Intelligent decision-making method and device based on agent and electronic equipment

Abstract

The application provides an intelligent decision-making method and device based on an agent and an electronic device, and relates to the technical field of artificial intelligence. Business requirement content of a financial business case is obtained; retrieval data matched with the financial business case is retrieved from a database based on the business requirement content; through a first representative agent and a second representative agent, multi-round debates are carried out based on the retrieval data, and a debate result is obtained; an intelligent strategy scheme corresponding to the business requirement content is obtained based on a decision-making strategy matched with the debate result, so as to improve the processing efficiency of the financial business case and the accuracy of the corresponding intelligent strategy scheme.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an intelligent decision-making method, device and electronic device based on intelligent agents. Background Technology

[0002] In scenarios where bank customers raise business needs related to banking fintech (such as loan risk assessment and review of the completeness of relevant contracts), it is often necessary to process the business cases through intelligent decision-making by human customer service or automated systems.

[0003] Traditional case handling methods rely on manual arbitration or rule engines. However, as case types become more complex and regulatory requirements become more refined, existing methods have significant shortcomings in terms of decision-making transparency, dynamic decision-making capabilities, and rigorous argumentation. This can easily lead to problems such as low decision-making efficiency, misjudgment of risks, and unstable decision results in business cases.

[0004] Therefore, improving the decision-making efficiency and accuracy of corresponding decision-making solutions for banking fintech-related business cases is an urgent problem to be solved. Summary of the Invention

[0005] This application provides an intelligent decision-making method, apparatus, and electronic device based on intelligent agents to improve the processing efficiency of financial business cases and the accuracy of corresponding intelligent strategy solutions.

[0006] In a first aspect, embodiments of this application provide an intelligent decision-making method based on an intelligent agent, including:

[0007] The business requirements for obtaining financial business case studies;

[0008] Based on business needs, retrieve data from the database that matches financial business cases;

[0009] The first and second representative agents conduct multiple rounds of debate based on the retrieved data to obtain the debate results.

[0010] Based on the decision-making strategy that matches the debate results, an intelligent strategy solution corresponding to the business requirements is obtained.

[0011] In one possible implementation, a first representative agent and a second representative agent conduct multiple rounds of debate based on retrieved data to obtain the debate results, including:

[0012] In each round of debate, the first representative agent sends the first argument information, which contains a subset of data. The data in the subset is either the tendency data corresponding to the first representative agent in the retrieved data, or new data obtained based on the argument information fed back by the second representative agent in the previous round of debate.

[0013] Based on the first argument information, the degree of debate between the first representative agent and the second representative agent is dynamically updated to obtain the updated degree of debate.

[0014] The second representative intelligent agent generates second argument information based on the first argument information. The second argument information includes rebuttal data corresponding to the data in the data subset.

[0015] Based on the first and second argumentation information, the comprehensive mismatch degree and comprehensive anchoring strength are obtained;

[0016] The debate ends when the updated degree of contention, overall mismatch, and overall anchoring strength meet the preset conditions, and the debate result is obtained based on the first and second argument information.

[0017] In one possible implementation, the updated debate degree includes a first debate degree corresponding to a first representative agent and a second debate degree corresponding to a second representative agent.

[0018] Based on the first argument information, the debate scores of the first representative agent and the second representative agent are dynamically updated to obtain the updated debate scores, including:

[0019] When the data in the data subset included in the first argument information is the tendency data corresponding to the first representative agent in the retrieved data, the initial degree of argument corresponding to the first representative agent is taken as the first degree of argument.

[0020] When the data in the data subset is new data obtained based on the argumentation information fed back by the second representative agent in the previous round of debate, the first effective support degree and the first mismatch degree of the new data are obtained through the first representative agent;

[0021] Based on the first effective support and the first mismatch, the first degree of controversy is obtained;

[0022] The second representative agent obtains the second effective support and the second mismatch of the data in the data subset.

[0023] The second degree of contention is derived based on the second effective support degree and the second mismatch degree.

[0024] In one possible implementation, based on the first and second argumentation information, the comprehensive mismatch degree and comprehensive anchoring strength are obtained, including:

[0025] To obtain the consensus point between the first and second argument information;

[0026] Perform self-disturbance tests on the consensus data corresponding to each consensus point to obtain the data mismatch degree corresponding to each consensus data point;

[0027] A comprehensive mismatch is obtained based on all the data mismatches.

[0028] The effective support of each consensus data point is obtained through a support calculation strategy; the support calculation strategy includes the adjudication cluster size and the clause matching degree.

[0029] A comprehensive effective support score is obtained based on the effective support of all data.

[0030] The overall anchoring strength is obtained by using the anchoring strength formula, based on the overall mismatch and overall effective support.

[0031] In one possible implementation, the debate terminates when the updated degree of contention, overall mismatch, and overall anchoring strength meet preset conditions, and the debate result is obtained based on the first and second argument information, including:

[0032] When the degree of debate is less than the degree of debate threshold after the update, and the overall anchoring strength is greater than the strength threshold, the preset convergence condition is met, and the debate ends.

[0033] Find the point of consensus between the first and second arguments, and use it as the result of the debate.

[0034] The debate ends when the number of rounds equals the rounds threshold and the overall mismatch is greater than the mismatch threshold, satisfying the preset exhaustion condition.

[0035] Obtain the multiple rounds of argument records corresponding to each round of debate, as the debate result.

[0036] In one possible implementation, an intelligent strategy solution corresponding to the business requirements is obtained based on a decision-making strategy that matches the debate outcome, including:

[0037] When the debate result is the consensus point between the first and second argument information in a round of debate, the decision strategy that matches the debate result is the agent's decision strategy.

[0038] The decision-making intelligent agent obtains intelligent strategy solutions corresponding to the business requirements based on consensus points.

[0039] When the debate outcome is a record of multiple rounds of argumentation, the decision-making strategy that matches the debate outcome is a manual decision-making strategy.

[0040] Obtain the results of human arbitration based on multiple rounds of argumentation records, and use them as intelligent strategy solutions corresponding to business requirements.

[0041] In one possible implementation, based on business needs, retrieval data matching the financial business case is obtained from the database, including:

[0042] Feature extraction is performed on the business requirement content to obtain the retrieval vector of the business requirement content; the retrieval vector includes semantic features, legal keywords and similar case features;

[0043] Based on semantic features and similarity features, target cases that meet the similarity requirements are obtained from the case library of the database;

[0044] Based on legal keywords, target provisions that meet the relevance requirements are obtained from the regulatory database.

