Enterprise risk identification method and device based on retrieval enhancement framework

Abstract

The invention relates to an enterprise risk identification method and device based on a retrieval enhancement framework. The method comprises the steps of obtaining a risk query statement of a user, and identifying a target enterprise corresponding to the risk query statement and risk query information corresponding to the risk query statement; on the basis of the risk query information, problem subject information of the risk query statement and current difficulty corresponding to the risk query statement are identified, and an interface calling strategy of the user is generated on the basis of the problem subject information and the current difficulty; calling an enterprise data query result corresponding to the risk query statement through an enterprise data interface of the target enterprise based on an interface calling strategy of the user, and identifying an enterprise risk analysis result of the target enterprise through a problem analysis model based on the enterprise data query result; and sending an enterprise risk analysis result to a user client of the user. By adopting the method, the accuracy of enterprise credit investigation and risk analysis with high-frequency data change can be improved.

Description

Technical Field

[0001] The present application relates to the technical fields of intelligent information processing, and corporate credit investigation and risk assessment, and in particular to a method and device for identifying corporate risks based on a retrieval enhancement framework. Background Art

[0002] With the increasing complexity of the business environment and the acceleration of globalization, traditional corporate credit and risk analysis tools can no longer meet the needs of timely, accurate and dynamic insights into high-speed changes and multi-dimensional data. Traditional credit models usually rely on static rules and simple scoring methods based on limited data sources, and are not flexible enough to respond to emerging new data and new risk factors. Therefore, how to improve the accuracy and flexibility of corporate credit and risk analysis is the current research focus.

[0003] Traditional corporate credit investigation and risk analysis methods generally rely on large static knowledge bases that are built in advance. The initial construction, data cleaning, and regular updates of these knowledge bases are not only costly and time-consuming, but also less flexible when faced with high-frequency data changes. For enterprise-level scenarios with large data volumes and frequent changes, it is more difficult to manually maintain a large static knowledge base that is always of high quality and highly available, resulting in poor accuracy in corporate credit investigation and risk analysis for high-frequency data changes. Summary of the invention

[0004] Based on this, it is necessary to provide an enterprise risk identification method, device, computer equipment, computer-readable storage medium and computer program product based on a retrieval enhancement framework to address the above technical problems.

[0005] In a first aspect, the present application provides an enterprise risk identification method based on a retrieval enhancement framework, comprising:

[0006] Acquire a risk query statement of a user, and identify a target enterprise corresponding to the risk query statement, and risk query information corresponding to the risk query statement;

[0007] Based on the risk query information, identifying the question subject information of the risk query statement and the current difficulty corresponding to the risk query statement, and generating the interface call strategy of the user based on the question subject information and the current difficulty;

[0008] Based on the user's interface calling strategy, calling the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise, and based on the enterprise data query result, identifying the enterprise risk analysis result of the target enterprise through the problem analysis model;

[0009] The enterprise risk analysis result is sent to the user client of the user.

[0010] Optionally, the identifying the target enterprise corresponding to the risk query statement and the risk query information corresponding to the risk query statement includes:

[0011] Identify the query text content corresponding to the risk query statement through a text recognition network, and identify key semantic information in the query text content through a semantic recognition network;

[0012] Based on each key semantic information, identifying each key semantic content corresponding to the risk query statement and the semantic information corresponding to each key semantic content, and using each key semantic content and the semantic information corresponding to each key semantic content as the risk query information corresponding to the risk query statement;

[0013] Based on the semantic information corresponding to each key semantic content, the enterprise entity content in each key semantic content is identified, and based on each enterprise entity content, the target enterprise corresponding to the risk query statement is identified through an enterprise name identification strategy.

[0014] Optionally, the identifying, based on the risk query information, question subject information of the risk query statement and a current difficulty level corresponding to the risk query statement includes:

[0015] Based on the semantic information of each key semantic content, the key question elements in each key semantic content are extracted through the question subject extraction strategy, and all the key question elements are used as the question subject information of the risk query statement;

[0016] Based on each of the key problem elements, the current difficulty of the risk query statement adaptation is identified through a difficulty identification matrix.

[0017] Optionally, the generating the user's interface calling strategy based on the question subject information and the current difficulty level includes:

[0018] Based on the current difficulty, identifying the interface calling method of the user, and based on each of the key problem elements, querying the interface calling database for the adaptation calling interface information corresponding to each key problem element;

[0019] Based on the adaptation calling interface information corresponding to each of the key problem elements, the target calling interface information called by the user needs is screened through the interface optimization calling strategy, and the target calling interface information and the interface calling method are used as the user's interface calling strategy.

[0020] Optionally, calling the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise based on the interface calling strategy of the user includes:

[0021] Based on each of the target call interface information, identifying each target call interface identifier and each call requirement information;

[0022] Based on the call requirement information corresponding to each target call interface identifier, calling the target call content corresponding to each call requirement information from the target call interface corresponding to each target call interface identifier through the interface call mode;

[0023] Based on the target call content corresponding to each of the call requirement information, it is filled into the call query template to obtain the enterprise data query result corresponding to the risk query statement.

[0024] Optionally, the problem analysis model includes sub-problem analysis models from multiple problem analysis angles, and the enterprise risk analysis result of the target enterprise is identified through the problem analysis model based on the enterprise data query result, including:

[0025] Splitting each target call content into call data groups corresponding to each problem analysis angle;

[0026] For each problem analysis angle, based on the call data group corresponding to the problem analysis angle, the problem analysis result corresponding to the problem analysis angle is identified through the sub-problem analysis model of the problem analysis angle;

[0027] Based on the problem analysis results corresponding to each of the problem analysis angles, a problem analysis conclusion of the target enterprise is generated, and the problem analysis conclusion is used as the enterprise risk analysis result of the target enterprise.

