Transaction risk detection method, device and equipment

Abstract

The embodiment of the invention provides a transaction risk detection method, device and equipment, and the method comprises the steps: receiving a risk analysis request for a target transaction object, obtaining risk analysis data corresponding to the target transaction object in response to the risk analysis request, and carrying out the risk analysis of the target transaction object through a pre-trained first analysis model, performing error identification processing on the risk analysis data, determining an error category to which the risk analysis data belongs, obtaining a plurality of error sub-categories corresponding to the error category, and performing error diagnosis processing on the risk analysis data by using a pre-trained second analysis model based on the plurality of error sub-categories, the method comprises the steps of obtaining risk analysis data of a target transaction object, obtaining diagnosis data corresponding to the risk analysis data, correcting the risk analysis data based on the diagnosis data, and determining whether the target transaction object has a transaction risk based on the corrected risk analysis data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a transaction risk detection method, device and equipment. BACKGROUND

[0002] With the continuous development of computer technology, automatic risk detection and risk analysis are widely used, and risk analysis report becomes the key basis for transaction risk decision-making. How to improve the accuracy of risk analysis data to protect user privacy and ensure transaction and data security has become the focus of various industries.

[0003] For example, the quality of risk analysis data can be detected by manual, but due to the large amount of risk analysis data and the complex data structure, the efficiency and accuracy of quality detection by manual are low, which seriously affects the transaction risk detection effect. It is necessary to provide a more rapid and reliable quality detection scheme of risk analysis data to improve the efficiency and accuracy of transaction risk detection. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a more rapid and reliable quality detection scheme of risk analysis data to improve the efficiency and accuracy of transaction risk detection.

[0005] In order to achieve the above technical solutions, the embodiments of the present application are implemented as follows: The transaction risk detection method provided by the embodiments of the present application comprises: receiving a risk analysis request for a target transaction object; in response to the risk analysis request, acquiring risk analysis data corresponding to the target transaction object; using a pre-trained first analysis model to perform error identification processing on the risk analysis data, determining an error category to which the risk analysis data belongs, and acquiring a plurality of error subcategories corresponding to the error category; using a pre-trained second analysis model to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories, to obtain diagnosis data corresponding to the risk analysis data, wherein the parameter magnitude of the second analysis model is greater than the parameter magnitude of the first analysis model; performing correction processing on the risk analysis data based on the diagnosis data, and determining whether the target transaction object has transaction risk based on the corrected risk analysis data.

[0006] The embodiment of the present specification provides a transaction risk detection device, the device comprises: a request receiving module configured to receive a risk analysis request for a target transaction object; a data acquisition module configured to acquire risk analysis data corresponding to the target transaction object in response to the risk analysis request; a first analysis module configured to perform error identification processing on the risk analysis data using a pre-trained first analysis model, determine an error category to which the risk analysis data belongs, and acquire a plurality of error sub-categories corresponding to the error category; a second analysis module configured to perform error diagnosis processing on the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model; and a risk detection module configured to perform correction processing on the risk analysis data based on the diagnosis data, and determine whether the target transaction object has a transaction risk based on the corrected risk analysis data.

[0007] The embodiment of the present specification provides a transaction risk detection device, the device comprises: a processor; and a memory arranged to store computer executable instructions, the executable instructions, when executed, cause the processor to: receive a risk analysis request for a target transaction object; acquire risk analysis data corresponding to the target transaction object in response to the risk analysis request; perform error identification processing on the risk analysis data using a pre-trained first analysis model, determine an error category to which the risk analysis data belongs, and acquire a plurality of error sub-categories corresponding to the error category; perform error diagnosis processing on the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model; perform correction processing on the risk analysis data based on the diagnosis data, and determine whether the target transaction object has a transaction risk based on the corrected risk analysis data.

[0008] The embodiment of the present specification further provides a storage medium for storing computer executable instructions, which, when executed by a processor, implement the following processes: receiving a risk analysis request for a target transaction object; in response to the risk analysis request, obtaining risk analysis data corresponding to the target transaction object; using a pre-trained first analysis model, performing error identification processing on the risk analysis data, determining an error category to which the risk analysis data belongs, and obtaining a plurality of error sub-categories corresponding to the error category; using a pre-trained second analysis model, performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories, to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model; performing correction processing on the risk analysis data based on the diagnosis data, and determining whether the target transaction object has a transaction risk based on the corrected risk analysis data.

[0009] The embodiment of the present specification further provides a computer program product, comprising a computer program which, when executed by a processor, implements the following processes: receiving a risk analysis request for a target transaction object; in response to the risk analysis request, obtaining risk analysis data corresponding to the target transaction object; using a pre-trained first analysis model, performing error identification processing on the risk analysis data, determining an error category to which the risk analysis data belongs, and obtaining a plurality of error sub-categories corresponding to the error category; using a pre-trained second analysis model, performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories, to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model; performing correction processing on the risk analysis data based on the diagnosis data, and determining whether the target transaction object has a transaction risk based on the corrected risk analysis data. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present specification or the prior art, brief introductions to the drawings needed in the embodiments or prior art descriptions will be given below. Obviously, the drawings in the following description are only some embodiments described in the present specification, and those skilled in the art can also obtain other drawings according to these drawings without creative labor; Figure 1 a schematic diagram of an implementation environment of a transaction risk detection method of the present specification; Figure 2 a schematic diagram of a processing process of a transaction risk detection method of the present specification; Figure 3a schematic diagram of a diagnostic data determination process of the present specification; Figure 4 a schematic diagram of a risk analysis data generation process of the present specification; Figure 5 a schematic diagram of a two-stage quality inspection process of the present specification; Figure 6 a schematic diagram of a model updating process of the present specification; Figure 7 a schematic diagram of another diagnostic data determination process of the present specification; Figure 8 a schematic diagram of another transaction risk detection apparatus of the present specification; Figure 9 a schematic diagram of a transaction risk detection apparatus of the present specification. DETAILED DESCRIPTION

[0011] The embodiments of the present specification provide various transaction risk detection methods, apparatuses and devices.