[0045] Based on semantic features, user information is obtained from the user profile database.

[0046] Based on the target case, target clause, and user information, search data matching financial business cases is obtained.

[0047] In one possible implementation, after retrieving search data matching the financial business case from the database based on business needs, the method further includes:

[0048] The retrieved data is subjected to perturbation testing based on a preset perturbation input to obtain perturbation results;

[0049] The difference between the disturbance result and the original disturbance result corresponding to the business requirement is calculated to obtain the difference result;

[0050] When the value of the difference result is greater than the difference threshold, an intelligent strategy solution corresponding to the business requirement content is obtained through manual arbitration.

[0051] Secondly, embodiments of this application provide an intelligent decision-making device based on an intelligent agent, comprising:

[0052] The acquisition unit is used to acquire the business requirements of financial business cases.

[0053] The data retrieval unit is used to retrieve data from the database that matches financial business cases based on business needs.

[0054] The agent debate unit is used to conduct multiple rounds of debate based on retrieved data through a first representative agent and a second representative agent to obtain the debate results.

[0055] The decision output unit is used to obtain intelligent strategy solutions corresponding to business requirements based on decision strategies that match the debate results.

[0056] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0057] The memory stores the instructions that the computer executes;

[0058] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0059] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0060] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0061] The intelligent decision-making method, device, and electronic device based on intelligent agents provided in this application firstly, through automated data retrieval based on business requirements, quickly locates relevant data, solving the problem that in complex financial business cases, data is often multi-dimensional and conflicting, and manual sorting is time-consuming and prone to omissions. Then, through a multi-round debate mechanism, a debate result is obtained. Based on the decision-making strategy matching the debate result, an intelligent strategy solution corresponding to the business requirements is obtained. In this way, automated data retrieval can complete data sorting and logical integration in a short time, compressing the case processing cycle. The established debate mechanism directly addresses the contradictions between data and seeks the optimal solution through iterative debate, thereby improving efficiency while ensuring the accuracy of the intelligent strategy solution. Therefore, this application, by constructing a data-driven, multi-agent debate, and logically interpretable automated decision-making framework, fundamentally solves the pain points of strong subjectivity, low efficiency, and difficulty in traceability in traditional decision-making, improving the processing efficiency of financial business decisions and the accuracy of corresponding intelligent strategy solutions. Attached Figure Description

[0062] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0063] Figure 1 A schematic diagram of an implementation environment provided for this application;

[0064] Figure 2 A flowchart illustrating the agent-based intelligent decision-making method provided in this application;

[0065] Figure 3 A flowchart illustrating the implementation of the agent-based intelligent decision-making method provided in this application;

[0066] Figure 4 A schematic diagram of the structure of the agent-based intelligent decision-making device provided in this application;

[0067] Figure 5 A schematic diagram of the structure of the electronic device provided in this application.

[0068] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0069] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0070] 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. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, and necessary confidentiality measures have been taken. They do not violate public order and good morals, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0071] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0072] Figure 1 This is a schematic diagram of an implementation environment provided by this application. The implementation environment includes a user terminal 10 and a server 20, which are connected via a wired or wireless connection.

[0073] Server 20 is used to obtain the business requirements of financial business cases transmitted by client 10; retrieve search data matching the financial business cases from the database based on the business requirements; conduct multiple rounds of debate based on the search data through a first representative agent and a second representative agent to obtain the debate results; and obtain an intelligent strategy solution corresponding to the business requirements based on the decision strategy matching the debate results.

[0074] It should be noted that, Figure 1In the implementation environment shown, server 20 can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. No restrictions are imposed here.

[0075] In existing technologies, traditional methods for handling financial business cases rely on manual arbitration or rule engines. However, with the increasing complexity of case types and the refinement of regulatory requirements, these methods suffer from significant shortcomings in decision-making transparency, dynamic decision-making capabilities, and rigorous argumentation. This leads to low efficiency in decision-making and processing of financial business cases, as well as unstable decision-making results. The intelligent decision-making method based on intelligent agents provided in this application addresses the pain points of strong subjectivity, low efficiency, and difficulty in traceability in traditional solutions for outputting decisions on financial business cases through an automated decision-making framework that includes data-driven approaches, multi-agent debate, and logically interpretable principles. This improves the processing efficiency of financial business cases and the accuracy of corresponding intelligent strategy solutions.

[0076] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0077] Figure 2 A flowchart illustrating the agent-based intelligent decision-making method provided in this application is shown below. Figure 2 As shown, the method includes:

[0078] S201, Business requirements for obtaining financial business cases.

[0079] The business requirements content can be various forms of business requirement information, such as requirement text.

[0080] In this embodiment, after obtaining a financial business case, the financial business case is analyzed to obtain the corresponding business requirements.

[0081] S202. Based on business needs, retrieve search data from the database that matches financial business cases.

[0082] The database can include multiple databases to store various types of information related to business needs. For example, it could include a case study database, a regulatory database, and a customer profile database. When financial business cases are related to loan risk assessment, the case study database could be a historical loan case database, storing information such as approval records, default cases, and repayment performance of similar loan projects. The regulatory database could be a risk control rules and regulations database, storing information such as loan guidelines from the State Financial Regulatory Commission, internal bank credit policies, and industry risk warnings. The customer profile database could store customer credit information and data archives.

[0083] In this embodiment, after extracting the business requirements of a financial business case, information matching the business requirements is retrieved from a pre-configured database to form retrieval data.

[0084] S203. Through the first representative agent and the second representative agent, multiple rounds of debate are conducted based on the retrieved data to obtain the debate results.

[0085] In this embodiment, the first and second representative agents represent simulated debaters from different perspectives, such as a customer rights representative and a risk control and compliance representative. In other embodiments, the first and second representative agents can be implemented by a single agent switching between multiple roles. For example, an LLM (Large Language Model) instance can alternate between the roles of "customer representative" and "risk control officer" using prompts, achieving multiple rounds of debate through internal simulated debate.

[0086] After the retrieval is completed and the retrieval data matching the financial business case is obtained, this embodiment conducts multiple rounds of debate based on the retrieval data through a preset debate mechanism, a first representative agent and a second representative agent, to simulate the multi-perspective balancing mechanism in the real decision-making process and obtain the debate results.