[0028] In a second aspect, the present application also provides an enterprise risk identification device based on a retrieval enhancement framework, comprising:

[0029] An acquisition module, used to acquire a risk query statement of a user, and identify a target enterprise corresponding to the risk query statement, and risk query information corresponding to the risk query statement;

[0030] A generating module, configured to identify, based on the risk query information, the question subject information of the risk query statement and the current difficulty level corresponding to the risk query statement, and to generate, based on the question subject information and the current difficulty level, an interface calling strategy for the user;

[0031] an identification module, configured to call the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise based on the interface call strategy of the user, and identify the enterprise risk analysis result of the target enterprise through the problem analysis model based on the enterprise data query result;

[0032] The sending module is used to send the enterprise risk analysis result to the user client of the user.

[0033] Optionally, the acquisition module is specifically used to:

[0034] Identify the query text content corresponding to the risk query statement through a text recognition network, and identify key semantic information in the query text content through a semantic recognition network;

[0035] Based on each key semantic information, identifying each key semantic content corresponding to the risk query statement and the semantic information corresponding to each key semantic content, and using each key semantic content and the semantic information corresponding to each key semantic content as the risk query information corresponding to the risk query statement;

[0036] Based on the semantic information corresponding to each key semantic content, the enterprise entity content in each key semantic content is identified, and based on each enterprise entity content, the target enterprise corresponding to the risk query statement is identified through an enterprise name identification strategy.

[0037] Optionally, the generating module is specifically used for:

[0038] Based on the semantic information of each key semantic content, the key question elements in each key semantic content are extracted through the question subject extraction strategy, and all the key question elements are used as the question subject information of the risk query statement;

[0039] Based on each of the key problem elements, the current difficulty of the risk query statement adaptation is identified through a difficulty identification matrix.

[0040] Optionally, the generating module is specifically used for:

[0041] Based on the current difficulty, identifying the interface calling method of the user, and based on each of the key problem elements, querying the interface calling database for the adaptation calling interface information corresponding to each key problem element;

[0042] Based on the adaptation calling interface information corresponding to each of the key problem elements, the target calling interface information called by the user needs is screened through the interface optimization calling strategy, and the target calling interface information and the interface calling method are used as the user's interface calling strategy.

[0043] Optionally, the identification module is specifically used to:

[0044] Based on each of the target call interface information, identifying each target call interface identifier and each call requirement information;

[0045] Based on the call requirement information corresponding to each target call interface identifier, calling the target call content corresponding to each call requirement information from the target call interface corresponding to each target call interface identifier through the interface call mode;

[0046] Based on the target call content corresponding to each of the call requirement information, it is filled into the call query template to obtain the enterprise data query result corresponding to the risk query statement.

[0047] Optionally, the identification module is specifically used to:

[0048] Splitting each target call content into call data groups corresponding to each problem analysis angle;

[0049] For each problem analysis angle, based on the call data group corresponding to the problem analysis angle, the problem analysis result corresponding to the problem analysis angle is identified through the sub-problem analysis model of the problem analysis angle;

[0050] Based on the problem analysis results corresponding to each of the problem analysis angles, a problem analysis conclusion of the target enterprise is generated, and the problem analysis conclusion is used as the enterprise risk analysis result of the target enterprise.

[0051] In a third aspect, the present application provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the methods in the first aspect are implemented.

[0052] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0053] In a fifth aspect, the present application provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0054] The above-mentioned enterprise risk identification method and device based on the retrieval enhancement framework obtains the user's risk query statement and identifies the target enterprise corresponding to the risk query statement and the risk query information corresponding to the risk query statement; based on the risk query information, identifies the problem subject information of the risk query statement and the current difficulty corresponding to the risk query statement, and generates the user's interface call strategy based on the problem subject information and the current difficulty; based on the user's interface call strategy, calls the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise, and identifies the enterprise risk analysis result of the target enterprise through the problem analysis model based on the enterprise data query result; sends the enterprise risk analysis result to the user client of the user. This solution does not require the construction of a large-scale static knowledge base, but implements refined control of the data range based on the permission strategy of the enterprise data interface. Then, when conducting risk queries through the enterprise's multi-data interface, this solution identifies the problem subject and performs difficulty analysis on the user's risk query statement, thereby generating the user's interface call strategy, so that when making enterprise data calls, it can autonomously generate enterprise data query results suitable for the current user's call requirements based on different call requirements and data change requirements. Finally, this solution can also perform risk analysis based on the called data, thereby avoiding the inefficiency and low precision of manual analysis. Finally, during intelligent analysis, this solution can adaptively adjust the generation method of the interface call strategy according to the results of the analysis, so that this solution can continue to evolve in a complex and changeable data environment, improve decision-making intelligence and the accuracy of answers, thereby comprehensively improving the accuracy of corporate credit investigation and risk analysis of high-frequency data changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A schematic diagram of a process of an enterprise risk identification method based on a retrieval enhancement framework in one embodiment;

[0057] Figure 2 A flowchart of an example of enterprise risk identification based on a retrieval enhancement framework in one embodiment;

[0058] Figure 3 A structural block diagram of an enterprise risk identification device based on a retrieval enhancement framework in one embodiment;