[0012] In order to enable persons skilled in the art to better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present specification. Obviously, the described embodiments are only a part of the embodiments of the present specification, rather than all the embodiments. Based on the embodiments in the present specification, all other embodiments obtained by persons skilled in the art without creative labor should belong to the protection scope of the present specification.

[0013] The embodiment of the present specification provides a quality detection scheme of risk analysis data which is faster and more reliable, so as to improve the efficiency and accuracy of transaction risk detection. In order to improve the accuracy of risk analysis data, the quality of risk analysis data can be detected by using evaluation indexes in actual application scenarios, wherein the evaluation indexes can include text overlap degree, word or N-gram matching degree, text semantic similarity, etc. Since the above indexes need high-quality manual reference text, for the text type of risk analysis report which is diversified and often has no unique standard answer, the cost and difficulty of obtaining reference text are high. Since only the surface matching is concerned, semantic, logical, factual errors or misuse of professional terms cannot be captured, there is a problem of lacking semantic understanding, and a single numerical score is output, which cannot provide detailed diagnosis of error type, location and reason, resulting in low accuracy of transaction risk detection. In addition, although the text semantic similarity and other indexes can be evaluated without reference text, their performance still usually depends on reference text, and for specific domain "factual illusion", "logical contradiction", "insufficient analysis depth" and other fine-grained professional errors, it is still impossible to give a clear type and location, and it is also impossible to provide detailed explanation, lacking fine-grained diagnosis, and also having the problems of large amount of calculation and high calculation cost. Alternatively, whether there are specific errors or compliance problems in risk analysis data can also be detected by pre-set expert rules, regular expressions, keyword tables, etc. However, due to the limited coverage, it is difficult to deal with diversified and complex semantic errors, logical errors and illusions in risk analysis data, and the formulation and maintenance of rules require a large amount of manual input, which is difficult to adapt to new risk types and generation modes of risk analysis data, and the maintenance cost is high. In addition, due to the difficulty in processing fuzzy or context-dependent errors, it is difficult to understand deep semantics and lack flexibility. Alternatively, professional risk analysts can also detect the quality of automatically generated risk analysis data. However, due to the slow speed of manual review, it is difficult to meet the quality detection demand of large-scale and high-frequency report generation, and a large amount of human resources need to be invested, resulting in high quality detection cost. In addition, due to the subjective differences of different auditors, the quality detection results are inconsistent, and although experts can point out errors, it is difficult to standardize, structure and quantify feedback, which is not conducive to the automation optimization of the quality detection process. Therefore, the embodiment of the present specification provides a quality detection scheme of risk analysis data which is faster and more reliable. In this scheme, a two-stage quality detection system can be constructed by using a first analysis model and a second analysis model, so as to realize fast error identification and classification by using the first analysis model with small parameter magnitude, and then realize fine diagnosis by using the second analysis model with large parameter magnitude, realizing the combination of fast screening and deep diagnosis, efficiently balancing the quality detection efficiency and depth, so as to quickly and accurately judge whether the target transaction object exists transaction risk based on the corrected risk analysis data, and improve the efficiency and accuracy of transaction risk detection. The specific processing can be referred to the specific content in the following embodiments.

[0014] The transaction risk detection method implemented by one or more embodiments of the present specification can be applied to the implementation environment of transaction risk detection, as shown in the following figure: Figure 1 The implementation environment at least includes: A client 100 and a server 200, in addition, the server 200 can be provided with various network models and algorithms, etc., wherein: The client 100 can run on a terminal device, which can be a mobile phone, a personal computer, a tablet computer, an e-book reader, a wearable device, an AR (Augmented Reality) and VR (Virtual Reality) based information interaction device, and a laptop computer, etc. The terminal device can install the client 100, and the client 100 can be an application, a browser, or a subprogram loaded in the application, etc.

[0015] The server 200 can run on a server, which can be one or more servers, or a server cluster composed of several servers, or a cloud server of a cloud computing platform, etc. The server can install the server 200, and the server 200 can be an application or a subprogram loaded in the application, etc. Various network models and algorithms can be integrated in the server 200, or the server 200 can call one or more of the various network models and algorithms to perform corresponding operations.

[0016] In addition, a database 300 can also be included, which can be set in the server running the server 200, or outside the server running the server 200. The database 300 can store transaction business data of transaction objects, risk analysis data, etc. for transaction business data.

[0017] In this implementation environment, the server 200 can receive a risk analysis request for a target transaction object sent by the client 100, and in response to the risk analysis request, the server 200 can obtain risk analysis data corresponding to the target transaction object, perform error identification processing on the risk analysis data using a pre-trained first analysis model, determine the error category to which the risk analysis data belongs, and obtain a plurality of error subcategories corresponding to the error category. Using a pre-trained second analysis model, based on the plurality of error subcategories, the risk analysis data is subjected to error diagnosis processing, and the diagnosis data corresponding to the risk analysis data is obtained, wherein the parameter order of the second analysis model can be greater than the parameter order of the first analysis model. Based on the diagnosis data, the risk analysis data is modified, and based on the modified risk analysis data, it is determined whether the target transaction object has a transaction risk. The server 200 can return the transaction risk detection result for the target object to the client 100.

[0018] As Figure 2 indicated in the specification, the embodiment provides a transaction risk detection method, and an execution subject of the method can be a server. The server can be an independent server, or a server cluster composed of multiple servers, etc. The server can be a background server of a service such as a financial service or an online shopping service, or a background server of an application program, etc. The method can specifically include the following steps. In step S202, a risk analysis request for a target transaction object is received.

[0019] The target object can be any user or account that can perform a transaction behavior.

[0020] In implementation, the server can determine a user who generates a transaction behavior within a preset detection period as the target object, or the server can determine the user as the target object when detecting that the user triggers a resource transfer transaction service, and trigger a risk analysis request for the target object.

