Mail processing method, device and system related to off-exchange derivatives

Abstract

This application provides a method, apparatus, and system for processing emails related to over-the-counter (OTC) derivatives. The method uses a trading intelligence agent composed of multiple primary models to perform precise intelligent analysis of emails related to OTC derivatives and dynamically adjusts the model weights corresponding to the primary models. When a first preset cumulative weight is not less than a first preset cumulative threshold, the output result of the trading intelligence agent is determined directly using the first processing result of the primary model corresponding to the model weight used to calculate the current first preset cumulative weight. This eliminates the need to wait for the response of the primary model before receiving the first processing result, thus ensuring the accuracy of the trading intelligence agent's output while significantly improving response speed, ultimately enhancing the accuracy and speed of email recognition related to OTC derivatives.

Description

Technical Field

[0001] This application relates to the field of email processing technology related to over-the-counter derivatives, and in particular to a method, apparatus and system for email processing related to over-the-counter derivatives. Background Technology

[0002] Over-the-counter (OTC) derivatives markets are characterized by complex structures (including options, swaps, snowball products, and shark fin products), cumbersome trading processes (including stages such as inquiry, quotation, locking in price, and confirmation), and significant differences in email formats and terminology among different financial institutions. Therefore, related technologies often suffer from inefficiencies, high risks of missed transactions, and frequent errors in content extraction when processing emails related to OTC derivatives. Thus, how to avoid these problems in processing emails related to OTC derivatives has become a pressing technical issue that needs to be addressed by those skilled in the art. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method, apparatus and system for processing emails related to over-the-counter derivatives, in order to solve the problems of low efficiency, high risk of missing transactions and frequent errors in content extraction when processing emails related to over-the-counter derivatives.

[0004] According to one aspect of this application, a method for processing emails related to over-the-counter derivatives is provided, comprising:

[0005] Receive emails about over-the-counter derivatives;

[0006] The emails related to over-the-counter derivatives are sent to the transaction judgment agent in the recognition layer, so that multiple first-level models in the transaction judgment agent process the emails related to over-the-counter derivatives respectively.

[0007] Initialize the first preset cumulative weight; after obtaining the corresponding first processing result for any of the first largest models, sum the model weights corresponding to the first largest model and the current first preset cumulative weights to update the current first preset cumulative weights, and determine whether the current first preset cumulative weights are not less than the first preset cumulative threshold.

[0008] When the current first preset cumulative weight is less than the first preset cumulative threshold, continue to wait for other first large models to obtain the corresponding first processing result, and re-execute the step of adding the model weight corresponding to the first large model and the current first preset cumulative weight to update the current first preset cumulative weight after any first large model obtains the corresponding first processing result, and determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0009] When the current first preset cumulative weight is not less than the first preset cumulative threshold, the output result of the transaction judgment agent is determined based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight.

[0010] When the output of the transaction judgment agent is a transaction email, it is processed using the email classification agent in the recognition layer, the business classification agent in the recognition layer, the extraction layer, and the supervision layer to obtain the recognition result of the email related to over-the-counter derivatives.

[0011] In some embodiments, the method further includes:

[0012] Determine the product type corresponding to the emails related to over-the-counter derivatives, whereby the product type includes standard products or complex products;

[0013] Based on the task of the transaction judgment agent, determine the model weight to be modified corresponding to the first large model;

[0014] Based on the product type, modify the model weights corresponding to the first large model to obtain the model weights corresponding to the first large model.

[0015] In some embodiments, the step of determining the output result of the transaction judgment agent based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight includes:

[0016] From the first processing results of the first largest model corresponding to the model weight used to calculate the first preset cumulative weight, select the first processing results with the largest number.

[0017] The first processing result with the largest number is determined as the output result of the transaction judgment agent.

[0018] In some embodiments, the step of processing emails related to off-exchange derivatives using the email classification agent in the identification layer, the business classification agent in the identification layer, the extraction layer, and the supervision layer includes:

[0019] Based on the emails related to over-the-counter derivatives, determine the output of the email classification agent;

[0020] Based on the emails related to over-the-counter derivatives, determine the output of the business classification agent;

[0021] When the output of the email classification agent is a confirmation category, based on the confirmation category and the output of the business classification agent, the confirmation element extraction agent in the extraction layer is used to process the emails related to off-exchange derivatives in order to determine the output of the confirmation element extraction agent.

[0022] Based on the output of the confirmation element extraction agent, the transaction task omission checking agent in the supervision layer is used to perform the check to obtain the first check result.

[0023] If the first check result is an omission, then send the first alarm signal;

[0024] If the first check result is not an omission, then based on the output result of the confirmed element extraction agent and the emails related to over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform a check to obtain a second check result.

[0025] If the second check result is that an element is missing, then a second alarm signal is sent;

[0026] If the second check result is not a missing element, then the element consistency check agent of the supervision layer is used to check and obtain the third check result.

[0027] If the results of the third check are inconsistent, a third alarm signal is sent.

[0028] If the third check result is consistent, then the output of the confirming element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

[0029] In some embodiments, the step of processing the emails related to over-the-counter derivatives using the confirmation element extraction agent in the extraction layer based on the confirmation category and the output of the business classification agent, to determine the output of the confirmation element extraction agent, includes:

[0030] The confirmation category, the output of the business classification agent, and the emails related to over-the-counter derivatives are sent to the confirmation element extraction agent in the extraction layer, so that multiple second-large models in the confirmation element extraction agent process the confirmation category, the output of the business classification agent, and the emails related to over-the-counter derivatives respectively.

[0031] Initialize the second preset cumulative weight; after obtaining the corresponding second processing result for any of the second major models, sum the model weights corresponding to the second major model and the current second preset cumulative weights to update the current second preset cumulative weights, and determine whether the current second preset cumulative weights are not less than the second preset cumulative threshold.

[0032] When the current second preset cumulative weight is less than the second preset cumulative threshold, continue to wait for other second major models to obtain the corresponding second processing result, and re-execute the step of adding the model weight corresponding to the second major model and the current second preset cumulative weight to update the current second preset cumulative weight after any second major model obtains the corresponding second processing result, and determine whether the current second preset cumulative weight is not less than the second preset cumulative threshold.

[0033] When the current second preset cumulative weight is not less than the second preset cumulative threshold, the output result of the confirmation element extraction agent is determined based on the second processing result of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight.

[0034] In some embodiments, the step of determining the output result of the confirming element extraction agent based on the second processing result of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight includes:

[0035] The second processing results of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight are merged to form a candidate element set;

[0036] Conflicting elements are selected from the set of candidate elements;

[0037] Count the number of conflicting elements in the candidate element set;

[0038] The conflicting elements with the largest number and the non-conflicting candidate elements in the set of candidate elements are used as the output of the confirmation element extraction agent.

[0039] In some embodiments, the method further includes:

[0040] When the output of the email classification agent is an inquiry category, based on the inquiry category and the output of the business classification agent, the inquiry element extraction agent in the extraction layer is used to process the emails related to over-the-counter derivatives to determine the output of the inquiry element extraction agent.

[0041] Based on the output of the inquiry element extraction agent and the relevant emails on over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform the check to obtain the fourth check result.