[0059] Figure 4 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0061] The enterprise risk identification method based on the retrieval enhancement framework provided in the embodiment of the present application can be applied to the application environment of enterprise risk identification based on the retrieval enhancement framework. The method can be applied to a terminal, a server, or a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers, laptops, mid-size computers, etc. The terminal does not need to build a large-scale static knowledge base, but instead implements refined control of the data range based on the permission policy of the enterprise data interface. Then, when conducting risk queries through the enterprise's multi-data interface, this solution identifies the problem subject and performs difficulty analysis on the user's risk query statement, thereby generating the user's interface call strategy, so that when making enterprise data calls, it can autonomously generate enterprise data query results suitable for the current user's call requirements based on different call requirements and data change requirements. Finally, this solution can also perform risk analysis based on the called data, thereby avoiding the inefficiency and low precision of manual analysis. Finally, during intelligent analysis, this solution can adaptively adjust the generation method of the interface call strategy according to the results of the analysis, so that this solution can continue to evolve in a complex and changeable data environment, improve decision-making intelligence and the accuracy of answers, thereby comprehensively improving the accuracy of corporate credit investigation and risk analysis of high-frequency data changes.

[0062] In an exemplary embodiment, Figure 1 As shown, a method for enterprise risk identification based on a search enhancement framework is provided, and the method is applied to a terminal as an example for explanation, including the following steps S101 to S104. Among them:

[0063] Step S101, obtaining a risk query statement of a user, and identifying a target enterprise corresponding to the risk query statement and risk query information corresponding to the risk query statement.

[0064] In this embodiment, the terminal responds to the user's natural language query operation, obtains the user's query statement information, and obtains a risk query statement, wherein the risk query statement, for example, queries the recent investment risk points of Company A. Then, the terminal identifies the target enterprise corresponding to the risk query statement, and the risk query information corresponding to the risk query statement. Among them, the risk query information includes the key semantic content of each query type, wherein the query type includes but is not limited to (enterprise risk consulting, credit score query, financial health assessment, etc.) The specific identification process will be described in detail later. Among them, the target enterprise is the enterprise that the user needs to query. When the user's risk query statement includes multiple enterprises with the same name and enterprises with fuzzy names, it is necessary to identify the target enterprise in the risk query statement through the multi-agent cluster preset by the terminal.

[0065] Step S102, based on the risk query information, identifying the question subject information of the risk query statement and the current difficulty corresponding to the risk query statement, and generating a user interface call strategy based on the question subject information and the current difficulty.

[0066] In this embodiment, the terminal identifies the problem subject information of the risk query statement and the current difficulty corresponding to the risk query statement based on the risk query information, and generates the user's interface call strategy based on the problem subject information and the current difficulty. Among them, the problem subject information is the main problem content in the above query statement, for example, "investment risk points", "recent credit changes", and "supply chain anomalies". Then, the difficulty identification strategy preset by the terminal identifies the current difficulty corresponding to the risk query statement. The specific identification process will be described in detail later. Among them, the current difficulty is divided according to the amount of enterprise content that needs to be queried, the complexity of obtaining the query enterprise content, and the difficulty of obtaining the query enterprise content. The specific division process will be described in detail later.

[0067] Step S103, based on the user's interface calling strategy, the enterprise data query result corresponding to the risk query statement is called through the enterprise data interface of the target enterprise, and based on the enterprise data query result, the enterprise risk analysis result of the target enterprise is identified through the problem analysis model.

[0068] In this embodiment, the terminal calls the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise based on the user's interface call strategy, and identifies the enterprise risk analysis result of the target enterprise through the problem analysis model based on the enterprise data query result. Among them, the problem analysis model is an analysis model based on different problem analysis angles of multi-agent agents. The analysis model can perform in-depth reasoning and analysis on the enterprise data query results from different dimensions, such as in-depth credit risk assessment, long-term trend prediction, and investment advice analysis. Among them, the analysis model package is a model that uses technologies such as multi-agent collaboration, reinforcement learning (RL), long-term memory (LTM), and retrieval-augmented generation (RAG).

[0069] Step S104: sending the enterprise risk analysis result to the user's client.

[0070] In this embodiment, the terminal sends the enterprise risk analysis results to the user's user client. Then, the terminal records the user's satisfaction, answer quality evaluation and actual decision effect (such as subsequent credit decision results) after receiving the answer, and stores these feedback information in LTM. Then, the terminal uses the feedback through the RL strategy module to continuously adjust the AP call strategy and Agent decision path, thereby adjusting the generation method of the interface call strategy, so that a more reasonable and efficient data source combination and answer strategy can be selected next time a similar problem is encountered.

[0071] Based on the above scheme, there is no need to build a large-scale static knowledge base, but based on the permission strategy of the enterprise data interface, the refined control of the data range is achieved. Then, when conducting risk queries through the enterprise multi-data interface, this scheme identifies the subject of the problem and analyzes the difficulty of the user's risk query statement, thereby generating the user's interface call strategy, so that when conducting enterprise data calls, it can autonomously generate enterprise data query results suitable for the current user's call requirements according to different call requirements and data change requirements. Finally, this scheme can also perform risk analysis based on the called data, thereby avoiding the inefficiency and low precision of manual analysis. Finally, during intelligent analysis, this scheme can adaptively adjust the generation method of the interface call strategy according to the results of the analysis, so that this scheme can continue to evolve in a complex and changeable data environment, improve decision-making intelligence and the accuracy of answers, thereby comprehensively improving the accuracy of corporate credit investigation and risk analysis for high-frequency data changes.