[0021] In step S204, in response to the risk analysis request, risk analysis data corresponding to the target transaction object is obtained.

[0022] The risk analysis data can include a risk analysis conclusion (such as whether it contains a risk, which risk it contains, a risk level, etc.), a derivation process of the risk analysis conclusion, a data source of original data cited in the derivation process, etc.

[0023] In implementation, in response to the risk analysis request, the server can obtain transaction service data of the target transaction object, which can include transaction data (such as transaction time, transaction platform, transaction channel, transaction object, transaction resource quantity, etc.), graph structure data corresponding to the target transaction object (used to represent historical transaction relationships between the target transaction object and other transaction objects), device information of the target transaction object used to trigger the target transaction service (such as device identification, device model, usage frequency, etc.), etc.

[0024] The server can use a pre-trained detection model to generate risk analysis data corresponding to the target transaction object based on the transaction service data, wherein the detection model can be a model constructed based on a preset deep learning algorithm.

[0025] Alternatively, the server can also send the transaction service data to an auditing party, and receive risk analysis data returned by the auditing party.

[0026] In addition, the determination method of the risk analysis data described above is an optional and implementable determination method. In actual application scenarios, there can be various different determination methods, and different determination methods can be selected according to different actual application scenarios. The embodiments of the present specification do not make specific limitations in this regard.

[0027] In step S206, the first analysis model pre-trained is used to perform error identification processing on the risk analysis data, to determine an error category to which the risk analysis data belongs, and to obtain a plurality of error subcategories corresponding to the error category.

[0028] The first analysis model can be a model constructed based on a preset machine learning algorithm.

[0029] In implementation, the server can input the risk analysis data into the first analysis model pre-trained to perform error identification processing on the risk analysis data, to determine the error category to which the risk analysis data belongs. Alternatively, to further improve the error identification speed in the first stage, the server can input a plurality of error types pre-defined and the risk analysis data into the first analysis model pre-trained, to perform error identification processing on whether the risk analysis data has the error pre-defined, to determine the error category to which the risk analysis data belongs.

[0030] After determining the error category to which the risk analysis data belongs, the server can determine a plurality of error subcategories corresponding to the error category based on a corresponding relationship between the error category and the error subcategory. Different error types can correspond to a plurality of different error subcategories, and the corresponding relationship between the error category and the error subcategory can be determined based on historical risk analysis data.

[0031] In step S208, the second analysis model pre-trained is used to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories, to obtain diagnosis data corresponding to the risk analysis data.

[0032] The parameter magnitude of the second analysis model can be greater than the parameter magnitude of the first analysis model. For example, the first analysis model can be a model constructed based on a 3-layer convolutional neural network, and the second analysis model can be a model constructed based on a 10-layer convolutional neural network. In this way, the first analysis model with a smaller parameter magnitude can be used to realize fast screening, and the second analysis model with a larger parameter magnitude can be used to realize deep positioning.

[0033] In implementation, the server can input the risk analysis data and the plurality of error subcategories into the second analysis model pre-trained, to obtain the diagnosis data corresponding to the risk analysis data.

[0034] The diagnostic data can include risk analysis data whether there is an error related to the error subcategory, and related diagnostic information of the existing error, such as whether correction is needed, correction cost, correction tool, and the like.

[0035] In step S210, the risk analysis data is corrected based on the diagnostic data, and whether the target transaction object has a transaction risk is determined based on the corrected risk analysis data.

[0036] In implementation, the server can determine the correction cost of the correction processing of the risk analysis data according to the diagnostic data. For example, the server can determine the number of errors existing in the risk analysis data, the calling cost of the correction tool required for the correction processing, and the like according to the diagnostic data, to determine the correction cost of the correction processing, so as to determine the corresponding correction strategy according to the correction cost, and then correct the risk analysis data based on the correction strategy.

[0037] For example, in the case of low correction cost, the correction processing can be performed through the manual correction strategy, and in the case of high correction cost, the correction processing can be performed through the model correction strategy.

[0038] After the correction processing, whether the target transaction object has a transaction risk can be determined based on the corrected risk analysis data. For example, the corrected risk analysis data can be sent to a preset auditing party, and whether the target transaction object has a transaction risk can be determined according to the judgment result of the preset auditing party.

[0039] In addition, in the case where it is determined that the target transaction object has a transaction risk, the corresponding risk control strategy can also be determined based on the corrected risk analysis data. The server can also store the corrected risk analysis data to ensure the traceability of the transaction risk detection process.

[0040] The embodiment of the specification provides a transaction risk detection method, receives a risk analysis request for a target transaction object, acquires risk analysis data corresponding to the target transaction object in response to the risk analysis request, performs error identification processing on the risk analysis data by using a pre-trained first analysis model, determines an error category to which the risk analysis data belongs, and acquires a plurality of error subcategories corresponding to the error category, performs error diagnosis processing on the risk analysis data based on the plurality of error subcategories by using a pre-trained second analysis model, and obtains diagnosis data corresponding to the risk analysis data, wherein the parameter magnitude of the second analysis model can be greater than the parameter magnitude of the first analysis model, the risk analysis data is corrected based on the diagnosis data, and whether the target transaction object has a transaction risk is determined based on the corrected risk analysis data. In this way, a two-stage quality inspection system can be constructed by using the first analysis model and the second analysis model to realize rapid error identification and classification by using the first analysis model with a smaller parameter magnitude, and then to realize fine diagnosis by using the second analysis model with a larger parameter magnitude, thereby realizing the combination of rapid screening and deep diagnosis, efficiently balancing the quality inspection efficiency and the quality inspection depth, and quickly and accurately judging whether the target transaction object has a transaction risk based on the corrected risk analysis data, thereby improving the transaction risk detection efficiency and accuracy.