[0042] If the fourth check result is that an element is missing, a fourth alarm signal is sent.

[0043] When the fourth check result is not a missing element, the element consistency check agent in the supervision layer is used to check and obtain the fifth check result.

[0044] If the results of the fifth check are inconsistent, a fifth alarm signal is sent;

[0045] If the fifth check result is consistent, then the output of the inquiry element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

[0046] In some embodiments, before the step of using the feature integrity checking agent in the supervision layer to check the output of the query feature extraction agent to obtain a fourth check result, the method further includes:

[0047] Based on the incremental modification identification agent in the extraction layer, the output result of the price inquiry element extraction agent is modified.

[0048] According to another aspect of this application, an email processing apparatus for over-the-counter derivatives is provided, comprising:

[0049] The acquisition unit is used to retrieve emails related to over-the-counter derivatives.

[0050] The processing unit is used to send the emails related to over-the-counter derivatives to the transaction judgment agent in the identification layer, so that multiple first-level models in the transaction judgment agent can process the emails related to over-the-counter derivatives respectively.

[0051] An initialization unit is used to initialize the first preset cumulative weight;

[0052] The judgment unit is used to, after obtaining the corresponding first processing result of any of the first largest models, add the model weights corresponding to the first largest model and the current first preset cumulative weights to update the current first preset cumulative weights, and determine whether the current first preset cumulative weights are not less than the first preset cumulative threshold.

[0053] The re-execution unit is used to continue waiting for other first-largest models to obtain the corresponding first processing result when the current first-preset cumulative weight is less than the first-preset cumulative threshold, and to re-execute the steps of adding the model weight corresponding to the first-largest model and the current first-preset cumulative weight to update the current first-preset cumulative weight after any first-largest model obtains the corresponding first processing result, and determining whether the current first-preset cumulative weight is not less than the first-preset cumulative threshold.

[0054] The determining unit is used to determine the output result of the transaction judgment agent based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight when the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0055] The identification unit is used to process the email classification agent, the business classification agent, the extraction layer, and the supervision layer in the identification layer when the output result of the transaction judgment agent is a transaction email, so as to obtain the identification result of the email related to the over-the-counter derivatives.

[0056] According to another aspect of this application, a system for processing emails related to over-the-counter derivatives is provided, for:

[0057] Receive emails about over-the-counter derivatives;

[0058] The emails related to over-the-counter derivatives are sent to the transaction judgment agent in the recognition layer, so that multiple first-level models in the transaction judgment agent process the emails related to over-the-counter derivatives respectively.

[0059] Initialize the first preset cumulative weight; after obtaining the corresponding first processing result for any of the first largest models, sum the model weights corresponding to the first largest model and the current first preset cumulative weights to update the current first preset cumulative weights, and determine whether the current first preset cumulative weights are not less than the first preset cumulative threshold.

[0060] When the current first preset cumulative weight is less than the first preset cumulative threshold, continue to wait for other first large models to obtain the corresponding first processing result, and re-execute the step of adding the model weight corresponding to the first large model and the current first preset cumulative weight to update the current first preset cumulative weight after any first large model obtains the corresponding first processing result, and determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0061] When the current first preset cumulative weight is not less than the first preset cumulative threshold, the output result of the transaction judgment agent is determined based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight.

[0062] When the output of the transaction judgment agent is a transaction email, it is processed using the email classification agent in the recognition layer, the business classification agent in the recognition layer, the extraction layer, and the supervision layer to obtain the recognition result of the email related to over-the-counter derivatives.

[0063] By employing the above technical solutions, embodiments of this application provide a method, apparatus, and system for processing emails related to over-the-counter (OTC) derivatives. This method uses a trading judgment agent composed of multiple first-large models to perform precise intelligent analysis of emails related to OTC derivatives and dynamically adjusts the model weights corresponding to the first-large models. When a first preset cumulative weight is not less than a first preset cumulative threshold, the first processing result of the first-large model corresponding to the model weight used to calculate the current first preset cumulative weight is directly used to determine the output result of the trading judgment agent. This eliminates the need to wait for a response from the first-large model that has not yet received its first processing result. This ensures the accuracy of the trading judgment agent's output while significantly improving response speed, ultimately enhancing the accuracy and speed of email recognition results related to OTC derivatives.

[0064] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0065] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0066] Figure 1 An exemplary flowchart of an email processing method for over-the-counter derivatives is shown;

[0067] Figure 2 An exemplary schematic diagram of a device for processing emails related to over-the-counter derivatives is shown. Detailed Implementation

[0068] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0069] Over-the-counter (OTC) derivatives markets are characterized by complex structures (including options, swaps, snowball products, and shark fin products), cumbersome trading processes (including stages such as inquiry, quotation, price locking, and confirmation), and significant differences in email formats and terminology among different financial institutions. Therefore, related technologies often encounter problems such as low email processing efficiency, high risk of missed transactions, and frequent errors in content extraction when handling emails related to OTC derivatives.

[0070] Therefore, how to avoid inefficiency, high risk of missing transactions, and frequent errors in extracting content when processing emails related to over-the-counter derivatives has become a technical problem that urgently needs to be solved by those skilled in the art.

[0071] To address the aforementioned issues, this application provides a method, apparatus, and system for processing emails related to over-the-counter (OTC) derivatives. This method utilizes a trading intelligence agent composed of multiple primary models to perform precise intelligent analysis of emails related to OTC derivatives and dynamically adjusts the model weights corresponding to the primary models. When a first preset cumulative weight is not less than a first preset cumulative threshold, the output result of the trading intelligence agent is directly determined using the first processing result of the primary model corresponding to the model weight used to calculate the current first preset cumulative weight. This eliminates the need to wait for a response from the primary model that has not yet received its first processing result. This ensures the accuracy of the trading intelligence agent's output while significantly improving response speed, ultimately enhancing the accuracy and speed of email recognition related to OTC derivatives.

[0072] This application provides a method for processing emails related to over-the-counter derivatives in some embodiments, such as... Figure 1 As shown, the method includes S100-S700.

[0073] S100, Receive emails regarding over-the-counter derivatives.

[0074] Because over-the-counter derivatives are characterized by complex structures, cumbersome trading processes, and significant differences in email formats and terminology among different financial institutions, the method in this application embodiment is specifically designed for emails related to over-the-counter derivatives.

[0075] S200: Send the emails related to over-the-counter derivatives to the transaction judgment agent in the identification layer, so that multiple first-level models in the transaction judgment agent process the emails related to over-the-counter derivatives respectively.

[0076] This application uses a three-layer architecture, specifically including an identification layer, an extraction layer, and a supervision layer.

[0077] In this embodiment, the trading decision-making agent processes emails related to over-the-counter derivatives by calling multiple primary models in parallel. Since each primary model has different advantages, the accurate output of the trading decision-making agent can be obtained subsequently using some or all of the primary model's processing results.

[0078] In some embodiments, the trading decision agent comprises four primary models. Exemplarily, the four primary models may include the DeepSeek model, the QWQ model, the Qwen model, and the Kimi model. The four primary models process emails based on information about over-the-counter derivatives in parallel.