[0072] Optionally, identifying the target enterprise corresponding to the risk query statement and the risk query information corresponding to the risk query statement includes: identifying the query text content corresponding to the risk query statement through a text recognition network, and identifying key semantic information in the query text content through a semantic recognition network; based on each key semantic information, identifying each key semantic content corresponding to the risk query statement and the semantic information corresponding to each key semantic content, and using each key semantic content and the semantic information corresponding to each key semantic content as the risk query information corresponding to the risk query statement; based on the semantic information corresponding to each key semantic content, identifying the enterprise entity content in each key semantic content, and based on each enterprise entity content, identifying the target enterprise corresponding to the risk query statement through an enterprise name recognition strategy.

[0073] In this embodiment, the terminal identifies the query text content corresponding to the risk query statement through a text recognition network, and identifies key semantic information in the query text content through a semantic recognition network. The text recognition network and the semantic recognition network are neural networks based on a large-scale language model (LLM).

[0074] Then, based on each key semantic information, the terminal identifies each key semantic content corresponding to the risk query statement and the semantic information corresponding to each key semantic content, and uses each key semantic content and the semantic information corresponding to each key semantic content as the risk query information corresponding to the risk query statement. Among them, each key semantic content is the language content corresponding to the semantic information related to enterprise risk and query in the risk query statement, for example, investment risk, stock risk, financial status, credit information and other language content.

[0075] Then, based on the semantic information corresponding to each key semantic content, the terminal identifies the enterprise entity content in each key semantic content, and based on each enterprise entity content, identifies the target enterprise corresponding to the risk query statement through the enterprise name identification strategy. Among them, the enterprise entity content is the language content containing the enterprise name, wherein the language content includes each enterprise name, fuzzy name, etc. Then, the terminal first uses the Agent to use the established enterprise name-EntID mapping library (such as the search of the internal database) to quickly match the unique enterprise ID (or enterprise name information) to obtain the target enterprise. When no match can be made, the terminal generates possible enterprise name information based on each enterprise name and fuzzy name, and feeds it back to the user client so that the user can directly filter it. Then, the terminal uses the enterprise name information filtered and determined by the user as the target enterprise.

[0076] Based on the above solution, the target enterprise corresponding to the risk query statement and the risk query information corresponding to the risk query statement are identified through enterprise entity identification, thereby improving the identification efficiency and accuracy.

[0077] Optionally, based on the risk query information, identify the question body information of the risk query statement and the current difficulty corresponding to the risk query statement, including: based on the semantic information of each key semantic content, extract the key question elements in each key semantic content through the question body extraction strategy, and use all the key question elements as the question body information of the risk query statement; based on each key question element, identify the current difficulty adapted to the risk query statement through a difficulty discrimination matrix.

[0078] In this embodiment, the terminal extracts the key question elements in each key semantic content through the question subject extraction strategy based on the semantic information of each key semantic content, and uses all the key question elements as the question subject information of the risk query statement. The method of extracting the key question elements is to extract the key question elements corresponding to each key semantic content in each key semantic content through the question subject Agent. The key question elements are equivalent to the main question content of the key semantic content.

[0079] Then, based on each key question element, the terminal identifies the current difficulty level that the risk query statement is adapted to through the difficulty identification matrix. Specifically, the terminal identifies the current difficulty level that the risk query statement containing each key question element is adapted to through the built-in strategy difficulty matrix, where the current difficulty level can be divided into simple questions, medium difficulty questions, and high difficulty and complex questions. For example:

[0080] Simple questions (low difficulty): such as "Query the basic information of Company A" or "Has Company A had any major negative news recently?"

[0081] Medium difficulty question: One or two APIs need to be called for data comparison (such as financial API and social information API).

[0082] High-difficulty and complex problems: require multi-dimensional data aggregation (finance, legal proceedings, investment events, macroeconomic indicators, etc.) and involve complex reasoning and evaluation.

[0083] Based on the above solution, after extracting the key problem elements, the difficulty of the problem is identified, thereby improving the accuracy of identifying the difficulty of the problem.

[0084] Optionally, based on the problem subject information and the current difficulty, a user's interface calling strategy is generated, including: based on the current difficulty, identifying the user's interface calling method, and based on each key problem element, querying the adaptation calling interface information corresponding to each key problem element in the interface calling database; based on the adaptation calling interface information corresponding to each key problem element, through the interface optimization calling strategy, screening each target calling interface information called by the user's needs, and using each target calling interface information and interface calling method as the user's interface calling strategy.

[0085] In this embodiment, the terminal identifies the user's interface call method based on the current difficulty, and based on each key question element, queries the interface call database for the adaptation call interface information corresponding to each key question element. Specifically, for example, if the question is simple and the local knowledge (information in the LLM parameters) is sufficient to answer, the agent directly outputs the answer, just like the "quick check for minor illnesses" in the outpatient department. If a small amount of general information is still needed, the web-search interface can be called for quick data retrieval (similar to an outpatient doctor going to a pharmacy to get ready-made medicine).

[0086] If the problem is slightly more difficult, it is necessary to call the enterprise data API to obtain real-time information (such as recent financial statements, industry data, and investment event lists). In this step, the RL strategy agent will dynamically select which APs to call, in what order, and in what priority to retrieve data based on historical feedback and current problem characteristics.