[0041] In actual application, the specific processing manner of the step S208 of performing error diagnosis processing on the risk analysis data based on the plurality of error subcategories by using the pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data can be various, and an optional determination method is provided as follows. Figure 3 As shown in the figure, the specific processing can include the following steps S2082-S2084.

[0042] In step S2082, business domain knowledge corresponding to a business domain to which the transaction business data belongs is acquired.

[0043] The business domain knowledge can include academic papers, review specifications, industry specifications, and the like, and the business domain can include a financial domain, an education domain, a financial domain, and the like.

[0044] In step S2084, the second analysis model is pre-trained, error diagnosis processing is performed on the risk analysis data based on the plurality of error subcategories and the business domain knowledge, and diagnosis data corresponding to the risk analysis data is obtained.

[0045] In implementation, the server can obtain the knowledge graph data constructed by the business domain knowledge, and utilize the pre-trained second analysis model to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories and the knowledge graph data, to obtain diagnosis data corresponding to the risk analysis data. In this way, the professional terms, logical relationships and factual information can be deeply understood, so as to identify deep semantic errors such as "fact illusion", "logical contradiction" and "insufficient analysis depth" which are difficult to be found by the traditional rule system.

[0046] In actual application, the diagnosis data can include one or more of the error subcategory to which the risk analysis data belongs, the error position, the error interpretation data, the error impact analysis data and the correction suggestion data.

[0047] In implementation, the diagnosis data can include a structured diagnosis report, which can include the following information: (1) Error type (i.e. error subcategory): accurate fine-grained error classification.

[0048] (2) Error position: specific paragraph, sentence or phrase in the original text of the risk analysis data.

[0049] (3) Error interpretation: why it is an error and which quality inspection standard is violated.

[0050] (4) Impact analysis: possible consequences of the error.

[0051] (5) Correction suggestion: targeted correction method or example.

[0052] In actual application, the specific processing manner of obtaining the risk analysis data corresponding to the target transaction object in step S204 can be various, and an optional determination method is provided as follows. Figure 4 As shown in FIG. 20, the specific processing can include the following steps S2042-S2044.

[0053] In step S2042, the risk detection requirement corresponding to the target transaction object and the transaction business data of the target transaction object are obtained.

[0054] The risk detection requirement can include the detection requirement of whether there is a risk and the generation requirement of the risk analysis data, the detection requirement can include the risk detection intensity, the risk judgment logic, the compliance judgment logic, the risk level division basis, etc., the generation requirement can include the analysis depth, the data volume of the generated data, the detail degree of the deduction process of the risk analysis, etc., and the transaction business data can include the transaction data, the graph structure data corresponding to the target transaction object, the device information of the target transaction object for triggering the target transaction business and the like.

[0055] In step S2044, the preset large language model is used to generate risk analysis data corresponding to the target transaction object based on the risk detection requirements and the transaction business data.

[0056] In implementation, as shown in Figure 5 , the server can use a dual-stage explainable quality control framework (DEQC) to perform efficient, fine-grained and explainable quality inspection on the risk analysis data generated by the large language model 1, thereby improving the data quality of the risk analysis data. This framework combines speed and depth, and ensures the comprehensiveness of quality inspection through a clearly defined error category system.

[0057] Firstly, the user can input the risk detection requirements corresponding to the target transaction object into the large language model. The large language model can generate risk analysis data C in combination with the transaction business data of the target transaction object, so as to take the risk analysis data C as the input data of the DEQC framework.

[0058] In actual application, the large language model can also be updated. The specific processing method of model updating can be various, and the following provides an optional determination method, as shown in Figure 6 , which can specifically include the following step S602 processing.

[0059] In step S602, the preset large language model is updated based on the corrected risk analysis data and the risk analysis data, to obtain an updated large language model.

[0060] In implementation, as shown in Figure 5 , the DEQC can decompose the quality inspection task of complex risk analysis data into two stages of cooperative work. The first stage can quickly identify the main quality problems and realize efficient filtering, and the second stage can perform in-depth diagnosis on the basis of the first stage and provide detailed explainable feedback.

[0061] Specifically, stage one can be a rapid risk identification and categorization stage, in which: The first analysis model is a lightweight QC LLM. For example, a fine-tuned, efficiency-first large language model 2 can be deployed, for example, the first analysis model can be based on Llama-2-7B or a smaller scale model.

[0062] The first analysis model can perform a preliminary and fast quality scan on the input risk analysis data. The core task is to determine whether there are major quality defects in the risk analysis data and to classify them into predefined error categories.

[0063] Specifically, the lightweight quality inspection model (i.e., the first analysis model) can be provided with concise preset prompt information to limit the first analysis model to focus on quickly identifying the overall quality status of the risk analysis data (e.g., "qualified" or "exists quality defects", etc.). The lightweight quality inspection model can output a preliminary quality inspection result based on the preset prompt information, which can include a binary judgment result of whether there is a quality problem (Pass / Fail) and a coarse-grained error category such as "accuracy problem", "professional problem", "completeness problem", etc.

[0064] The first stage emphasizes reasoning speed, aiming to achieve real-time or near-real-time preliminary screening, quickly filtering out obviously unqualified risk analysis data, and guiding the problems in the risk analysis data to the next stage.

[0065] Stage two can be a detailed diagnosis and explainable report generation stage, in which: The second analysis model can be an in-depth quality inspection model (In-depth QC LLM), such as a large language model 3 with stronger capabilities and larger parameters, for example, the second analysis model can be GPT-4 or a deeply fine-tuned Llama-2-70B model.

[0066] Based on the identification of the main error categories in the previous stage, the second analysis model can perform comprehensive and detailed error diagnosis on the risk analysis data and generate diagnosis data containing detailed and explainable structured diagnosis reports G.

[0067] Specifically, the risk analysis data and the main error categories output by the first stage, as well as the multiple error subcategories contained therein, can be input into the in-depth quality inspection model (i.e., the second analysis model). Detailed analysis instructions are provided to the in-depth quality inspection model, requiring it to conduct a comprehensive review of the risk analysis data according to the predefined fine-grained error category system (including factual illusion, term misuse, logical contradiction, analysis depth deficiency, key information omission, bias tendency, syntax error, etc. Error subcategories).