[0079] In this embodiment, the first major model uses a standardized model call template provided by the Prompt template library to ensure that multiple first major models receive consistent task descriptions. It is understood that processing emails related to over-the-counter derivatives constitutes the task received by the first major model.

[0080] S300. Initialize the first preset cumulative weight. For example, the initialized first preset cumulative weight, i.e., the current first preset cumulative weight, is 0.

[0081] S400. After obtaining the corresponding first processing result for any of the first largest models, the model weights corresponding to the first largest model and the current first preset cumulative weights are summed to update the current first preset cumulative weights, and it is determined whether the current first preset cumulative weights are not less than the first preset cumulative threshold.

[0082] It is understandable that multiple primary models may not complete the processing of emails related to over-the-counter derivatives at the same time. Once a primary model obtains its corresponding first processing result, the model weight corresponding to that primary model is summed with the current first preset cumulative weight.

[0083] For example, the transaction judgment agent includes four primary models: Primary Model A, Primary Model B, Primary Model C, and Primary Model D. Primary Model A is the first to be processed. At this time, the model weight of Primary Model A (e.g., 0.4) is summed with the current first preset cumulative weight (since Primary Model A is the first to be processed, the current first preset cumulative weight is the initialized first preset cumulative weight, i.e., 0), and the current first preset cumulative weight is updated (at this time, the current first preset cumulative weight is 0.4). Then, it is determined whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0084] S500. When the current first preset cumulative weight is less than the first preset cumulative threshold, continue to wait for other first large models to obtain the corresponding first processing result, and re-execute the step of adding the model weight corresponding to the first large model and the current first preset cumulative weight to update the current first preset cumulative weight after any first large model obtains the corresponding first processing result, and determining whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0085] S600. When the current first preset cumulative weight is not less than the first preset cumulative threshold, the output result of the transaction judgment agent is determined based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight.

[0086] Based on the previous example, the first preset cumulative threshold is 0.8. Since the current first preset cumulative weight, 0.4, is less than the first preset cumulative threshold, 0.8, we continue to wait for other first-largest models to process and obtain the corresponding first processing result. Then, we re-execute the step of summing the model weight corresponding to the first-largest model and the current first preset cumulative weight to update the current first preset cumulative weight after any first-largest model obtains the corresponding first processing result, and determining whether the current first preset cumulative weight is not less than the first preset cumulative threshold. For example, when re-executing the above steps, the first-largest model B processes emails related to over-the-counter derivatives and obtains the corresponding first processing result. We sum the model weight corresponding to the first-largest model B (e.g., 0.3) and the current first preset cumulative weight, 0.4, to update the current first preset cumulative weight (the current first preset cumulative weight is 0.7). We continue to determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0087] Since the first preset cumulative threshold is 0.8, the current first preset cumulative weight is still less than the first preset cumulative threshold. Therefore, we continue to wait for other first-largest models to process and obtain the corresponding first processing result. Then, we re-execute the step of summing the model weight corresponding to the first-largest model and the current first preset cumulative weight to update the current first preset cumulative weight after obtaining the corresponding first processing result for any first-largest model, and determining whether the current first preset cumulative weight is not less than the first preset cumulative threshold. For example, when re-executing the above steps, the model weight corresponding to the first-largest model C (e.g., 0.2) and the current first preset cumulative weight, i.e., 0.7, are summed to update the current first preset cumulative weight (the current first preset cumulative weight is 0.9). We continue to determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold. Since the current first preset cumulative weight, i.e. 0.9, is not less than the first preset cumulative threshold, i.e. 0.8, the output result of the transaction judgment agent is determined based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight. That is, the output result of the transaction judgment agent is determined based on the first processing result corresponding to the first large model A, the first large model B, and the first large model C, without waiting for the first large model D to obtain the corresponding first processing result.

[0088] In this embodiment, the sum of model weights of the first large model that yields the first processing result is calculated. When the first preset cumulative threshold is obtained, a decision is directly triggered. That is, the output result of the transaction judgment agent is determined directly based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight, without waiting for all first large models to respond (i.e., without waiting for all first large models to process and obtain the corresponding first processing result). In this way, the transaction judgment agent can integrate the first processing results of multiple first large models, ensuring the accuracy of the output result of the transaction judgment agent while significantly improving the response speed.

[0089] In this embodiment, the model weights corresponding to different first-largest models may be different, and are determined through dynamic adjustment. Specifically, in some embodiments, the method further includes: determining the product type corresponding to the email about over-the-counter derivatives, wherein the product type includes standard products or complex products.

[0090] Based on the task of the transaction judgment agent, determine the model weight to be modified corresponding to the first large model;

[0091] Based on the product type, modify the model weights corresponding to the first large model to obtain the model weights corresponding to the first large model.

[0092] In this embodiment, different primary models possess different strengths; some excel at understanding financial terminology, others at handling Chinese contexts, and still others at reasoning through long texts. Therefore, different weights can be assigned to the primary models for different tasks. Then, the weights of the primary models corresponding to the primary models are further modified based on the product type to adapt to that product.

[0093] For example, the task of the transaction judgment agent is transaction recognition. In this case, the model weights of multiple primary models are relatively balanced, and the model weights of the primary models are similar. In one example, the multiple primary models include the DeepSeek model, the Qwen model, the Kimi model, and the DeepSeek-R1 model, with corresponding model weights of 0.3, 0.3, 0.3, and 0.1, respectively. Based on this, the model weights of the primary models can be modified according to the product type to obtain the model weights corresponding to the primary models. For example, standard products prioritize processing speed, so the model weights of the slower primary models can be reduced, while the model weights of the faster primary models can be increased. In one example, the DeepSeek-R1 model is an inference model. When dealing with the same problem, the DeepSeek-R1 model takes more time to process data, so its model weight is reduced. Complex products require deep inference, so the weights of inference models can be increased; for example, the model weight of the DeepSeek-R1 model can be increased.

[0094] In this embodiment, the model weights are dynamically adjusted based on the task and product type corresponding to the intelligent agent.

[0095] In some embodiments, the step of determining the output result of the transaction judgment agent based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight includes:

[0096] From the first processing results of the first largest model corresponding to the model weight used to calculate the first preset cumulative weight, select the first processing results with the largest number.

[0097] The first processing result with the largest number is determined as the output result of the transaction judgment agent.

[0098] For example, the first processing result of the first largest model corresponding to the model weight used to calculate the first preset cumulative weight includes the first processing result of the first largest model A, the first processing result of the first largest model B, and the first processing result of the first largest model C. If the first processing result of the first largest model A is the same as the first processing result of the first largest model B, but different from the first processing result of the first largest model C, the first processing result with the largest number of selections is the first processing result of the first largest model A. The first processing result of the first largest model A is determined as the transaction judgment agent. This embodiment can also be referred to as using a voting strategy to determine the output result of the transaction judgment agent.

[0099] If the output of the transaction judgment agent is not a transaction email, then the subsequent steps will not be executed.

[0100] The process of steps S100 to S600 above can be summarized as the process of determining the output result of the trading agent based on emails related to over-the-counter derivatives.