[0087] S204. Based on the decision-making strategy that matches the debate results, obtain an intelligent strategy solution corresponding to the business requirements.

[0088] In this embodiment, the decision-making strategy corresponds to a variety of intelligent strategy scheme generation strategies. After obtaining the debate results through multiple rounds of debate, the strategy that best matches the debate results is selected from the decision-making strategies, thereby forming an intelligent strategy scheme.

[0089] The intelligent decision-making method based on intelligent agents provided in this application firstly locates relevant data quickly through automated data retrieval based on business requirements, solving the problem that in complex financial business cases, data is often multi-dimensional and conflicting, and manual sorting is time-consuming and prone to omissions. Then, a multi-round debate mechanism is used to obtain the debate results. Based on the decision-making strategy matching the debate results, an intelligent strategy solution corresponding to the business requirements is obtained. In this way, automated data retrieval can complete data sorting and logical integration in a short time, compressing the case processing cycle. The established debate mechanism directly addresses the contradictions between data and seeks the optimal solution through iterative debate, thereby improving efficiency while ensuring the accuracy of the intelligent strategy solution.

[0090] Furthermore, regarding the multi-round debate mechanism, on the one hand, a dialectical debate framework is constructed by introducing a first representative agent (such as a customer rights representative) and a second representative agent (such as a risk control and compliance representative). The two agents debate from different perspectives, simulating the multi-perspective balancing mechanism in real-world decision-making. This ensures that the final debate result is no longer a conclusion from a single perspective, but rather an equilibrium output formed after multiple rounds of adversarial testing. This significantly reduces the impact of individual subjectivity on decision-making and improves the consistency of intelligent strategy solutions for similar cases.

[0091] On the other hand, the "multi-round debate" is not a black-box process, but rather a structured and traceable chain of reasoning based on retrieved data. In this way, the final intelligent strategy solution can be linked to the key data and reasoning paths used in the debate process, forming an auditable "data-claim-conclusion" mapping. This makes the decision outcome no longer an unknowable output, but a result supported by transparent logic, greatly enhancing users' (clients, regulators) acceptance and trust in the intelligent decision-making solution.

[0092] Furthermore, this application abstracts the decision-making process for financial business cases into a standardized computational framework of "retrieval-debate-decision," transforming a non-standard process that originally relied on human experience into a repeatable and quantifiable algorithmic process. This allows for the simultaneous processing of multiple financial business cases while ensuring that each case undergoes the same rigorous process. This not only reduces institutions' reliance on scarce expert resources but also provides a technological foundation for the large-scale and intelligent operation of banking fintech-related businesses, making it particularly suitable for various high-frequency business scenarios.

[0093] In an exemplary embodiment of this application, the step of obtaining a debate result by conducting multiple rounds of debate based on retrieved data through a first representative intelligent agent and a second representative intelligent agent may specifically include:

[0094] In each round of debate, the first representative agent sends the first argument information, which contains a subset of data. The data in the subset is either the tendency data corresponding to the first representative agent in the retrieved data, or new data obtained based on the argument information fed back by the second representative agent in the previous round of debate.

[0095] Based on the first argument information, the degree of debate between the first representative agent and the second representative agent is dynamically updated to obtain the updated degree of debate.

[0096] The second representative intelligent agent generates second argument information based on the first argument information. The second argument information includes rebuttal data corresponding to the data in the data subset.

[0097] Based on the first and second argumentation information, the comprehensive mismatch degree and comprehensive anchoring strength are obtained;

[0098] The debate ends when the updated degree of contention, overall mismatch, and overall anchoring strength meet the preset conditions, and the debate result is obtained based on the first and second argument information.

[0099] Among them, "degree of contention" refers to the degree to which an agent adheres to its own position and is unwilling to compromise during a debate. "Mismatch degree" refers to the robustness of the data. "Anchoring strength" is a dimensionless value that establishes a quantitative "rational decision-making phase transition point" with a corresponding strength threshold, transforming the originally vague judgment of "when to automatically output a decision plan" that relies on human experience into a clear and calculable scientific threshold. The preset condition is the convergence condition for multi-round debates; iterative debates stop when the condition is met.

[0100] In this embodiment, during the first round of debate after obtaining retrieval data matching the financial business case, tendency data corresponding to the first representative agent is extracted from the retrieval data and used as a subset of data included in the first argument information. Then, the first representative agent sends the first argument information to the second representative agent, and based on the first argument information, the degree of argument between the first and second representative agents is dynamically updated to obtain the updated degree of argument. The second representative agent generates rebuttal data corresponding to the data in the data subset, forming the second argument information. In each subsequent round of debate, the subset of data included in the first argument information corresponding to the first representative agent is new data obtained based on the argument information fed back by the second representative agent in the previous round of debate.

[0101] In each round of debate, the comprehensive mismatch degree and comprehensive anchoring strength are obtained based on the first and second argument information. The debate ends when the updated debate degree, comprehensive mismatch degree, and comprehensive anchoring strength meet the preset conditions, and the debate result is obtained based on the first and second argument information.

[0102] Thus, through the above embodiments, this application achieves automation, rationalization, and rapid convergence of multi-agent debate by constructing a closed-loop debate process that includes dynamic debate intensity updates, adversarial argument generation, and multi-dimensional quantitative evaluation (comprehensive mismatch degree and comprehensive anchoring strength). Specifically, dynamically adjusting the debate intensity based on data quality ensures the adaptability of the debate behavior, adversarial data exchange promotes the in-depth evolution of the argument, and the quantitative evaluation of comprehensive mismatch degree and anchoring strength provides an objective and verifiable criterion for terminating the debate, ultimately enabling the efficient production of results supported by strong data.

[0103] In another exemplary embodiment, the updated dispute degree mentioned in the above embodiment includes a first dispute degree corresponding to a first representative agent and a second dispute degree corresponding to a second representative agent; correspondingly, the step of dynamically updating the dispute degrees of the first and second representative agents based on the first argument information to obtain the updated dispute degree may specifically include:

[0104] When the data in the data subset included in the first argument information is the tendency data corresponding to the first representative agent in the retrieved data, the initial degree of argument corresponding to the first representative agent is taken as the first degree of argument.