[0087] Afterwards, the terminal uses the interface optimization calling strategy to screen the target calling interface information called by the user based on the adaptation calling interface information corresponding to each key problem element, and uses the target calling interface information and the interface calling method as the user's interface calling strategy. Among them, the interface optimization calling strategy is an RL strategy, and the optimization goals of the RL strategy include: reducing unnecessary AP calls, giving priority to calling high-value data sources, speeding up data return speed, and improving the accuracy of the final answer. Among them, the generation method of the RL strategy is based on the Agenic-RAG framework, and the reinforcement learning (RL) strategy is the engine of the system's self-evolution. The RL strategy continuously selects and optimizes the API call path, data retrieval strategy, and decision-making process between agents, so that the system can always maintain high adaptability and high answer quality in a rapidly changing enterprise data environment. Drawing on the research ideas of OpenAl et al. in RLfrom Human Feedback (RLHF) and ByteDance in user behavior feedback optimization, the present invention incorporates multiple signals such as user stay time, page interaction behavior, feedback score and business decision effectiveness into the RL optimization goal.

[0088] Traditional RLHF fusion: Similar to the RLHF (Reinforcement Learning Based on Human Feedback) framework proposed by OpenAI, the terminal can use human annotators or internal experts to evaluate the quality of answers in the early stage to provide initial reward signals for the RL strategy. For example, when the expert believes that the answer is accurate and valuable, the corresponding API call path and data screening strategy will be given positive rewards, otherwise they will be punished.

[0089] Introduction of user behavior signals: Once the system is online and interacts with real users, user behavior can be used as the main source of feedback signals. In the corporate credit inquiry page, various user behavior data can be used as feedback indicators:

[0090] Page dwell time (DwellTime): The length of time a user stays on the answer page can reflect the attractiveness and usefulness of the answer to the user. A longer dwell time or users repeatedly browsing related answers and data usually means that the answer is more valuable.

[0091] Scroll and click interactions: Whether users further expand details, click related links, or view additional data descriptions provides feedback on the depth and ease of use of the answer.

[0092] Active user feedback: Simple star ratings, likes / dislikes, text feedback, etc. can be directly used to adjust RL strategies.

[0093] Business decision feedback: In the actual application of corporate credit investigation, some answers will affect downstream business decisions (such as lending, risk rating, and investment judgment). If the user's business decision performance after referring to the answer is better (such as reducing the bad debt rate and increasing the investment success rate), it can be regarded as the answer has actual value to the user, and the RL strategy will give higher rewards to the data call path and decision-making process.

[0094] Among them, the reward function design: Combine the latest research ideas and integrate various user feedback and business indicators into the reward function. For example, the reward function can be a weighted combination of the following forms:

[0095] R=a*manual annotation score+β*normalized value of page dwell time+y*user click / interaction index+6*business decision effectiveness measurement+…

[0096] Different weights can be set through experiments and business goals, and RL strategies constantly try to balance short-term user satisfaction (such as dwell time) with long-term business decision effectiveness.

[0097] Strategy Improvement: After the reward function is optimized, the RL strategy will make changes to the API call and data screening strategies based on the reward signal, tending to choose decision paths that have historically received higher comprehensive rewards, thereby improving the overall value of the answer and user experience.

[0098] Based on the above solution, the interface calling strategy of the user is generated based on the question subject information and the current difficulty, thereby improving the generation efficiency of the interface calling strategy.

[0099] Optionally, based on the user's interface calling strategy, the enterprise data query result corresponding to the risk query statement is called through the enterprise data interface of the target enterprise, including: based on each target calling interface information, identifying each target calling interface identifier and each calling requirement information; based on the calling requirement information corresponding to each target calling interface identifier, calling the target calling content corresponding to each calling requirement information from the target calling interface corresponding to each target calling interface identifier through the interface calling method; based on the target calling content corresponding to each calling requirement information, filling it into the calling query template to obtain the enterprise data query result corresponding to the risk query statement.

[0100] In this embodiment, the terminal identifies each target call interface identifier and each call requirement information based on each target call interface information. Based on the call requirement information corresponding to each target call interface identifier, the target call content corresponding to each call requirement information is called from the target call interface corresponding to each target call interface identifier through the interface call method. Among them, each target call interface identifier is the interface identification information of the enterprise data interface called by the terminal, which is used to identify the enterprise data structure to be called. Then, the terminal fills the call query template based on the target call content corresponding to each call requirement information to obtain the enterprise data query result corresponding to the risk query statement. Among them, the calling method is to query the interface data through the RAG (retrieval enhancement generation) module. Among them, the terminal fine-tunes the business field of the large pre-trained language model (LLM) (for example, vertical segmentation optimization for corporate credit reporting), accesses other professional large models (foreign aid doctors) through the interface, and a collaborative analysis mechanism similar to multidisciplinary consultation, thereby improving the professionalism, accuracy and explanatory power of the answer.

[0101] Among them, the RAG module plays a key role in "data integration and intelligence enhancement" in the Agenic-RAG framework. Its core idea is: based on the basic capabilities of LLM, the latest and most relevant enterprise data and historical context obtained from real-time API retrieval and LTM (long-term memory) are injected into the LM answering process, so that the answer not only relies on the general knowledge of LLM, but also makes full use of dynamic business data to achieve accurate and real-time enterprise credit Q&A and risk analysis.

[0102] The functions of the RAG module also include:

[0103] Industry-customized fine-tuning: In the medical analogy, although general practitioners (general LMs) have extensive knowledge reserves, they still lack specialized experience in complex vertical fields such as corporate credit reporting. This module introduces domain fine-tuning technology for corporate credit reporting scenarios, which enables LLMs to have more professional understanding and reasoning capabilities on corporate credit reporting issues by retraining or prompting engineering enhancements on a large amount of corporate credit data, financial statements, legal judgments, industry reports, and regulatory policy documents.