[0068] At the same time, the in-depth quality inspection model can query the domain knowledge base during the diagnosis process to verify the accuracy of professional concepts, data, and logic in the report.

[0069] In addition, to enhance the explainability of the model output results, the server can also guide the large language model 3 to gradually reason through the prompt engineering technology such as CoT (Chain-of-Thought) or WoT (Workshop-of-Thoughts), and explain in detail the type, specific location, cause, potential impact of each error, and provide specific modification suggestions.

[0070] Finally, as Figure 5 shown, quality inspection feedback and continuous improvement can be performed based on the corrected risk analysis data and the risk analysis data, for example, the risk analysis data and the diagnosis data can be sent to decision makers to assist the decision makers in understanding the reported problems and making manual corrections or decisions.

[0071] On the other hand, the diagnosis data can also be used as LLM optimization guidance for iterative training and optimization of the above-mentioned large language model 1 (i.e., the large language model used to generate risk analysis data) and / or large language model 2 (i.e., the large language model used to perform preliminary quality judgment), forming a closed-loop quality improvement mechanism.

[0072] Among them, the server can determine the error recall rate (Error Recall Rate, ERR) and the diagnosis accuracy rate (Diagnosis Precision Rate, DPR) based on the corrected risk analysis data and the risk analysis data, where ERR can be used to measure the proportion of all actual existing errors identified by the first analysis system, ERR = number of correctly identified errors / total number of actual existing errors, and DPR can be used to measure the proportion of actual true errors in the errors diagnosed by the quality inspection system, DPR = number of correctly identified and diagnosed errors / total number of diagnosed errors.

[0073] The server can perform iterative update processing on the model (such as the first analysis model and / or the second analysis model) based on ERR and / or DPR.

[0074] Alternatively, the server can also use the reinforcement learning human feedback (RLHF) algorithm to update the above-mentioned large language model 1 and / or large language model 2 based on the corrected risk analysis data and the risk analysis data, and in addition, to save data processing costs, the above-mentioned large language model 1 and large language model 2 can be the same large language model, i.e., the lightweight quality inspection model can also be used to generate risk analysis data based on risk detection requirements and transaction business data, and generate preliminary quality inspection results for the risk analysis data.

[0075] The DEQC does not rely on manually written standard reference reports for quality evaluation, but directly analyzes the internal quality of the risk analysis data, the consistency with the instructions, and the matching degree with the domain knowledge through a lightweight quality inspection model and a deep quality inspection model. This design enables the DEQC to deal with risk analysis data that lacks a unique standard answer, effectively overcoming the "dependence on reference text" disadvantage.

[0076] In addition, both stages of DEQC can be used for large language models. Since large language models have strong semantic understanding and reasoning capabilities, DEQC can overcome the "lack of semantic understanding and lack of explainability" disadvantage. For example, the deep quality inspection model can combine domain business knowledge to deeply understand professional terms, logical relationships, and factual information, thereby identifying deep semantic errors such as "fact illusion", "logical contradiction", and "insufficient analysis depth" that traditional rule systems cannot detect. In addition, through CoT and WoT prompting engineering, the deep quality inspection model can generate structured diagnostic data, clearly indicating the error type, location, cause, and modification suggestions, providing explainability and helping to optimize and update large language models.

[0077] In addition, the lightweight quality inspection model in stage one can also be used for filtering risk analysis data. For example, the first analysis model can perform preliminary quality evaluation on multiple risk analysis data and filter out risk analysis data with more than a preset number of problems, i.e., most unqualified data can be quickly filtered out, avoiding resource waste caused by deep analysis of all risk analysis data.

[0078] The deep quality inspection model in stage two performs fine-grained diagnosis only on errors that need further investigation after preliminary judgment. This divide-and-conquer strategy allows the entire quality inspection process to maintain depth while improving overall efficiency, avoiding long reasoning times caused by a single model performing all tasks at once, and overcoming the "difficulty in balancing efficiency and depth" disadvantage.

[0079] In addition, DEQC can achieve automated quality inspection. Since machine quality inspection is more consistent than manual inspection, it can reduce subjective differences while improving quality inspection efficiency and reducing dependence on manual review and related costs. It can also provide diagnostic data including structured and explainable diagnostic reports, improving the efficiency of risk analysts for manual review and modification, transforming from "reviewers" to "decision makers", and overcoming the "low efficiency, high cost, and subjectivity" disadvantage.

[0080] In actual application, the specific processing manner of the risk analysis data corresponding to the diagnosis data obtained by using the pre-trained second analysis model to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories in step S208 can be various, and an optional determination method is provided as follows, which can specifically include the following steps S2086-S2088. Figure 7 As shown, the specific processing manner of the risk analysis data corresponding to the diagnosis data obtained by using the pre-trained second analysis model to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories in step S208 can be various, and an optional determination method is provided as follows, which can specifically include the following steps S2086-S2088.

[0081] In step S2086, the target analysis data corresponding to each error subcategory in the historical risk analysis data is obtained.

[0082] The historical risk analysis data can be risk analysis data for historical transaction objects.

[0083] In step S2088, the pre-trained second analysis model is used to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories and the target analysis data corresponding to each error subcategory, to obtain diagnosis data corresponding to the risk analysis data.

[0084] In implementation, the target analysis data corresponding to each error subcategory can be used to provide case guidance for the second analysis model, so as to improve the determination accuracy of the diagnosis data.

[0085] In actual application, the specific processing manner of the risk analysis data corresponding to the diagnosis data obtained by using the pre-trained second analysis model to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories in step S208 can be various, and an optional determination method is provided as follows, which can specifically include the following steps A1-A2.