[0105] When the data in the data subset is new data obtained based on the argumentation information fed back by the second representative agent in the previous round of debate, the first effective support degree and the first mismatch degree of the new data are obtained through the first representative agent;

[0106] Based on the first effective support and the first mismatch, the first degree of controversy is obtained;

[0107] The second representative agent obtains the second effective support and the second mismatch of the data in the data subset.

[0108] The second degree of contention is derived based on the second effective support degree and the second mismatch degree.

[0109] In this context, the first and second representative agents each have an initial level of contention. In the first round of debate, the initial level of contention is dynamically updated. In subsequent rounds of debate, the level of contention generated in the previous round is dynamically updated.

[0110] In addition to the data subset, the first argument information also includes the conclusions corresponding to the data in the data subset. Correspondingly, valid support refers to the degree of match between the data and the conclusions.

[0111] In this embodiment, when calculating the corresponding second degree of contention through the second representative agent B, the second effective support and second mismatch of the data in the data subset are first obtained. Specifically, the second effective support is ρd (activation degree of E_A in B's context), which means mapping the data E_A of the first representative agent A to B's position and knowledge system, and evaluating the degree to which these data support or influence B's original viewpoint.

[0112] In this embodiment, B's knowledge system can be the tendency data corresponding to the second representative agent in the retrieved data, or the data contained in the argument information fed back by the second representative agent in the previous round of debate. B's position can be B's main claim, which is obtained through the data corresponding to the knowledge system: claim = argmax(ρd_data), which means taking the maximum value among the effective support between all data and the corresponding conclusion, and determining the conclusion corresponding to the maximum value as B's main claim.

[0113] For example, A sends argument EA to B: Data 1: Article 20 of the "xxx Policy" (ρd=0.9, clearly stated in the original text). Data 2: Similar case "The client had a default record but still got approved" (ρd=0.8, details are vague).

[0114] B. Contextual shift:

[0115] Data 1: B believes that the policy only applies when "the customer's reason for breach of contract can be appealed" → activation level drops to 0.6.

[0116] Data 2: B found that the bank had a system failure in this case → inconsistent with the current case, and the activation rate dropped to 0.3.

[0117] Weighted average (assuming the two data points have equal weights): ρd (activation degree) = (0.6 + 0.3) / 2 = 0.45.

[0118] The second mismatch degree is dr (E_A perturbation test), which is obtained by performing a self-perturbation test on the data E_A. For example, a slight semantic perturbation (such as restating or deleting details) is performed on each data in E_A, and the difference between the conclusions corresponding to the data before and after the perturbation is calculated as the mismatch degree of that data, thus obtaining the second mismatch degree corresponding to the data subset.

[0119] Secondly, the anchoring score S_A→B is calculated based on the second effective support and the second mismatch: S_A→B=ρd(activation degree of E_A in the context of B)-dr(perturbation test of E_A).

[0120] The final update received the second highest level of controversy. β=0.3. Wherein, β represents the degree of contention corresponding to the second agent in the previous round, and β is the learning rate coefficient.

[0121] It should be noted that the process of obtaining the first degree of contention for the first representative agent A is similar. First, the first effective support and first mismatch of the new data in the context of A in the argument information sent by the second representative agent are calculated. From this, the corresponding anchoring score SB→A is obtained. Then, the updated first degree of contention is obtained based on the dynamic update formula and the learning rate coefficient. .

[0122] Thus, through the above embodiments, this application enables the dynamic debate degree update mechanism based on data quality quantitative assessment to adaptively adjust the agent's position persistence according to the validity and stability of the opponent's argument, thereby significantly improving the convergence efficiency and decision quality of multi-agent debate, and ensuring that the debate process quickly focuses on core data and leads to rational consensus.

[0123] In another exemplary embodiment, the step of obtaining the comprehensive mismatch degree and comprehensive anchoring strength based on the first and second argumentation information may specifically include:

[0124] To obtain the consensus point between the first and second argument information;

[0125] Perform self-disturbance tests on the consensus data corresponding to each consensus point to obtain the data mismatch degree corresponding to each consensus data point;

[0126] A comprehensive mismatch is obtained based on all the data mismatches.

[0127] The effective support of each consensus data point is obtained through a support calculation strategy; the support calculation strategy includes the adjudication cluster size and the clause matching degree.

[0128] A comprehensive effective support score is obtained based on the effective support of all data.

[0129] The overall anchoring strength is obtained by using the anchoring strength formula, based on the overall mismatch and overall effective support.

[0130] The overall mismatch degree and overall anchoring strength can be obtained through a decision-making agent. The decision-making agent, along with the first and second representative agents, adopts a MAI (Multi-Agent Collaborative Intelligence) architecture, which includes a three-layer decision-making structure.

[0131] (1) Bait layer: The debate intensity of the agent is controlled by the behavior regulation parameter αc(0-1);

[0132] (2) Filtering layer: CRIT (Critical Reasoning Interrogative Tribunal) judge, used to evaluate the clarity, consistency, data support and falsifiability of the argument;

[0133] (3) Persistent layer: A memory system with transactional characteristics that stores assertion trees with anchored scores and regulatory audit trails.

[0134] In this system, the first and second representative agents are located in the bait layer. The CRIT judge in the filtering layer performs a four-quadrant test (clarity, consistency, data support, and falsifiability) on the first and second argument information; if any one of these tests falls below a threshold, the system blocks the query and returns a corrected query. The decision agent is located in the persistence layer.

[0135] The decision-making agent possesses transactional characteristics in the persistence layer, used to store assertion trees with anchored scores and regulatory audit trails, specifically:

[0136] Generate a structured assertion tree, which clearly and completely shows each step of the reasoning process from business requirements (root node) to the final decision (intelligent strategy solution) in a tree structure.

[0137] Calculate the anchor score S_evi = ρd_evi - dr_evi for each data point in the consensus data, where ρd_evi is the effective support of each data point for the conclusion corresponding to the consensus point, and dr_evi is the mismatch of each data point.

[0138] Generate a regulatory audit trail, including: the original text and ID of the cited regulatory provisions, the case number and ruling of similar precedents, the curve of changes in the agent's degree of contention, and a snapshot of the S-value (overall anchoring strength) at the moment when the intelligent strategy solution is decided.