[0104] Business knowledge injection: LLM is fine-tuned through industry data (such as typical corporate cases, financial indicator explanations, and risk rating model documents) so that it can provide more explanatory, feasible, and practical answers to questions such as corporate debt structure, investment risks, upstream and downstream fluctuations in the industrial chain, and changes in regulatory policies.

[0105] Interfaced large model access: The RAG module is not limited to one LM instance. When encountering specific problems (such as legal compliance analysis, industrial chain risk assessment, and macroeconomic impact measurement), the RAG module can call other large models optimized specifically for this field through predefined interfaces - these external models can be regarded as "specialists" in their respective fields.

[0106] Multi-model consultation mechanism: Similar to expert consultation in the medical field, a group of doctors (multiple models) collaborate to discuss the patient's complex condition. Here, multiple dedicated large models conduct parallel analysis and cross-validation on the same corporate credit issue. The RAG module can integrate answers and opinions from different specialist models (such as legal models, financial models, industrial chain models, and social information analysis models) and select the best result through voting, weighting or RL strategies.

[0107] Dynamic expansion and adaptation: When the enterprise data ecology or regulatory environment changes, or new professional needs arise, external large models can be added or replaced (similar to introducing new specialists) without major changes to the entire framework. Only the configuration needs to be updated through the interface at the RAG layer. In this way, the system can quickly adapt to new situations and maintain long-term evolution capabilities.

[0108] Data injection: When RAG receives data packets from API retrieval and LTM history, it encapsulates this information into contextual prompts and injects them into the input of LLM. For external aid models, the same data context is provided in a similar manner. Model answer generation: LLM and / or external aid models begin to generate answers after receiving prompts. Vertically fine-tuned LLMs have a high-level understanding of corporate credit issues and can perform deep reasoning on data. Specialized external aid large models provide professional insights and credible evidence in their respective fields.

[0109] Result aggregation: The RAG module integrates the results returned from each model, which may include weighting and comparing different answers, or even meta-evaluating the answers by a meta-model (similar to the expert group leader arbitrating the opinions of various specialists) to obtain the best comprehensive answer.

[0110] Real-time data enhancement: Traditional LLM answers lag behind model training time, while the RAG module uses API real-time data to make the answers timely - for example, the answers to questions such as "Company A's recent investment risks" can refer to the latest financial statements and social information in real time.

[0111] Explanation and evidence presentation: In the medical analogy, patients not only want doctors to give a diagnosis, but also need to understand the evidence behind it. Similarly, in the RAG process, the system can annotate the data source, the API called, and the relevant historical record summary in the answer results to improve the explainability and credibility of the answer.

[0112] Feedback loop: The RAG module works with the RL strategy to judge the quality of multi-model collaborative answers through user feedback and system internal evaluation. When it is found that some external aid models perform better for certain types of questions, the RL strategy can give priority to them to participate in the consultation and optimize API calls and data combination strategies.

[0113] Sustainable evolution: As time goes by and the environment changes, the RAG module will accumulate experience and gradually improve the calling strategy of large models in different fields, so that it can continuously enhance the accuracy and business value of answers in corporate credit analysis.

[0114] Based on the above solution, there is no need to build a large-scale static knowledge base. Instead, based on the permission strategy of the enterprise data interface, fine-grained control of the data scope is achieved to generate enterprise data query results, thereby improving the accuracy and efficiency of generating enterprise data query results.

[0115] Optionally, the problem analysis model includes sub-problem analysis models of multiple problem analysis angles, and based on the enterprise data query results, the problem analysis model is used to identify the enterprise risk analysis results of the target enterprise, including: splitting each target call content into call data groups corresponding to each problem analysis angle; for each problem analysis angle, based on the call data group corresponding to the problem analysis angle, identifying the problem analysis result corresponding to the problem analysis angle through the sub-problem analysis model of the problem analysis angle; based on the problem analysis results corresponding to each problem analysis angle, generating a problem analysis conclusion for the target enterprise, and using the problem analysis conclusion as the enterprise risk analysis result for the target enterprise.

[0116] In this embodiment, the terminal splits each target call content into call data groups corresponding to each problem analysis angle. Then, for each problem analysis angle, based on the call data group corresponding to the problem analysis angle, the terminal identifies the problem analysis result corresponding to the problem analysis angle through the sub-problem analysis model of the problem analysis angle. Finally, based on the problem analysis results corresponding to each problem analysis angle, the terminal generates the problem analysis conclusion of the target enterprise, and uses the problem analysis conclusion as the enterprise risk analysis result of the target enterprise. Specifically, when generating the problem analysis conclusion of the target enterprise, the terminal identifies the sub-analysis conclusion corresponding to each problem analysis result, and then summarizes all the sub-analysis conclusions, queries the analysis conclusion of the target enterprise in the problem analysis database, and obtains the enterprise risk analysis result of the target enterprise. Among them, in the problem analysis database, the sub-analysis conclusions contained in the analysis conclusions of different enterprises.

[0117] Based on the above solution, by splitting the analysis angles of each problem, the enterprise risk analysis results of each target enterprise are identified, thereby improving the accuracy and comprehensiveness of the analysis of the target enterprise.