[0086] In step A1, prompt information corresponding to each error subcategory is generated based on the target analysis data corresponding to the error subcategory.

[0087] The prompt information can include an error reasoning process corresponding to the error subcategory.

[0088] In implementation, the server can use a pre-set large language model to generate an error reasoning process for determining the error subcategory of the target analysis data, and determine the prompt information corresponding to each error subcategory according to the error reasoning process.

[0089] In step A2, the pre-trained second analysis model is used to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories, the target analysis data corresponding to each error subcategory, and the prompt information, to obtain diagnosis data corresponding to the risk analysis data.

[0090] In implementation, the server can guide the model to complete a specific task by providing a small number of examples in the input to improve its performance on zero-shot or few-shot tasks, i.e., the server can utilize the target analysis data and prompt information corresponding to each error subcategory to improve the diagnostic performance of the second analysis model.

[0091] In practical application, the error categories can include an accuracy error category, a professionalism error category, and a completeness error category, the accuracy error category can include one or more error subcategories of factual illusion, data reference error, and logical contradiction error, the professionalism error category can include one or more error subcategories of term misuse, insufficient analysis depth, and insufficient conclusion support, and the completeness error category can include one or more error subcategories of objectivity and neutrality, readability and structure.

[0092] The accuracy error category can be used to represent that the text content has inaccuracies, untruths, or contradictions in facts, data, or logic, affecting the credibility and effectiveness of the information. The factual illusion subcategory can be used to represent that the text contains false or fabricated information that does not conform to objective facts, training data, or external knowledge. The data reference error subcategory can be used to represent that the data referenced in the text is inaccurate, outdated, or unreliable. The logical contradiction error subcategory can be used to represent that the text content has self-contradictions or reasoning inconsistencies in logic.

[0093] The professionalism error category can be used to represent that the text has deficiencies in term usage, analysis depth, or conclusion support, failing to reflect professional competence and rigor. The term misuse subcategory can be used to represent that the text incorrectly uses professional terms, resulting in inaccurate information transmission or ambiguity. The insufficient analysis depth subcategory can be used to represent that the text's analysis of the problem stops at the surface and fails to delve into the essence or propose valuable insights. The insufficient conclusion support subcategory can be used to represent that the text's conclusion lacks sufficient and strong argument support.

[0094] The completeness error category can be used to represent that the text has defects in objectivity, neutrality, readability, or structure, failing to fully and clearly convey information. The objectivity and neutrality subcategory can be used to represent that the text content has bias or bias, failing to present facts objectively and neutrally. The readability and structure subcategory can be used to represent that the text has problems in language expression, paragraph arrangement, or logical connection, affecting reading fluency and understanding effect.

[0095] Thus, a multi-dimensional and fine-grained error category system covering "accuracy (factitious illusion, data reference error, logical contradiction)", "professionalism (term misuse, insufficient analysis depth, insufficient conclusion support)", "completeness", "objectivity and neutrality", and "readability and structure" is established according to the characteristics of the risk analysis data. By introducing the above fine-grained error classification system for risk analysis data, factitious illusion, logical contradiction and professional errors that are difficult to be found by traditional methods can be accurately identified and explained.

[0096] The embodiment of the present specification provides a transaction risk detection method, receiving a risk analysis request for a target transaction object, in response to the risk analysis request, obtaining risk analysis data corresponding to the target transaction object, using a pre-trained first analysis model to perform error identification processing on the risk analysis data, determining the error category to which the risk analysis data belongs, and obtaining a plurality of error subcategories corresponding to the error category, using a pre-trained second analysis model, based on the plurality of error subcategories, performing error diagnosis processing on the risk analysis data, obtaining diagnosis data corresponding to the risk analysis data, wherein the parameter magnitude of the second analysis model can be greater than the parameter magnitude of the first analysis model, based on the diagnosis data, performing correction processing on the risk analysis data, and based on the corrected risk analysis data, determining whether the target transaction object has a transaction risk. In this way, a two-stage quality inspection system can be constructed by the first analysis model and the second analysis model to realize fast error identification and classification by the first analysis model with smaller parameter magnitude, and then realize fine diagnosis by the second analysis model with larger parameter magnitude, realizing the combination of fast screening and deep diagnosis, efficiently balancing the quality inspection efficiency and the quality inspection depth, to quickly and accurately determine whether the target transaction object has a transaction risk based on the corrected risk analysis data, improving the transaction risk detection efficiency and accuracy.

[0097] The above is the transaction risk detection method provided by the embodiment of the present specification. Based on the same idea, the embodiment of the present specification also provides a transaction risk detection device, as shown in Figure 8 .

[0098] The transaction risk detection device includes a request receiving module 801, a data acquisition module 802, a first analysis module 803, a second analysis module 804, and a risk detection module 805, wherein: The request receiving module 801 is configured to receive a risk analysis request for a target transaction object; The data acquisition module 802 is configured to obtain risk analysis data corresponding to the target transaction object in response to the risk analysis request; The first analysis module 803 is configured to perform error identification processing on the risk analysis data by using a pre-trained first analysis model, determine an error category to which the risk analysis data belongs, and obtain a plurality of error sub-categories corresponding to the error category. The second analysis module 804 is configured to perform error diagnosis processing on the risk analysis data based on the plurality of error sub-categories by using a pre-trained second analysis model, to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model. The risk detection module 805 is configured to perform correction processing on the risk analysis data based on the diagnosis data, and determine whether the target transaction object has a transaction risk based on the corrected risk analysis data.

[0099] In the embodiments of the present specification, the second analysis module 804 is configured to: obtain business domain knowledge corresponding to a business domain to which the transaction business data belongs; perform error diagnosis processing on the risk analysis data based on the plurality of error sub-categories and the business domain knowledge by using the pre-trained second analysis model, to obtain diagnosis data corresponding to the risk analysis data.

[0100] In the embodiments of the present specification, the diagnosis data includes one or more of an error sub-category to which the risk analysis data belongs, an error position, error explanation data, error impact analysis data, and correction suggestion data.