[0139] In other embodiments, the SagaLLM persistence layer can be removed, each round of debate can run independently, and the final results can be log-aggregated.

[0140] In this embodiment, a decision-making agent obtains the comprehensive mismatch degree and comprehensive anchoring strength based on the first and second argumentation information. Specifically, the consensus point between the first and second argumentation information is first obtained. The consensus point can be a conclusion or data, such as a conclusion or fact that both parties explicitly agree on (e.g., "the customer has a high credit score"); or both parties citing the same data without disagreement (e.g., both parties agree on the validity of a certain legal provision). After determining the consensus point, a self-perturbation test is performed on each consensus data corresponding to the consensus point to obtain the data mismatch degree corresponding to each consensus data; the comprehensive mismatch degree is obtained based on all the data mismatch degrees.

[0141] Through the support calculation strategy, obtain the data effective support for each consensus data; the support calculation strategy includes the adjudication cluster size and the clause matching degree. In this embodiment, through the support calculation strategy, calculate the scores of each item of the preset type of each consensus data, and then comprehensively obtain the data effective support. The scores of each item can include the authority score: the legal / official effectiveness level of the data source; the relevance score: the semantic relevance between the data content and the consensus point; the timeliness score: whether the data is currently valid or outdated; the specificity score: the clarity and refinement degree of the data content. Then, based on all the data effective supports, obtain the comprehensive effective support.

[0142] Finally, through the anchoring strength formula, based on the comprehensive mismatch degree and the comprehensive effective support, obtain the comprehensive anchoring strength. The comprehensive anchoring strength formula can be expressed as: S = ρd - dr - γlogk.

[0143] Among them, ρd represents the comprehensive effective support, dr represents the comprehensive mismatch degree, k represents the anchoring budget, γ represents the adaptive regularization parameter, γ = 0.2 (during the economic stability period), which is usually a fixed value and can be dynamically adjusted according to the macroeconomic index after a period of time. The anchoring budget k refers to the maximum number of data units allowed, which is used to control the calculation cost and information overload. The anchoring budget k can be a fixed value, for example, uniformly set to 10. It can also be dynamically adjusted according to the case risk level: for ordinary risk cases, k ≤ 10; for major risk cases, k ≤ 20. And it dynamically expands with the comprehensive mismatch degree dr. If the comprehensive anchoring strength is less than the strength threshold and k < kmax after a certain round, trigger the "need more data" branch, expand the data retrieval, and increase the k value. In the next round of debate, the anchoring budget is the increased k.

[0144] The anchoring strength formula adopts the UCCT (Unified Contextual Control Theory) theoretical framework, which is used to quantify the combination strength between external constraints (such as legal provisions and business rules) and the internal pattern library of the large-scale language model.

[0145] In addition, the decision-making agent is also set with an initial argument degree (αc_C = 0.3). When the debate reaches a stalemate (such as the comprehensive mismatch degree is too high or the round is close to T_max), αc_C can be temporarily increased to 0.6, so that C takes the lead and actively proposes a compromise plan.

[0146] In this way, through the above embodiments of the present application, by constructing a refined calculation framework that starts from the consensus data, quantitatively evaluates its stability (comprehensive mismatch degree) and support strength (comprehensive effective support), and finally aggregates into the comprehensive anchoring strength, the dual objective measurement of the stability and confidence of the debate result is realized, ensuring that the final intelligent strategy plan is based on a solid and stable data consensus.

[0147] In another exemplary embodiment, the debate terminates when the updated degree of contention, overall mismatch degree, and overall anchoring strength meet preset conditions. The step of obtaining the debate result based on the first argument information and the second argument information may specifically include:

[0148] When the degree of debate is less than the degree of debate threshold after the update, and the overall anchoring strength is greater than the strength threshold, the preset convergence condition is met, and the debate ends.

[0149] Find the point of consensus between the first and second arguments, and use it as the result of the debate.

[0150] The debate ends when the number of rounds equals the rounds threshold and the overall mismatch is greater than the mismatch threshold, satisfying the preset exhaustion condition.

[0151] Obtain the multiple rounds of argument records corresponding to each round of debate, as the debate result.

[0152] In an exemplary embodiment, the initial debate scores of the first representative agent and the second representative agent are 0.6 and 0.85, respectively, and the debate score thresholds are 0.3 and 0.5, respectively. It is understood that the updated debate scores can only be comparable to the debate score thresholds after at least two rounds of debate. In the first round of debate, the first representative agent uses the initial values; therefore, it is necessary to determine whether preset conditions are met after multiple rounds of debate. Preset conditions may include preset convergence conditions and preset exhaustion conditions.

[0153] Condition 1 (preset convergence condition): αc_A<0.3 and αc_B<0.5 and the overall anchoring strength is greater than the strength threshold. The consensus point between the first argument information and the second argument information is the debate result.

[0154] Condition 2 (preset exhaustion condition): t=T (maximum number of rounds) but the overall mismatch degree dr>0.3, marked as "requires manual arbitration", the transaction memory stores the record of each round of debate as the debate result for reference.

[0155] In addition, the preset conditions may also include insufficient anchoring conditions. When the overall anchoring strength is greater than the strength threshold and the dynamic anchoring budget k has reached k_max, the "expert review" process under manual arbitration is triggered.

[0156] Thus, through the above embodiments, this application ensures that the multi-agent debate process can end in a timely manner when a high-quality consensus is reached to improve efficiency, or be decisively terminated when a deadlock is reached to control costs, by using a clear, multi-indicator joint determination debate termination mechanism (including a "convergence condition" based on the degree of argumentation and anchoring strength and a "depletion condition" based on the number of rounds and mismatch). This achieves an optimized balance between efficiency and quality in dynamic debate and ensures that the output results (consensus points or multi-round records) always have clear decision-making basis and auditable value.

[0157] In another exemplary embodiment provided in this application, the step of obtaining an intelligent strategy solution corresponding to the business requirements based on a decision strategy matching the debate result after obtaining the debate result in the above embodiment may specifically include:

[0158] When the debate result is the consensus point between the first and second argument information in a round of debate, the decision strategy that matches the debate result is the agent's decision strategy.

[0159] The decision-making intelligent agent obtains intelligent strategy solutions corresponding to the business requirements based on consensus points.

[0160] When the debate outcome is a record of multiple rounds of argumentation, the decision-making strategy that matches the debate outcome is a manual decision-making strategy.