[0118] This application also provides an example of enterprise risk identification based on a retrieval enhancement framework, such as Figure 2 As shown, the specific processing process includes the following steps:

[0119] Step S201, obtaining a user's risk query statement.

[0120] Step S202: identifying the query text content corresponding to the risk query statement through a text recognition network, and identifying key semantic information in the query text content through a semantic recognition network.

[0121] Step S203, based on each key semantic information, identify each key semantic content corresponding to the risk query statement and the semantic information corresponding to each key semantic content, and use each key semantic content and the semantic information corresponding to each key semantic content as the risk query information corresponding to the risk query statement.

[0122] Step S204, based on the semantic information corresponding to each key semantic content, the enterprise entity content in each key semantic content is identified, and based on each enterprise entity content, the target enterprise corresponding to the risk query statement is identified through the enterprise name identification strategy.

[0123] Step S205 , based on the semantic information of each key semantic content, the key question elements in each key semantic content are extracted through the question subject extraction strategy, and all the key question elements are used as the question subject information of the risk query statement.

[0124] Step S206: Based on each key problem element, identify the current difficulty level adapted to the risk query statement through a difficulty discrimination matrix.

[0125] Step S207: Based on the current difficulty level, identify the user's interface call method, and based on each key problem element, query the corresponding adapted call interface information for each key problem element in the interface call database.

[0126] Step S208: Based on the corresponding adapted call interface information for each key problem element, filter the target call interface information required by the user through an interface optimized call strategy, and use the target call interface information and the interface call method as the user's interface call strategy.

[0127] Step S209: Based on the target call interface information, identify each target call interface identifier and each call requirement information.

[0128] Step S210: Based on the call requirement information corresponding to each target call interface identifier, through the interface call method, call the target call content corresponding to each call requirement information from the target call interface corresponding to each target call interface identifier.

[0129] Step S211: Fill the target call content corresponding to each call requirement information into the call query template to obtain the enterprise data query result corresponding to the risk query statement.

[0130] Step S212: Split each target call content into call data groups corresponding to each problem analysis angle.

[0131] Step S213: For each problem analysis angle, based on the call data group corresponding to the problem analysis angle, identify the problem analysis result corresponding to the problem analysis angle through the sub-problem analysis model of the problem analysis angle.

[0132] Step S214: Based on the problem analysis results corresponding to each problem analysis angle, generate the problem analysis conclusion of the target enterprise, and use the problem analysis conclusion as the enterprise risk analysis result of the target enterprise.

[0133] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0134] Based on the same inventive concept, the embodiment of the present application also provides a device for enterprise risk identification based on a search enhancement framework for implementing the enterprise risk identification method based on a search enhancement framework. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the device for enterprise risk identification based on a search enhancement framework provided below can refer to the limitations of the enterprise risk identification method based on a search enhancement framework above, and will not be repeated here.

[0135] In an exemplary embodiment, Figure 3 As shown, an enterprise risk identification device based on a retrieval enhancement framework is provided, comprising: an acquisition module 310, a generation module 320, an identification module 330 and a sending module 340, wherein:

[0136] The acquisition module 310 is used to acquire the risk query statement of the user, and identify the target enterprise corresponding to the risk query statement, and the risk query information corresponding to the risk query statement;

[0137] A generating module 320, configured to identify, based on the risk query information, the question subject information of the risk query statement and the current difficulty level corresponding to the risk query statement, and to generate, based on the question subject information and the current difficulty level, an interface calling strategy for the user;

[0138] The identification module 330 is used to call the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise based on the interface call strategy of the user, and identify the enterprise risk analysis result of the target enterprise through the problem analysis model based on the enterprise data query result;

[0139] The sending module 340 is used to send the enterprise risk analysis result to the user client of the user.

[0140] Optionally, the acquisition module 310 is specifically configured to:

[0141] Identify the query text content corresponding to the risk query statement through a text recognition network, and identify key semantic information in the query text content through a semantic recognition network;

[0142] Based on each key semantic information, identifying each key semantic content corresponding to the risk query statement and the semantic information corresponding to each key semantic content, and using each key semantic content and the semantic information corresponding to each key semantic content as the risk query information corresponding to the risk query statement;

[0143] Based on the semantic information corresponding to each key semantic content, the enterprise entity content in each key semantic content is identified, and based on each enterprise entity content, the target enterprise corresponding to the risk query statement is identified through an enterprise name identification strategy.

[0144] Optionally, the generating module 320 is specifically configured to:

[0145] Based on the semantic information of each key semantic content, the key question elements in each key semantic content are extracted through the question subject extraction strategy, and all the key question elements are used as the question subject information of the risk query statement;

[0146] Based on each of the key problem elements, the current difficulty of the risk query statement adaptation is identified through a difficulty identification matrix.

[0147] Optionally, the generating module 320 is specifically configured to:

[0148] Based on the current difficulty, identifying the interface calling method of the user, and based on each of the key problem elements, querying the interface calling database for the adaptation calling interface information corresponding to each key problem element;

[0149] Based on the adaptation calling interface information corresponding to each of the key problem elements, the target calling interface information called by the user needs is screened through the interface optimization calling strategy, and the target calling interface information and the interface calling method are used as the user's interface calling strategy.

[0150] Optionally, the identification module 330 is specifically used to:

[0151] Based on each of the target call interface information, identifying each target call interface identifier and each call requirement information;

[0152] Based on the call requirement information corresponding to each target call interface identifier, calling the target call content corresponding to each call requirement information from the target call interface corresponding to each target call interface identifier through the interface call mode;

[0153] Based on the target call content corresponding to each of the call requirement information, it is filled into the call query template to obtain the enterprise data query result corresponding to the risk query statement.