[0101] In the embodiments of the present specification, the data acquisition module 802 is configured to: acquire a risk detection requirement corresponding to the target transaction object and transaction business data of the target transaction object; generate risk analysis data corresponding to the target transaction object based on the risk detection requirement and the transaction business data by using a pre-set large language model.

[0102] In the embodiments of the present specification, the apparatus further includes: The model updating module is configured to update the pre-set large language model based on the corrected risk analysis data and the risk analysis data, to obtain an updated large language model.

[0103] In the embodiments of the present specification, the second analysis module 804 is configured to: obtain target analysis data corresponding to each of the error sub-categories in historical risk analysis data; The pre-trained second analysis model is used to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories and the target analysis data corresponding to each error subcategory, to obtain diagnosis data corresponding to the risk analysis data.

[0104] In the embodiments of the present specification, the second analysis module 804 is configured to: generate prompt information corresponding to each error subcategory based on the target analysis data corresponding to the error subcategory, wherein the prompt information includes an error reasoning process corresponding to the error subcategory; The pre-trained second analysis model is used to perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories and the target analysis data corresponding to each error subcategory and the prompt information, to obtain diagnosis data corresponding to the risk analysis data.

[0105] In the embodiments of the present specification, the error categories include accuracy error categories, professionalism error categories, and integrity error categories, the accuracy error categories include one or more error subcategories of factual illusion, data reference error, and logical contradiction error, the professionalism error categories include one or more error subcategories of term misuse, insufficient analysis depth, and insufficient conclusion support, and the integrity error categories include one or more error subcategories of objectivity and neutrality, and readability and structure.

[0106] The embodiments of the present specification provide a transaction risk detection apparatus, which receives a risk analysis request for a target transaction object, acquires risk analysis data corresponding to the target transaction object in response to the risk analysis request, performs error identification processing on the risk analysis data by using a pre-trained first analysis model, determines an error category to which the risk analysis data belongs, acquires a plurality of error subcategories corresponding to the error category, performs error diagnosis processing on the risk analysis data based on the plurality of error subcategories by using a pre-trained second analysis model, obtains diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model can be greater than a parameter magnitude of the first analysis model, performs correction processing on the risk analysis data based on the diagnosis data, and determines whether the target transaction object has a transaction risk based on the corrected risk analysis data. In this way, a two-stage quality inspection system can be constructed by using the first analysis model and the second analysis model to realize rapid error identification and classification by using the first analysis model with a smaller parameter magnitude, and then to realize fine diagnosis by using the second analysis model with a larger parameter magnitude, so as to efficiently balance the quality inspection efficiency and the quality inspection depth, to quickly and accurately determine whether the target transaction object has a transaction risk based on the corrected risk analysis data, and to improve the transaction risk detection efficiency and accuracy.

[0107] The transaction risk detection device provided by the embodiments of the present specification can be a terminal device or a server, etc. Figure 9

[0108] The transaction risk detection device can be a terminal device or a server, etc.

[0109] As shown in Figure 9 The device 900 is mainly composed of a communication interface 902, a user interface 904, a processor 906 and a data storage 908, which are connected and communicate with each other through a system bus, a network or other connection mechanism 910. The communication interface 902 enables the device 900 to communicate with other devices, access networks and transmission networks through analog or digital modulation. For example, the communication interface 902 can include a chipset and an antenna for wireless communication with a radio access network or an access point. In addition, the communication interface 902 can also be a wired interface such as an Ethernet, token ring or USB port, or a wireless interface such as Wifi, Bluetooth, Global Positioning System (GPS) or Wide Area Wireless Interface (e.g. WiMAX or LTE). Of course, the communication interface 902 can also support other forms of physical layer interface and standard or proprietary communication protocol. The communication interface 902 can also include multiple physical communication interfaces, such as a Wifi interface, a Bluetooth interface and a wide area wireless interface, etc.

[0110] The user interface 904 includes receiving user input and providing output to the user. Therefore, the user interface 904 can include input components such as a keypad, a keyboard, a sensitive or presence panel, a computer mouse, a trackball, a joystick, a microphone, a still camera and a video camera, and can also include output components such as a display screen (which can be combined with a touch-sensitive panel), a CRT, an LCD, an LED, a display using DLP technology, a printer, other known or future developed similar devices, etc. The user interface 904 can also generate audible output through a speaker, a speaker jack, an audio output port, an audio output device, a headset and other known or future developed similar devices. In some embodiments, the user interface 904 can include software, circuitry or other forms of logic capable of transmitting data to and receiving data from external user input / output devices. In addition or alternatively, the device 900 can support remote access from other devices through the communication interface 902 or another physical interface (not shown). The user interface 904 can be configured to receive user input, the position and movement of which can be indicated by the indicators or cursors described herein. The user interface 904 can also be configured as a display device for rendering or displaying a text segment.

[0111] The processor 906 can include one or more general-purpose processors and / or special-purpose processors. ​

[0112] Data storage 908 can include one or more volatile storage components and can be integrated in whole or in part with processor 906. Data storage 908 can include removable and / or non-removable storage components.

[0113] Processor 906 is capable of executing program instructions stored in data storage 908 (e.g., compiled or interpreted program logic and / or machine code) to implement various functionality described herein. Data storage 908 can include non-transitory computer-readable media having stored thereon program instructions that, when executed by device 900, enable device 900 to carry out any of the methods, processes, or functionality disclosed in the specification and / or drawings. Execution of the program instructions 918 by processor 906 can cause processor 906 to utilize data 912.

[0114] For example, program instructions 918 can include an operating system 922 (e.g., an operating system kernel, device drivers, and / or other modules) installed on device 900, as well as one or more applications 920 (e.g., a browser, a social application, or a gaming application). Similarly, data 912 can include operating system data 916 and application data 914. Operating system data 916 is primarily accessible to operating system 922, while application data 914 is primarily accessible to one or more applications 920. Application data 914 can be located in a file system that is visible or hidden to a user of device 900.