[0161] Obtain the results of human arbitration based on multiple rounds of argumentation records, and use them as intelligent strategy solutions corresponding to business requirements.

[0162] Thus, through the above embodiments, this application achieves seamless integration and optimal division of labor between automation and human intervention by using a dual-path decision-making strategy (agent adjudication or human arbitration) that intelligently matches the debate outcome status (successful consensus or deadlock record). This ensures that routine cases that can reach a consensus can be handled efficiently to leverage the effectiveness of automation, while complex disputed cases can be accurately transferred to human intervention with complete structured debate records as decision support. This improves overall processing efficiency while ensuring the reliability, compliance, and acceptability of the final intelligent strategy solution for all types of financial business cases.

[0163] In an exemplary embodiment provided in this application, the step of retrieving search data from a database that matches a financial business case based on business requirements may specifically include:

[0164] Feature extraction is performed on the business requirement content to obtain the retrieval vector of the business requirement content; the retrieval vector includes semantic features, legal keywords and similar case features;

[0165] Based on semantic features and similarity features, target cases that meet the similarity requirements are obtained from the case library of the database;

[0166] Based on legal keywords, target provisions that meet the relevance requirements are obtained from the regulatory database.

[0167] Based on semantic features, user information is obtained from the user profile database.

[0168] Based on the target case, target clause, and user information, search data matching financial business cases is obtained.

[0169] In this embodiment, business requirement content X is obtained, such as X{customer ID, business type, amount involved, customer level, regulatory referral mark}. Then, features are extracted from business requirement content X to form a retrieval vector Q=[X, legal keywords, similar case features]. Next, based on semantic features X and similar case features, target cases meeting similarity requirements are obtained from the case library of the database; based on legal keywords, target provisions meeting relevance requirements are obtained from the regulatory library of the database; based on semantic features X, user information, such as CLV (Customer Lifetime Value) and historical security records, is obtained from the user profile library of the database. Finally, based on the target cases, target provisions, and user information, retrieval data matching financial business cases is obtained.

[0170] Thus, through the above embodiments, this application achieves efficient and comprehensive location of structured data highly relevant to current financial business cases from massive databases by integrating a precise retrieval mechanism that integrates multi-dimensional features (semantics, law, cases, user profiles). This provides a solid, multi-dimensional, and highly relevant data foundation for subsequent multi-agent debates, thereby significantly improving the decision-making accuracy, efficiency, and interpretability of the entire decision-making system.

[0171] In another exemplary embodiment, after retrieving search data matching the financial business case from the database based on business needs, an intermediate decision-making step is further included, which may specifically include:

[0172] The retrieved data is subjected to perturbation testing based on a preset perturbation input to obtain perturbation results;

[0173] The difference between the disturbance result and the original disturbance result corresponding to the business requirement is calculated to obtain the difference result;

[0174] When the value of the difference result is greater than the difference threshold, an intelligent strategy solution corresponding to the business requirement content is obtained through manual arbitration.

[0175] In this embodiment, perturbation tests are performed on the retrieved data from the database to determine whether a financial business case is a "highly controversial case" requiring intermediate decision-making. First, perturbed retrieval data is obtained based on preset perturbation inputs and retrieval data. When the number of perturbation inputs m=5, it can include X1-X5.

[0176] X1: Emotional de-biasing rewriting (removal of strong words);

[0177] X2: Timing disturbance (change "3 AM" to "abnormal time period");

[0178] X3-X5: Randomly remove 1-2 data units (the smallest information fragment obtained from semantic segmentation or fact extraction in the retrieved data).

[0179] Effective support is calculated on the perturbed search data, where effective support represents the relevance of the data to the business requirements. Specifically, the majority decision cluster size |Cmaj| is calculated using similar target cases. Clause matching is calculated using relevant target clauses. The normalized CLV value is obtained using user information such as Customer Lifetime Value (CLV) and historical security records. Finally, the effective support is calculated using the following formula:

[0180] Ρd i =w1·|Cmaj| / N+w2·clause matching degree+w3·CLV normalized value;

[0181] Where w1=0.5, w2=0.3, w3=0.2, and i corresponds to the number of disturbance inputs. When the number of disturbance inputs m=5, i=1, 2, 3, 4, 5.

[0182] The disturbance result S was obtained by calculating the anchoring strength formula. i : , where dr i The value is zero, k represents the anchor budget, and γ represents the adaptive regularization parameter. The anchor budget k can be obtained based on the business requirements. For example, if the business requirements represent an amount greater than 100,000 yuan or there is a regulatory referral flag, k=15; otherwise, it is 6.

[0183] The difference between the disturbance result and the original disturbance result corresponding to the business requirement is calculated to obtain the difference result dr'. Where S0 is the original perturbation result, and m is the number of perturbation inputs.

[0184] When the value of the difference result is greater than the difference threshold (e.g., 0.5), the financial business case is marked as a "high-dispute case", and an intelligent strategy solution corresponding to the business requirements is obtained through manual arbitration.

[0185] Thus, through the above embodiments, this application introduces a disturbance testing and difference quantification evaluation mechanism based on retrieved data, which can automatically identify and screen "highly controversial" or "highly uncertain" cases that are highly sensitive to input information (or data representation) in the early stages of the decision-making process, and directly transfer them to manual arbitration. This effectively avoids investing unstable and volatile cases into the subsequent time-consuming multi-agent debate process, and significantly improves the resource utilization efficiency and decision reliability of the entire system while ensuring the quality of high-risk case processing.

[0186] In an exemplary embodiment of the present application, when implementing the agent-based intelligent decision-making method provided by the present application in a software system, it is implemented through an input layer, a core processing unit, and an output layer. Please refer to Figure 3 , Figure 3 which is a schematic flowchart for implementing the agent-based intelligent decision-making method provided by the present application. As Figure 3 shown, the steps are as follows:

[0187] Requirement text X and case feature extraction: Input a financial business case, extract the requirement text as the content of business requirements, and extract key information from the text.

[0188] Data retrieval and initialization of the anchored budget allocation k value: According to the case features, obtain the retrieval features from the database; allocate the anchored budget k according to the case severity and initialize the k value.

