. A system for implementing the user-centric large language model driven text opinion mining method according to any one of claims 1 to 7, characterized in that: include: The user portrait generation module generates a user portrait based on the text to be detected through the large language model (LLM); The context scene generation module is used to generate the context scene based on the text to be detected through the large language model LLM; The user opinion mining module uses the Large Language Model (LLM) to act as a user agent, inputs user profiles and context scenarios into the user agent, simulates the user's interaction process in the scenario, and mines user opinions; The final opinion generation module is used to use the large language model (LLM) to play the role of expert agent and management agent. The management agent judges the opinions of the user agent and outputs the final opinion based on the discussion results of the expert agent and the user agent.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images