[0101] S700. When the output of the transaction judgment agent is a transaction email, the email classification agent in the recognition layer, the business classification agent in the recognition layer, the extraction layer, and the supervision layer are used for processing to obtain the recognition result of the email related to over-the-counter derivatives.

[0102] In this embodiment, the complete processing of emails related to over-the-counter derivatives includes an identification stage, an extraction stage, and a verification stage. The identification stage is completed by the identification layer, the extraction stage by the extraction layer, and the verification stage by the verification layer. Therefore, when the output of the transaction judgment agent is determined to be a transaction email, the email classification agent and business classification agent in the identification layer, the agent in the extraction layer, and the agent in the verification layer need to continue processing to finally obtain the identification result of the emails related to over-the-counter derivatives.

[0103] The architecture in this embodiment includes safeguards such as JSON extraction and error recovery. The architecture also includes a Prompt Lab, which continuously optimizes system performance through template configuration, testing and verification, and result evaluation. This architectural design effectively separates business logic, technical capabilities, and quality assurance, ensuring clear responsibilities and loose coupling between modules, thus facilitating system expansion and maintenance.

[0104] In some embodiments, the step of processing emails related to off-exchange derivatives using the email classification agent in the identification layer, the business classification agent in the identification layer, the extraction layer, and the supervision layer includes:

[0105] Based on the emails related to over-the-counter derivatives, determine the output of the email classification agent;

[0106] In this embodiment, the output of the mail classification agent includes inquiry category, price lock category, confirmation category, or redemption category. It should be noted that inquiry, price lock, confirmation, and redemption are the transaction stages of over-the-counter derivatives trading.

[0107] In this embodiment, the step of determining the output of the email classification agent based on emails related to over-the-counter derivatives is similar to the process described above for determining the output of the trading agent based on emails related to over-the-counter derivatives, and will not be repeated here. In this embodiment, the email classification agent also includes multiple large models. However, unlike the process of determining the output of the trading agent based on emails related to over-the-counter derivatives, the large models used may be different, the model weights set for the large models may be different (the method for setting model weights is the same as the method for setting the model weights of the first large model described above), and the input data to the large models are different.

[0108] Based on the emails related to over-the-counter derivatives, determine the output of the business classification agent.

[0109] In this embodiment, the output of the business classification agent, i.e., the business classification, includes options products, swap transactions, snowball products, or shark fin products. It should be noted that options products, swap transactions, snowball products, and shark fin products all belong to over-the-counter derivatives.

[0110] In this embodiment, the step of determining the output of the business classification agent based on emails related to over-the-counter derivatives is similar to the process described above for determining the output of the trading agent based on emails related to over-the-counter derivatives, and will not be repeated here. In this embodiment, the email classification agent also includes multiple large models. However, unlike the process of determining the output of the trading agent based on emails related to over-the-counter derivatives, the large models used may be different, the model weights set for the large models may be different (the method for setting model weights is the same as the method for setting the model weights of the first large model described above), and the input data to the large models are different.

[0111] When the output of the email classification agent is a confirmation category, based on the confirmation category and the output of the business classification agent, the confirmation element extraction agent in the extraction layer is used to process the emails related to off-exchange derivatives in order to determine the output of the confirmation element extraction agent.

[0112] In this embodiment, the outputs of the email classification agent and the business classification agent assist the confirmation element extraction agent in processing emails related to over-the-counter derivatives, enabling the inquiry element agent to extract elements more accurately based on the confirmation category and business classification, thus obtaining accurate output results from the confirmation element extraction agent. In this embodiment, the elements can be extraction amount and fee rate, etc.

[0113] In some embodiments, the step of processing the emails related to over-the-counter derivatives using the confirmation element extraction agent in the extraction layer based on the confirmation category and the output of the business classification agent, to determine the output of the confirmation element extraction agent, includes:

[0114] The confirmation category, the output of the business classification agent, and the emails related to over-the-counter derivatives are sent to the confirmation element extraction agent in the extraction layer, so that multiple second-large models in the confirmation element extraction agent process the confirmation category, the output of the business classification agent, and the emails related to over-the-counter derivatives respectively.

[0115] Initialize the second preset cumulative weight;

[0116] After obtaining the corresponding second processing result for any of the second major models, the model weights corresponding to the second major models and the current second preset cumulative weights are summed to update the current second preset cumulative weights, and it is determined whether the current second preset cumulative weights are not less than the second preset cumulative threshold.

[0117] In this embodiment, the method for setting the weights of the second model corresponding to the second largest model is the same as the method for setting the weights of the first model corresponding to the first largest model described above, and will not be repeated here.

[0118] When the current second preset cumulative weight is less than the second preset cumulative threshold, continue to wait for other second major models to obtain the corresponding second processing result, and re-execute the step of adding the model weight corresponding to the second major model and the current second preset cumulative weight to update the current second preset cumulative weight after any second major model obtains the corresponding second processing result, and determine whether the current second preset cumulative weight is not less than the second preset cumulative threshold.

[0119] When the current second preset cumulative weight is not less than the second preset cumulative threshold, the output result of the confirmation element extraction agent is determined based on the second processing result of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight.

[0120] In this embodiment, from initializing the second preset cumulative weight to waiting for other second-largest models to obtain their corresponding second processing results when the current second preset cumulative weight is less than the second preset cumulative threshold, and then re-executing the step of summing the model weights of the second-largest model and the current second preset cumulative weight to update the current second preset cumulative weight after any second-largest model obtains its corresponding second processing result, and determining whether the current second preset cumulative weight is not less than the second preset cumulative threshold, is the same process as the step mentioned above of initializing the first preset cumulative weight to waiting for other first-largest models to obtain their corresponding first processing results when the current first preset cumulative weight is less than the first preset cumulative threshold, and then re-executing the step of summing the model weights of the first-largest model and the current first preset cumulative weight to update the current first preset cumulative weight after any first-largest model obtains its corresponding first processing result, and determining whether the current first preset cumulative weight is not less than the first preset cumulative threshold. Therefore, it will not be described in detail here. However, the process of determining the output result of the confirmation element extraction agent based on the second processing result of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight and the process of determining the output result of the transaction judgment agent based on the first processing result of the first largest model corresponding to the model weight used to calculate the current first preset cumulative weight, as described above, can be different.

[0121] Specifically, in some embodiments, the step of determining the output result of the confirming element extraction agent based on the second processing result of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight can be referred to as an integration strategy, including:

[0122] The second processing results of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight are merged to form a candidate element set.

[0123] For example, multiple second-largest models include the Qwen model, the Deepseek model, and the Kimi model. Each second-largest model outputs a corresponding second-processing result. For instance, the second-processing result output by the Qwen model includes: value date: 2025-08-20; principal: 3000 yuan; underlying asset: null; option type: null. The second-processing result output by the Deepseek model includes: value date: null; principal: 3000 yuan; underlying asset: Kweichow Moutai; option type: binary call. The second-processing result output by the Kimi model includes: value date: 2025-08-20; principal: 3000 yuan; underlying asset: Kweichow Moutai; option type: null. The second-processing results output by the Qwen model, the Deepseek model, and the Kimi model are merged to obtain the candidate element set. The candidate set of elements includes: Value date: 2025-08-20; Principal: 3000 yuan; Underlying asset: null; Option type: null; Value date: null; Principal: 3000 yuan; Underlying asset: Kweichow Moutai; Option type: Call binary; Value date: 2025-08-20; Principal: 3000 yuan; Underlying asset: Kweichow Moutai; Option type: null.