[0189] MACI three-agent debate. Initialize three agents: Agent A (customer rights representative), argument degree αc = 0.6, Agent B (risk control and compliance officer), argument degree αc = 0.85, Agent C (decision maker), argument degree αc = 0.3. Multiple rounds of debate and data exchange: A and B conduct multiple rounds of debate based on the retrieved data, exchange arguments and data, and C is responsible for recording the consensus points into the transaction memory.

[0190] CRIT judge four-quadrant test: Each round of argument is submitted to the CRIT judge for a four-quadrant test (clarity, consistency, data support, falsifiability). If the test passes, proceed to the next step. If the test fails, return to the debate session to correct the argument.

[0191] Persistence of the transaction memory SagaLLM: The passed arguments and consensus points are stored in the SagaLLM transaction memory.

[0192] UCCT anchored strength S calculation.

[0193] Threshold determination: Determine whether S > θ (θ is the threshold, default 0.5). If S ≥ θ and dr < 0.3, output and generate a decision letter (including the S value and the complete data chain). If S < θ but k < k_max (the upper limit of the number of data has not been reached), return to the data retrieval link to expand the data. If S < θ and k ≥ k_max or the number of rounds exceeds the limit, transfer to manual arbitration and attach a snapshot of the transaction memory.

[0194] Figure 4 is a schematic structural diagram of the agent-based intelligent decision-making device provided by the present application. As Figure 4 shown, the agent-based intelligent decision-making device 40 includes:

[0195] An acquisition unit 401, configured to acquire the business requirement content of a financial business case;

[0196] The data retrieval unit 402 is used to retrieve data from the database that matches financial business cases based on business needs.

[0197] The agent debate unit 403 is used to conduct multiple rounds of debate based on retrieved data through a first representative agent and a second representative agent to obtain the debate results.

[0198] The decision output unit 404 is used to obtain an intelligent strategy solution corresponding to the business requirements based on the decision strategy that matches the debate results.

[0199] In one possible implementation, the agent debate unit 403 is further configured to send first argument information through a first representative agent in each round of debate. The first argument information includes a data subset, the data in the data subset being either the tendency data corresponding to the first representative agent in the retrieved data, or new data obtained based on the argument information fed back by the second representative agent in the previous round of debate.

[0200] Based on the first argument information, the degree of debate between the first representative agent and the second representative agent is dynamically updated to obtain the updated degree of debate.

[0201] The second representative intelligent agent generates second argument information based on the first argument information. The second argument information includes rebuttal data corresponding to the data in the data subset.

[0202] Based on the first and second argumentation information, the comprehensive mismatch degree and comprehensive anchoring strength are obtained;

[0203] The debate ends when the updated degree of contention, overall mismatch, and overall anchoring strength meet the preset conditions, and the debate result is obtained based on the first and second argument information.

[0204] In one possible implementation, the updated debate degree includes the first debate degree corresponding to the first representative agent and the second debate degree corresponding to the second representative agent; the agent debate unit 403 is also used to take the initial debate degree corresponding to the first representative agent as the first debate degree when the data in the data subset contained in the first argument information is the tendency data corresponding to the first representative agent in the search data.

[0205] When the data in the data subset is new data obtained based on the argumentation information fed back by the second representative agent in the previous round of debate, the first effective support degree and the first mismatch degree of the new data are obtained through the first representative agent;

[0206] Based on the first effective support and the first mismatch, the first degree of controversy is obtained;

[0207] The second representative agent obtains the second effective support and the second mismatch of the data in the data subset.

[0208] The second degree of contention is derived based on the second effective support degree and the second mismatch degree.

[0209] In one possible implementation, the agent debate unit 403 is also used to obtain a consensus point between the first argument information and the second argument information;

[0210] Perform self-disturbance tests on the consensus data corresponding to each consensus point to obtain the data mismatch degree corresponding to each consensus data point;

[0211] A comprehensive mismatch is obtained based on all the data mismatches.

[0212] The effective support of each consensus data point is obtained through a support calculation strategy; the support calculation strategy includes the adjudication cluster size and the clause matching degree.

[0213] A comprehensive effective support score is obtained based on the effective support of all data.

[0214] The overall anchoring strength is obtained by using the anchoring strength formula, based on the overall mismatch and overall effective support.

[0215] In one possible implementation, the agent debate unit 403 is further configured to terminate the debate when the debate degree is less than the debate degree threshold after the update and the overall anchoring strength is greater than the strength threshold, thus satisfying a preset convergence condition.

[0216] Find the point of consensus between the first and second arguments, and use it as the result of the debate.

[0217] The debate ends when the number of rounds equals the rounds threshold and the overall mismatch is greater than the mismatch threshold, satisfying the preset exhaustion condition.

[0218] Obtain the multiple rounds of argument records corresponding to each round of debate, as the debate result.

[0219] In one possible implementation, the decision output unit 404 is also used to determine the agent's decision strategy when the debate result is a consensus point between the first and second argument information in a round of debate.

[0220] The decision-making intelligent agent obtains intelligent strategy solutions corresponding to the business requirements based on consensus points.

[0221] When the debate outcome is a record of multiple rounds of argumentation, the decision-making strategy that matches the debate outcome is a manual decision-making strategy.

[0222] Obtain the results of human arbitration based on multiple rounds of argumentation records, and use them as intelligent strategy solutions corresponding to business requirements.

[0223] In one possible implementation, the data retrieval unit 402 is further used to extract features from the business requirement content to obtain a retrieval vector of the business requirement content; the retrieval vector includes semantic features, legal keywords, and similar case features;

[0224] Based on semantic features and similarity features, target cases that meet the similarity requirements are obtained from the case library of the database;

[0225] Based on legal keywords, target provisions that meet the relevance requirements are obtained from the regulatory database.

[0226] Based on semantic features, user information is obtained from the user profile database.

[0227] Based on the target case, target clause, and user information, search data matching financial business cases is obtained.

[0228] In one possible implementation, the device further includes an intermediate adjustment unit, which, after retrieving search data from the database that matches the financial business case based on the business requirements, performs a perturbation test on the search data based on a preset perturbation input to obtain the perturbation result.

[0229] The difference between the disturbance result and the original disturbance result corresponding to the business requirement is calculated to obtain the difference result;

[0230] When the value of the difference result is greater than the difference threshold, an intelligent strategy solution corresponding to the business requirement content is obtained through manual arbitration.