[0154] Optionally, the identification module 330 is specifically used to:

[0155] Splitting each target call content into call data groups corresponding to each problem analysis angle;

[0156] For each problem analysis angle, based on the call data group corresponding to the problem analysis angle, the problem analysis result corresponding to the problem analysis angle is identified through the sub-problem analysis model of the problem analysis angle;

[0157] Based on the problem analysis results corresponding to each of the problem analysis angles, a problem analysis conclusion of the target enterprise is generated, and the problem analysis conclusion is used as the enterprise risk analysis result of the target enterprise.

[0158] Each module in the enterprise risk identification device based on the retrieval enhancement framework can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0159] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 4As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for identifying enterprise risks based on a retrieval enhancement framework is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0160] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0161] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, steps corresponding to the enterprise risk identification method based on the retrieval enhancement framework are implemented.

[0162] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps corresponding to the enterprise risk identification method based on the retrieval enhancement framework are implemented.

[0163] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements steps corresponding to the enterprise risk identification method based on a retrieval enhancement framework.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0165] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0166] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0167] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for enterprise risk identification based on a retrieval enhancement framework, characterized in that: The method comprises: Acquire a risk query statement of a user, and identify a target enterprise corresponding to the risk query statement, and risk query information corresponding to the risk query statement; Based on the risk query information, identifying the question subject information of the risk query statement and the current difficulty corresponding to the risk query statement, and generating the interface call strategy of the user based on the question subject information and the current difficulty; Based on the user's interface calling strategy, calling the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise, and based on the enterprise data query result, identifying the enterprise risk analysis result of the target enterprise through the problem analysis model; The enterprise risk analysis result is sent to the user client of the user.

2. The method according to claim 1, characterized in that The identifying the target enterprise corresponding to the risk query statement and the risk query information corresponding to the risk query statement includes: Identify the query text content corresponding to the risk query statement through a text recognition network, and identify key semantic information in the query text content through a semantic recognition network; Based on each key semantic information, identifying each key semantic content corresponding to the risk query statement and the semantic information corresponding to each key semantic content, and using each key semantic content and the semantic information corresponding to each key semantic content as the risk query information corresponding to the risk query statement; Based on the semantic information corresponding to each key semantic content, the enterprise entity content in each key semantic content is identified, and based on each enterprise entity content, the target enterprise corresponding to the risk query statement is identified through an enterprise name identification strategy.

3. The method according to claim 2, characterized in that The step of identifying, based on the risk query information, the question subject information of the risk query statement and the current difficulty level corresponding to the risk query statement includes: Based on the semantic information of each key semantic content, the key question elements in each key semantic content are extracted through the question subject extraction strategy, and all the key question elements are used as the question subject information of the risk query statement; Based on each of the key problem elements, the current difficulty of the risk query statement adaptation is identified through a difficulty identification matrix.

4. The method according to claim 3, characterized in that The generating the user's interface calling strategy based on the question subject information and the current difficulty level includes: Based on the current difficulty, identifying the interface calling method of the user, and based on each of the key problem elements, querying the interface calling database for the adaptation calling interface information corresponding to each key problem element; Based on the adaptation calling interface information corresponding to each of the key problem elements, the target calling interface information called by the user needs is screened through the interface optimization calling strategy, and the target calling interface information and the interface calling method are used as the user's interface calling strategy.

5. The method according to claim 4, characterized in that The calling of the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise based on the interface calling strategy of the user includes: Based on each of the target call interface information, identifying each target call interface identifier and each call requirement information; Based on the call requirement information corresponding to each target call interface identifier, calling the target call content corresponding to each call requirement information from the target call interface corresponding to each target call interface identifier through the interface call mode; Based on the target call content corresponding to each of the call requirement information, it is filled into the call query template to obtain the enterprise data query result corresponding to the risk query statement.

6. The method according to claim 1, characterized in that The problem analysis model includes sub-problem analysis models from multiple problem analysis angles. The enterprise risk analysis results of the target enterprise are identified based on the enterprise data query results through the problem analysis model, including: Splitting each target call content into call data groups corresponding to each problem analysis angle; For each problem analysis angle, based on the call data group corresponding to the problem analysis angle, the problem analysis result corresponding to the problem analysis angle is identified through the sub-problem analysis model of the problem analysis angle; Based on the problem analysis results corresponding to each of the problem analysis angles, a problem analysis conclusion of the target enterprise is generated, and the problem analysis conclusion is used as the enterprise risk analysis result of the target enterprise.

7. An enterprise risk identification device based on a retrieval enhancement framework, characterized in that: The device comprises: An acquisition module, used to acquire a risk query statement of a user, and identify a target enterprise corresponding to the risk query statement, and risk query information corresponding to the risk query statement; A generating module, configured to identify, based on the risk query information, the question subject information of the risk query statement and the current difficulty level corresponding to the risk query statement, and to generate, based on the question subject information and the current difficulty level, an interface calling strategy for the user; an identification module, configured to call the enterprise data query result corresponding to the risk query statement through the enterprise data interface of the target enterprise based on the interface call strategy of the user, and identify the enterprise risk analysis result of the target enterprise through the problem analysis model based on the enterprise data query result; The sending module is used to send the enterprise risk analysis result to the user client of the user.

8. A computer 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 6 are implemented.

9. 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 6 are implemented.

10. A computer program product, comprising a computer program, 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 6 are implemented.

Images