[0124] Conflicting elements are selected from the set of candidate elements.

[0125] In this embodiment, conflicting elements refer to results of the same type that have different specific contents. For example, for the same interest date type, the Qwen model outputs 2025-08-20; the Deepseek model outputs null; and the Kimi model outputs 2025-08-20. In this case, 2025-08-20 and null are conflicting elements.

[0126] The number of conflicting elements in the candidate element set is counted.

[0127] Following the previous example, there are 2 instances of 2025-08-20 and 1 instance of null.

[0128] The conflicting elements with the largest number and the non-conflicting candidate elements in the set of candidate elements are used as the output of the confirmation element extraction agent.

[0129] In this embodiment of the application, a candidate element without conflict refers to a specific result of the same type. For example, if the interest start date output by the three second largest models is 2025-08-20, then 2025-08-20 does not conflict and is therefore a candidate element without conflict.

[0130] It is understandable that the principal, underlying asset, and option type are determined in the same way as the value date to determine whether they are candidate or conflicting elements, and will not be repeated here. In the embodiments of this application, the candidate element set is screened and merged to finally obtain the output result of the comprehensive confirmation element extraction agent.

[0131] Based on the output of the confirmation element extraction agent, the transaction task omission checking agent in the supervision layer is used to perform the check to obtain the first check result.

[0132] In this embodiment of the application, the transaction omission check agent is used to check whether the trader forgot to record the transaction after confirming the transaction and sending a confirmation email to the customer.

[0133] In this embodiment, based on the output of the confirmation element extraction agent, the transaction task omission checking agent in the supervision layer is used to perform checks to obtain a first check result. This is similar to the process described above for determining the output of the trading agent based on emails related to OTC derivatives, and will not be repeated here. The transaction task omission checking agent also includes multiple large models. Unlike the process of determining the output of the trading agent based on emails related to OTC derivatives, the large models used may be different, the model weights set for the large models may be different (the method for setting model weights is the same as the method for setting the model weights of the first large model described above), and the input data to the large models are different.

[0134] If the first check result is an omission, then send the first alarm signal;

[0135] In this embodiment of the application, if the first check result is an omission, a first alarm signal is sent so that the trader can handle the omission in a timely manner.

[0136] If the first check result is not an omission, then based on the output result of the confirmed element extraction agent and the emails related to over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform a check to obtain a second check result.

[0137] For example, if the feature integrity check agent determines that an email about over-the-counter derivatives should contain 10 features, but the feature extraction agent's output only includes 8 features, then the second check result is determined to be feature missing. If the feature extraction agent's output includes 10 features, then the second check result is determined not to be feature missing.

[0138] In this embodiment, based on the output of the confirmation element extraction agent and the emails related to OTC derivatives, the element integrity checking agent in the supervision layer is used to perform a check to obtain a second check result. This is similar to the process described above of determining the output of the trading agent based on emails related to OTC derivatives, and will not be repeated here. The element integrity checking agent also includes multiple large models. Unlike the process of determining the output of the trading agent based on emails related to OTC derivatives, the large models used may be different, the model weights set for the large models may be different (the method for setting model weights is the same as the method for setting model weights for the first large model described above), and the input data for the large models are different.

[0139] If the second check result indicates that an element is missing, a second alarm signal is sent.

[0140] In this embodiment of the application, when the first check result is missing, a second alarm signal is sent to notify the trader to check the emails related to over-the-counter derivatives.

[0141] If the second check result is not a missing element, then the element consistency check agent of the supervision layer is used to perform the check to obtain the third check result.

[0142] In this embodiment, when the feature extraction agent outputs the corresponding output result, it needs to be stored. After storage, a consistency check needs to be performed between the stored result and the elements in the output result of the feature extraction agent. That is, the feature consistency check agent is used to detect whether the output result of the feature extraction agent and the stored elements are consistent.

[0143] In this embodiment, the feature consistency checking agent of the supervision layer is used to perform checks to obtain a third check result. This is similar to the process described above of determining the output result of the trading agent based on emails related to OTC derivatives, and will not be repeated here. In this embodiment, the feature consistency checking agent also includes multiple large models. Unlike the process of determining the output result of the trading agent based on emails related to OTC derivatives, the large models used may be different, the model weights set for the large models may be different (the method of setting model weights is the same as the method of setting model weights for the first large model described above), and the input data to the large models are different.

[0144] If the results of the third check are inconsistent, a third alarm signal is sent.

[0145] In this embodiment, if the stored elements and the output directly output by the confirmation element extraction agent are the same, the third check result is determined to be consistent. If they are different, the third check result is determined to be inconsistent, and a third alarm signal is sent to allow traders to promptly check emails related to over-the-counter derivatives.

[0146] If the third check result is consistent, then the output of the confirming element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

[0147] In some embodiments, the method further includes:

[0148] When the output of the email classification agent is an inquiry category, based on the inquiry category and the output of the business classification agent, the inquiry element extraction agent in the extraction layer is used to process the emails related to over-the-counter derivatives to determine the output of the inquiry element extraction agent.

[0149] In this embodiment, based on the output results of the inquiry category and the business classification agent, the inquiry element extraction agent in the extraction layer processes the emails related to over-the-counter derivatives to determine the output result of the inquiry element extraction agent. This process is the same as the process described above, where the confirmation element extraction agent in the extraction layer processes the emails related to over-the-counter derivatives based on the output results of the confirmation category and the business classification agent to determine the output result of the confirmation element extraction agent, and will not be repeated here. In this embodiment, the inquiry element extraction agent also includes multiple large models. The difference between processing the emails related to over-the-counter derivatives using the confirmation element extraction agent in the extraction layer based on the output results of the confirmation category and the business classification agent to determine the output result of the confirmation element extraction agent is that the large models used may be different, the model weights corresponding to the large models may be different, and the input data to the large models may be different.

[0150] In this embodiment, the output results of the email classification agent and the business classification agent assist the inquiry element extraction agent of the extraction layer in processing emails related to over-the-counter derivatives, so that the inquiry element agent can extract elements more accurately for inquiry categories and business classifications, that is, obtain accurate output results of the inquiry element extraction agent.

[0151] Based on the output of the inquiry element extraction agent and the relevant emails on over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform the check to obtain the fourth check result.

[0152] In this embodiment, the element integrity checking agent has been described above and will not be repeated here.

[0153] In some embodiments, before the step of using the feature integrity checking agent in the supervision layer to check the output of the query feature extraction agent to obtain a fourth check result, the method further includes:

[0154] Based on the incremental modification identification agent in the extraction layer, the output result of the price inquiry element extraction agent is modified.

[0155] In this embodiment, the incremental modification identification agent is specifically designed to identify incremental modifications to emails related to over-the-counter derivatives. For example, the incremental modification could be "the rate has been changed to 8.5%" in a quote inquiry email.