[0231] The intelligent decision-making device based on intelligent agents provided in this embodiment can execute the methods provided in the above method embodiments. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0232] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0233] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0234] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0235] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0236] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0237] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0238] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0239] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0240] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0241] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0242] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0243] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0244] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0245] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0246] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0247] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An intelligent decision-making method based on intelligent agents, characterized in that, include: The business requirements for obtaining financial business case studies; Based on the aforementioned business requirements, search data matching the aforementioned financial business case is retrieved from the database. The first and second representative agents conduct multiple rounds of debate based on the retrieved data to obtain the debate results. Based on the decision-making strategy that matches the debate results, an intelligent decision-making solution corresponding to the business requirement is obtained.

2. The agent-based intelligent decision-making method according to claim 1, characterized in that, The process involves multiple rounds of debate based on the retrieved data, conducted by a first representative agent and a second representative agent, to obtain the debate results, including: In each round of debate, the first representative agent sends first argument information, which includes a subset of data. The data in the subset of data is the tendency data corresponding to the first representative agent in the retrieved data, or is new data obtained based on the argument information fed back by the second representative agent in the previous round of debate. Based on the first argument information, the degree of dispute between the first representative agent and the second representative agent is dynamically updated to obtain the updated degree of dispute. The second representative agent generates second argument information based on the first argument information, and the second argument information includes rebuttal data corresponding to the data in the data subset. Based on the first and second argumentation information, the comprehensive mismatch degree and comprehensive anchoring strength are obtained; The debate terminates when the updated degree of contention, overall mismatch, and overall anchoring strength meet preset conditions, and the debate result is obtained based on the first argument information and the second argument information.

3. The agent-based intelligent decision-making method according to claim 2, characterized in that, The updated debate degree includes the first debate degree corresponding to the first representative agent and the second debate degree corresponding to the second representative agent; The step of dynamically updating the dispute degree of the first representative agent and the second representative agent based on the first argument information to obtain the updated dispute degree includes: When the data in the data subset included in the first argument information is the tendency data corresponding to the first representative agent in the retrieved data, the initial degree of argument corresponding to the first representative agent is taken as the first degree of argument. When the data in the data subset is new data obtained based on the argumentation information fed back by the second representative agent in the previous round of debate, the first effective support degree and the first mismatch degree of the new data are obtained through the first representative agent; Based on the first effective support and the first mismatch, the first degree of contention is obtained; The second representative agent obtains the second effective support and the second mismatch of the data in the data subset. Based on the second effective support degree and the second mismatch degree, the second controversy degree is obtained.

4. The agent-based intelligent decision-making method according to claim 2, characterized in that, The process of obtaining the comprehensive mismatch degree and comprehensive anchoring strength based on the first and second argumentation information includes: Obtain the consensus point between the first argument information and the second argument information; Perform a self-disturbance test on the consensus data corresponding to each consensus point to obtain the data mismatch degree corresponding to each consensus data; A comprehensive mismatch is obtained based on all the aforementioned data mismatches; The effective support of each consensus data point is obtained through a support calculation strategy; the support calculation strategy includes the adjudication cluster size and the clause matching degree. A comprehensive effective support score is obtained based on the effective support scores of all the aforementioned data. The overall anchoring strength is obtained using the anchoring strength formula, based on the overall mismatch and the overall effective support.

5. The agent-based intelligent decision-making method according to claim 2, characterized in that, The debate terminates when the updated degree of contention, overall mismatch, and overall anchoring strength meet preset conditions. The debate result is obtained based on the first and second argument information, including: When the updated degree of contention is less than the degree of contention threshold and the comprehensive anchoring strength is greater than the strength threshold, the preset convergence condition is met, and the debate ends. Obtain the consensus point between the first and second argument information as the debate result; When the number of rounds of debate equals the rounds threshold and the overall mismatch degree is greater than the mismatch threshold, a preset exhaustion condition is met, and the debate terminates. Obtain the multiple rounds of argument records corresponding to each round of debate, as the debate result.

6. The agent-based intelligent decision-making method according to claim 5, characterized in that, The intelligent strategy solution corresponding to the business requirement content is obtained based on the decision-making strategy matching the debate result, including: When the debate result is a consensus point between the first argument information and the second argument information in a round of debate, the decision strategy that matches the debate result is the agent decision strategy. The decision-making agent obtains the intelligent strategy solution corresponding to the business requirement based on the consensus point. When the debate result is a record of multiple rounds of argumentation, the decision-making strategy that matches the debate result is a manual decision-making strategy. Obtain the human arbitration result based on the multi-round argumentation records, and use it as the intelligent strategy solution corresponding to the business requirement.

7. The agent-based intelligent decision-making method according to any one of claims 1 to 6, characterized in that, The step of retrieving data from the database that matches the financial business case based on the business requirements includes: Feature extraction is performed on the business requirement content to obtain a retrieval vector for the business requirement content; the retrieval vector includes semantic features, legal keywords, and similar case features; Based on the semantic features and the similarity features, target cases that meet the similarity requirements are obtained from the case library of the database; Based on the legal keywords, target provisions that meet the relevance requirements are obtained from the regulatory database of the database; Based on the semantic features, user information is obtained from the user profile database of the database; Based on the target case, the target clause, and the user information, search data matching the financial business case is obtained.

8. The agent-based intelligent decision-making method according to claim 7, characterized in that, After retrieving search data from the database that matches the financial business case based on the stated business requirements, the process further includes: The retrieved data is subjected to a perturbation test based on a preset perturbation input to obtain the perturbation result; The difference between the disturbance result and the original disturbance result corresponding to the business requirement content is calculated to obtain the difference result; When the value of the difference result is greater than the difference threshold, an intelligent strategy solution corresponding to the business requirement content is obtained through manual arbitration.

9. An intelligent decision-making device based on an intelligent agent, characterized in that, include The acquisition unit is used to acquire the business requirements of financial business cases. The data retrieval unit is used to retrieve search data from the database that matches the financial business case based on the business requirements. The agent debate unit is used to conduct multiple rounds of debate based on the retrieved data through a first representative agent and a second representative agent to obtain the debate results. The decision output unit is used to obtain an intelligent strategy solution corresponding to the business requirement content based on the decision strategy that matches the debate result.

10. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the electronic device to implement the agent-based intelligent decision-making method as described in any one of claims 1 to 8.

Images