[0156] In this embodiment, the incremental modification identification agent also includes multiple large models. Unlike the process of determining the output of the trading agent based on emails related to over-the-counter derivatives, the large models used may be different, the model weights set for the large models may be different (the method for setting model weights is the same as the method for setting the model weights of the first large model described above), and the input data to the large models are different.

[0157] If the fourth check result is that an element is missing, a fourth alarm signal is sent.

[0158] When the fourth check result is not a missing element, the element consistency check agent in the supervision layer is used to check and obtain the fifth check result.

[0159] In this embodiment of the application, the element consistency checking agent can also be used to detect whether the output of the query element extraction agent is consistent with the stored elements.

[0160] In this embodiment, the element consistency check agent has been described above and will not be repeated here.

[0161] If the results of the fifth check are inconsistent, a fifth alarm signal is sent;

[0162] If the fifth check result is consistent, then the output of the inquiry element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

[0163] In some embodiments, when the output of the email classification agent is a price-locked category, the price-locked element extraction agent in the extraction layer is used to process the emails related to over-the-counter derivatives based on the price-locked category and the output of the business classification agent, so as to determine the output of the price-locked element extraction agent.

[0164] In this embodiment, based on the output results of the locking-in category and the business classification agent, the locking-in element extraction agent in the extraction layer processes the emails related to OTC derivatives to determine the output result of the locking-in element extraction agent. This process is the same as the process described above, where the locking-in element extraction agent in the extraction layer processes the emails related to OTC derivatives based on the output results of the confirmation category and the business classification agent to determine the output result of the confirmation element extraction agent, and will not be repeated here. In this embodiment, the locking-in element extraction agent also includes multiple large models. The difference between this and processing the emails related to OTC derivatives using the confirmation element extraction agent in the extraction layer based on the output results of the confirmation category and the business classification agent to determine the output result of the confirmation element extraction agent is that the large models used may be different, the model weights corresponding to the large models may be different, and the input data to the large models may be different.

[0165] Based on the output of the locked-price element extraction agent and the relevant emails on over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform the check to obtain the sixth check result.

[0166] In this embodiment, the element integrity checking agent has been described above and will not be repeated here.

[0167] In some embodiments, before the step of using the feature integrity checking agent in the supervision layer to perform a check based on the output of the locked-price feature extraction agent and the emails related to OTC derivatives to obtain a sixth check result, the method further includes:

[0168] Based on the incremental modification identification agent in the extraction layer, the output result of the price-locking element extraction agent is modified.

[0169] In this embodiment, the incremental modification recognition agent has been described above and will not be repeated here.

[0170] If the sixth check result is that an element is missing, then a sixth alarm signal is sent;

[0171] When the sixth check result is not a missing element, the element consistency check agent in the supervision layer is used to perform the check to obtain the seventh check result.

[0172] In this embodiment, the element consistency check agent has been described above and will not be repeated here.

[0173] If the results of the seventh check are inconsistent, a seventh alarm signal is sent;

[0174] If the seventh check result is consistent, then the output of the price-locking element extraction agent is determined to be the identification result of the email related to OTC derivatives.

[0175] In some embodiments, the method further includes:

[0176] When the output of the email classification agent is a quotation category, based on the quotation category and the output of the business classification agent, the quotation element extraction agent in the extraction layer is used to process the emails related to over-the-counter derivatives to determine the output of the quotation element extraction agent.

[0177] In this embodiment, based on the output results of the pricing category and the business classification agent, the pricing element extraction agent in the extraction layer processes the emails related to OTC derivatives to determine the output result of the pricing element extraction agent. This process is the same as the process described above, where the confirmation element extraction agent in the extraction layer processes the emails related to OTC derivatives based on the output results of the confirmation category and the business classification agent to determine the output result of the confirmation element extraction agent, and will not be repeated here. In this embodiment, the pricing element extraction agent also includes multiple large models. The difference between this and processing the emails related to OTC derivatives using the confirmation element extraction agent in the extraction layer based on the output results of the confirmation category and the business classification agent to determine the output result of the confirmation element extraction agent is that the large models used may be different, the model weights corresponding to the large models may be different, and the input data to the large models may be different.

[0178] Based on the output of the pricing element extraction agent and the emails related to over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform the check to obtain the eighth check result.

[0179] In this embodiment, the element integrity checking agent has been described above and will not be repeated here.

[0180] If the eighth check result is that an element is missing, then an eighth alarm signal is sent.

[0181] When the eighth check result is not a missing element, the element consistency check agent in the supervision layer is used to check and obtain the ninth check result.

[0182] In this embodiment, the element consistency check agent has been described above and will not be repeated here.

[0183] If the results of the ninth check are inconsistent, a ninth alarm signal is sent.

[0184] If the results of the ninth check are consistent, then the output of the pricing element extraction agent is determined to be the identification result of the email related to OTC derivatives.

[0185] The method in this application embodiment enables each agent to call multiple large models (e.g., DeepSeek, Qwen, Kimi, etc.) in parallel to perform accurate intelligent analysis of emails related to over-the-counter derivatives. A hierarchical agent architecture (i.e., a three-layer architecture) is adopted to realize pipelined processing of identification, extraction, and verification, and the collaborative efficiency of the models is optimized through dynamic model weight adjustment and a preset cumulative threshold response mechanism, that is, there is no need to wait for all large models to output the corresponding processing results.

[0186] By leveraging the differentiated strengths of different large models—for example, DeepSeek excels at understanding financial terminology, Qwen at processing Chinese context, and Kimi at reasoning about long texts—weights are assigned adaptively to tasks, achieving complementary advantages in model capabilities.

[0187] By employing a multi-strategy fusion mechanism (voting and integration strategies), the illusions and biases of a single model are effectively eliminated. This allows for accurate identification of emails at different transaction stages, such as inquiry, price locking, and confirmation. Key elements such as customer name, transaction amount, product type, and fee period are precisely extracted, and data quality is ensured through multi-layered verification mechanisms. The resulting accurate email identification results related to OTC derivatives significantly improve the efficiency of OTC derivatives email processing, reduce operational risks, and achieve intelligent management of the entire transaction process.

[0188] By applying the technical solution of the above embodiments, the method can perform accurate and intelligent analysis of emails related to over-the-counter derivatives based on a trading judgment agent composed of multiple first-large models, and dynamically adjust the model weights corresponding to the first-large models. When the first preset cumulative weight is not less than the first preset cumulative threshold, the output result of the trading judgment agent is determined directly using the first processing result of the first-large model corresponding to the model weight used to calculate the current first preset cumulative weight, without waiting for the response of the first-large model that has not yet received the first processing result. This ensures the accuracy of the output result of the trading judgment agent while significantly improving the response speed, ultimately improving the accuracy and speed of the identification results of emails related to over-the-counter derivatives.

[0189] In some embodiments, as a specific implementation of the email processing method for over-the-counter derivatives described in the above embodiments, some embodiments of this application also provide an email processing apparatus for over-the-counter derivatives, such as... Figure 2 As shown, it includes:

[0190] Acquisition unit 201 is used to retrieve emails related to over-the-counter derivatives;

[0191] Processing unit 202 is used to send the emails related to over-the-counter derivatives to the transaction judgment agent in the identification layer, so that multiple first-level models in the transaction judgment agent process the emails related to over-the-counter derivatives respectively.

[0192] Initialization unit 203 is used to initialize the first preset cumulative weight;

[0193] The judgment unit 204 is used to, after obtaining the corresponding first processing result of any of the first large models, add the model weight corresponding to the first large model and the current first preset cumulative weight to update the current first preset cumulative weight, and determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0194] The re-execution unit 205 is used to continue waiting for other first-largest models to obtain the corresponding first processing result when the current first preset cumulative weight is less than the first preset cumulative threshold, and to re-execute the steps of adding the model weight corresponding to the first-largest model and the current first preset cumulative weight to update the current first preset cumulative weight after any first-largest model obtains the corresponding first processing result, and determining whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0195] The determining unit 206 is used to determine the output result of the transaction judgment agent based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight when the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0196] The identification unit 207 is used to process the email classification agent, the business classification agent, the extraction layer and the supervision layer in the identification layer when the output result of the transaction judgment agent is a transaction email, so as to obtain the identification result of the email related to the over-the-counter derivatives.

[0197] In some embodiments, as a specific implementation of the email processing method related to over-the-counter derivatives described in the above embodiments, some embodiments of this application also provide an email processing system related to over-the-counter derivatives, for:

[0198] Receive emails about over-the-counter derivatives;

[0199] The emails related to over-the-counter derivatives are sent to the transaction judgment agent in the recognition layer, so that multiple first-level models in the transaction judgment agent process the emails related to over-the-counter derivatives respectively.

[0200] Initialize the first preset cumulative weight; after obtaining the corresponding first processing result for any of the first largest models, sum the model weights corresponding to the first largest model and the current first preset cumulative weights to update the current first preset cumulative weights, and determine whether the current first preset cumulative weights are not less than the first preset cumulative threshold.

[0201] When the current first preset cumulative weight is less than the first preset cumulative threshold, continue to wait for other first large models to obtain the corresponding first processing result, and re-execute the step of adding the model weight corresponding to the first large model and the current first preset cumulative weight to update the current first preset cumulative weight after any first large model obtains the corresponding first processing result, and determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold.

[0202] When the current first preset cumulative weight is not less than the first preset cumulative threshold, the output result of the transaction judgment agent is determined based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight.

[0203] When the output of the transaction judgment agent is a transaction email, it is processed using the email classification agent in the recognition layer, the business classification agent in the recognition layer, the extraction layer, and the supervision layer to obtain the recognition result of the email related to over-the-counter derivatives.

[0204] It should be noted that other corresponding descriptions of the functional units involved in the email processing system related to over-the-counter derivatives provided in the embodiments of this application can be found in the corresponding descriptions in the email processing method related to over-the-counter derivatives provided in the above embodiments, and will not be repeated here.

[0205] This application also provides a computer device, specifically a personal computer, server, network device, etc. The computer device includes a bus, processor, memory, and communication interface, and may also include input / output interfaces and a display device. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores location information. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0206] Those skilled in the art will understand that the structure of the computer device described above is only a partial structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components, or combine certain components, or have different component arrangements.

[0207] In one embodiment, a computer-readable storage medium is also provided, which may be non-volatile or volatile, and a computer program is stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0208] In one embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0209] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0210] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. 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 methods.

[0211] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0212] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for processing emails related to over-the-counter derivatives, characterized in that, include: Receive emails about over-the-counter derivatives; The emails related to over-the-counter derivatives are sent to the transaction judgment agent in the recognition layer, so that multiple first-level models in the transaction judgment agent process the emails related to over-the-counter derivatives respectively. Initialize the first preset cumulative weight; After obtaining the corresponding first processing result for any of the first largest models, the model weights corresponding to the first largest model and the current first preset cumulative weights are summed to update the current first preset cumulative weights, and it is determined whether the current first preset cumulative weights are not less than the first preset cumulative threshold. When the current first preset cumulative weight is less than the first preset cumulative threshold, continue to wait for other first large models to obtain the corresponding first processing result, and re-execute the step of adding the model weight corresponding to the first large model and the current first preset cumulative weight to update the current first preset cumulative weight after any first large model obtains the corresponding first processing result, and determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold. When the current first preset cumulative weight is not less than the first preset cumulative threshold, the output result of the transaction judgment agent is determined based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight. When the output of the transaction judgment agent is a transaction email, it is processed using the email classification agent in the recognition layer, the business classification agent in the recognition layer, the extraction layer, and the supervision layer to obtain the recognition result of the email related to OTC derivatives. The step of using the email classification agent, the business classification agent, the extraction layer, and the supervision layer in the identification layer to process the data and obtain the identification results of emails related to off-exchange derivatives includes: Based on the emails related to over-the-counter derivatives, determine the output of the email classification agent; Based on the emails related to over-the-counter derivatives, determine the output of the business classification agent; When the output of the email classification agent is a confirmation category, based on the confirmation category and the output of the business classification agent, the confirmation element extraction agent in the extraction layer is used to process the emails related to off-exchange derivatives to determine the output of the confirmation element extraction agent. Based on the output of the confirmation element extraction agent, the transaction task omission checking agent in the supervision layer is used to perform the check to obtain the first check result. If the first check result is an omission, then send the first alarm signal; If the first check result is not an omission, then based on the output result of the confirmed element extraction agent and the emails related to over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform a check to obtain a second check result. If the second check result is that an element is missing, a second alarm signal is sent; If the second check result is not a missing element, then the element consistency check agent of the supervision layer is used to check and obtain the third check result. If the third check result is inconsistent, a third alarm signal is sent; If the third check result is consistent, then the output of the confirming element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

2. The method according to claim 1, characterized in that, Also includes: Determine the product type corresponding to the emails related to over-the-counter derivatives, whereby the product type includes standard products or complex products; Based on the task of the transaction judgment agent, determine the model weight to be modified corresponding to the first large model; Based on the product type, modify the model weights corresponding to the first large model to obtain the model weights corresponding to the first large model.

3. The method according to claim 1, characterized in that, The step of determining the output result of the transaction judgment agent based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight includes: From the first processing results of the first largest model corresponding to the model weight used to calculate the first preset cumulative weight, select the first processing results with the largest number. The first processing result with the largest number is determined as the output result of the transaction judgment agent.

4. The method according to claim 1, characterized in that, The step of processing the emails related to over-the-counter derivatives using the confirmation element extraction agent in the extraction layer based on the output results of the confirmation category and the business classification agent, in order to determine the output results of the confirmation element extraction agent, includes: The confirmation category, the output of the business classification agent, and the emails related to over-the-counter derivatives are sent to the confirmation element extraction agent in the extraction layer, so that multiple second-large models in the confirmation element extraction agent process the confirmation category, the output of the business classification agent, and the emails related to over-the-counter derivatives respectively. Initialize the second preset cumulative weight; after obtaining the corresponding second processing result for any of the second major models, sum the model weights corresponding to the second major model and the current second preset cumulative weights to update the current second preset cumulative weights, and determine whether the current second preset cumulative weights are not less than the second preset cumulative threshold. When the current second preset cumulative weight is less than the second preset cumulative threshold, continue to wait for other second major models to obtain the corresponding second processing result, and re-execute the step of adding the model weight corresponding to the second major model and the current second preset cumulative weight to update the current second preset cumulative weight after any second major model obtains the corresponding second processing result, and determine whether the current second preset cumulative weight is not less than the second preset cumulative threshold. When the current second preset cumulative weight is not less than the second preset cumulative threshold, the output result of the confirmation element extraction agent is determined based on the second processing result of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight.

5. The method according to claim 4, characterized in that, The step of determining the output result of the confirmed element extraction agent based on the second processing result of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight includes: The second processing results of the second largest model corresponding to the model weight used to calculate the current second preset cumulative weight are merged to form a candidate element set; Conflicting elements are selected from the set of candidate elements; Count the number of conflicting elements in the candidate element set; The conflicting elements with the largest number and the non-conflicting candidate elements in the set of candidate elements are used as the output of the confirmation element extraction agent.

6. The method according to claim 1, characterized in that, Also includes: When the output of the email classification agent is an inquiry category, based on the inquiry category and the output of the business classification agent, the inquiry element extraction agent in the extraction layer is used to process the emails related to over-the-counter derivatives to determine the output of the inquiry element extraction agent. Based on the output of the inquiry element extraction agent and the relevant emails on over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform the check to obtain the fourth check result. If the fourth check result is that an element is missing, a fourth alarm signal is sent. When the fourth check result is not a missing element, the element consistency check agent in the supervision layer is used to check and obtain the fifth check result. If the results of the fifth check are inconsistent, a fifth alarm signal is sent; If the fifth check result is consistent, then the output of the inquiry element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

7. The method according to claim 6, characterized in that, Before the step of using the feature integrity checking agent in the supervision layer to check the output result of the query feature extraction agent to obtain the fourth check result, the method further includes: Based on the incremental modification identification agent in the extraction layer, the output result of the price inquiry element extraction agent is modified.

8. An email processing apparatus for over-the-counter derivatives, characterized in that, include: The acquisition unit is used to retrieve emails related to over-the-counter derivatives. The processing unit is used to send the emails related to over-the-counter derivatives to the transaction judgment agent in the identification layer, so that multiple first-level models in the transaction judgment agent can process the emails related to over-the-counter derivatives respectively. An initialization unit is used to initialize the first preset cumulative weight; The judgment unit is used to, after obtaining the corresponding first processing result of any of the first largest models, add the model weights corresponding to the first largest model and the current first preset cumulative weights to update the current first preset cumulative weights, and determine whether the current first preset cumulative weights are not less than the first preset cumulative threshold. The re-execution unit is used to continue waiting for other first-largest models to obtain the corresponding first processing result when the current first-preset cumulative weight is less than the first-preset cumulative threshold, and to re-execute the steps of adding the model weight corresponding to the first-largest model and the current first-preset cumulative weight to update the current first-preset cumulative weight after any first-largest model obtains the corresponding first processing result, and determining whether the current first-preset cumulative weight is not less than the first-preset cumulative threshold. The determining unit is used to determine the output result of the transaction judgment agent based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight when the current first preset cumulative weight is not less than the first preset cumulative threshold. The identification unit is used to process the email classification agent, the business classification agent, the extraction layer, and the supervision layer in the identification layer when the output result of the transaction judgment agent is a transaction email, so as to obtain the identification result of the email related to the over-the-counter derivatives. The identification unit is specifically used for: Based on the emails related to over-the-counter derivatives, determine the output of the email classification agent; Based on the emails related to over-the-counter derivatives, determine the output of the business classification agent; When the output of the email classification agent is a confirmation category, based on the confirmation category and the output of the business classification agent, the confirmation element extraction agent in the extraction layer is used to process the emails related to off-exchange derivatives to determine the output of the confirmation element extraction agent. Based on the output of the confirmation element extraction agent, the transaction task omission checking agent in the supervision layer is used to perform the check to obtain the first check result. If the first check result is an omission, then send the first alarm signal; If the first check result is not an omission, then based on the output result of the confirmed element extraction agent and the emails related to over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform a check to obtain a second check result. If the second check result is that an element is missing, a second alarm signal is sent; If the second check result is not a missing element, then the element consistency check agent of the supervision layer is used to check and obtain the third check result. If the third check result is inconsistent, a third alarm signal is sent; If the third check result is consistent, then the output of the confirming element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

9. A system for processing emails related to over-the-counter derivatives, characterized in that, Used for: Receive emails about over-the-counter derivatives; The emails related to over-the-counter derivatives are sent to the transaction judgment agent in the recognition layer, so that multiple first-level models in the transaction judgment agent process the emails related to over-the-counter derivatives respectively. Initialize the first preset cumulative weight; After obtaining the corresponding first processing result for any of the first largest models, the model weights corresponding to the first largest model and the current first preset cumulative weights are summed to update the current first preset cumulative weights, and it is determined whether the current first preset cumulative weights are not less than the first preset cumulative threshold. When the current first preset cumulative weight is less than the first preset cumulative threshold, continue to wait for other first large models to obtain the corresponding first processing result, and re-execute the step of adding the model weight corresponding to the first large model and the current first preset cumulative weight to update the current first preset cumulative weight after any first large model obtains the corresponding first processing result, and determine whether the current first preset cumulative weight is not less than the first preset cumulative threshold. When the current first preset cumulative weight is not less than the first preset cumulative threshold, the output result of the transaction judgment agent is determined based on the first processing result of the first large model corresponding to the model weight used to calculate the current first preset cumulative weight. When the output of the transaction judgment agent is a transaction email, it is processed using the email classification agent in the recognition layer, the business classification agent in the recognition layer, the extraction layer, and the supervision layer to obtain the recognition result of the email related to OTC derivatives. The step of using the email classification agent, the business classification agent, the extraction layer, and the supervision layer in the identification layer to process the data and obtain the identification results of emails related to off-exchange derivatives includes: Based on the emails related to over-the-counter derivatives, determine the output of the email classification agent; Based on the emails related to over-the-counter derivatives, determine the output of the business classification agent; When the output of the email classification agent is a confirmation category, based on the confirmation category and the output of the business classification agent, the confirmation element extraction agent in the extraction layer is used to process the emails related to off-exchange derivatives to determine the output of the confirmation element extraction agent. Based on the output of the confirmation element extraction agent, the transaction task omission checking agent in the supervision layer is used to perform the check to obtain the first check result. If the first check result is an omission, then send the first alarm signal; If the first check result is not an omission, then based on the output result of the confirmed element extraction agent and the emails related to over-the-counter derivatives, the element integrity checking agent in the supervision layer is used to perform a check to obtain a second check result. If the second check result is that an element is missing, a second alarm signal is sent; If the second check result is not a missing element, then the element consistency check agent of the supervision layer is used to check and obtain the third check result. If the third check result is inconsistent, a third alarm signal is sent; If the third check result is consistent, then the output of the confirming element extraction agent is determined to be the identification result of the email related to over-the-counter derivatives.

Images