[0115] Applications 920 can communicate with operating system 912 through one or more application programming interfaces (APIs). These APIs facilitate applications 920 reading from and / or writing to application data 914, transmitting or receiving information via communication interface 902, receiving or displaying information on user interface 904, and the like.

[0116] In some terminology, applications 920 can be referred to simply as "apps." Furthermore, applications 920 can be downloaded to device 900 through one or more online application stores or application markets. However, applications can also be installed on device 900 through other means, such as through a web browser or a physical interface on device 900 (e.g., a USB port).

[0117] In particular embodiments, the transaction risk detection device includes a memory, and one or more programs, wherein the one or more programs are stored in the memory and the one or more programs can include one or more modules, and each module can include a set of computer-executable instructions for a transaction risk detection device and configured to be executed by one or more processors of the one or more programs include computer-executable instructions for: receive a risk analysis request for a target transaction object; In response to the risk analysis request, obtain risk analysis data corresponding to the target transaction object; Perform error identification processing on the risk analysis data using a pre-trained first analysis model, determine an error category to which the risk analysis data belongs, and obtain a plurality of error sub-categories corresponding to the error category; Perform error diagnosis processing on the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model, wherein the parameter magnitude of the second analysis model is greater than the parameter magnitude of the first analysis model, to obtain diagnosis data corresponding to the risk analysis data. Based on the diagnosis data, correct the risk analysis data, and based on the corrected risk analysis data, determine whether the target transaction object has a transaction risk.

[0118] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the transaction risk detection device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0119] The embodiment of the specification provides a transaction risk detection device, which receives a risk analysis request for a target transaction object, obtains risk analysis data corresponding to the target transaction object in response to the risk analysis request, performs error identification processing on the risk analysis data using a pre-trained first analysis model, determines an error category to which the risk analysis data belongs, and obtains a plurality of error sub-categories corresponding to the error category, performs error diagnosis processing on the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model, wherein the parameter magnitude of the second analysis model can be greater than the parameter magnitude of the first analysis model, corrects the risk analysis data based on the diagnosis data, and determines whether the target transaction object has a transaction risk based on the corrected risk analysis data. In this way, a two-stage quality inspection system can be constructed by the first analysis model and the second analysis model to realize rapid error identification and classification by the first analysis model with a smaller parameter magnitude, and then realize fine diagnosis by the second analysis model with a larger parameter magnitude, thereby combining rapid screening and deep diagnosis, efficiently balancing the quality inspection efficiency and the quality inspection depth, and quickly and accurately judging whether the target transaction object has a transaction risk based on the corrected risk analysis data, thereby improving the transaction risk detection efficiency and accuracy.

[0120] Further, based on the above Figures 1 to 7The one or more embodiments of the specification also provide a storage medium for storing computer executable instruction information, and in a specific embodiment, the storage medium can be a U disk, an optical disk, a hard disk, etc. The computer executable instruction information stored in the storage medium, when executed by a processor, can implement the following processes: receiving a risk analysis request for a target transaction object; in response to the risk analysis request, obtaining risk analysis data corresponding to the target transaction object; using a pre-trained first analysis model, performing error identification processing on the risk analysis data, determining an error category to which the risk analysis data belongs, and obtaining a plurality of error sub-categories corresponding to the error category; using a pre-trained second analysis model, performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories, to obtain diagnosis data corresponding to the risk analysis data, wherein the parameter magnitude of the second analysis model is greater than the parameter magnitude of the first analysis model; based on the diagnosis data, performing correction processing on the risk analysis data, and based on the corrected risk analysis data, determining whether the target transaction object has a transaction risk.

[0121] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the above-mentioned storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0122] The embodiment of the present specification provides a storage medium, receives a risk analysis request for a target transaction object, acquires risk analysis data corresponding to the target transaction object in response to the risk analysis request, performs error identification processing on the risk analysis data by using a pre-trained first analysis model, determines an error category to which the risk analysis data belongs, and acquires a plurality of error sub-categories corresponding to the error category, performs error diagnosis processing on the risk analysis data based on the plurality of error sub-categories by using a pre-trained second analysis model, and obtains diagnosis data corresponding to the risk analysis data, wherein the parameter magnitude of the second analysis model can be greater than the parameter magnitude of the first analysis model, the risk analysis data is corrected based on the diagnosis data, and whether the target transaction object has a transaction risk is determined based on the corrected risk analysis data. In this way, a two-stage quality inspection system can be constructed by the first analysis model and the second analysis model to realize rapid error identification and classification by the first analysis model with a smaller parameter magnitude, and then to realize fine diagnosis by the second analysis model with a larger parameter magnitude, realizing the combination of rapid screening and deep diagnosis, efficiently balancing the quality inspection efficiency and the quality inspection depth, and quickly and accurately judging whether the target transaction object has a transaction risk based on the corrected risk analysis data, improving the transaction risk detection efficiency and accuracy.

[0123] Further, based on the above Figures 1 to 7 The one or more embodiments of the present specification also provide a computer program product comprising a computer program, wherein the computer program in the computer program product, when executed by a processor, can implement the following processes: receiving a risk analysis request for a target transaction object; acquiring risk analysis data corresponding to the target transaction object in response to the risk analysis request; performing error identification processing on the risk analysis data by using a pre-trained first analysis model, determining an error category to which the risk analysis data belongs, and acquiring a plurality of error sub-categories corresponding to the error category; performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories by using a pre-trained second analysis model, and obtaining diagnosis data corresponding to the risk analysis data, wherein the parameter magnitude of the second analysis model is greater than the parameter magnitude of the first analysis model; correcting the risk analysis data based on the diagnosis data, and determining whether the target transaction object has a transaction risk based on the corrected risk analysis data.

[0124] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for the above-mentioned computer program product embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0125] The embodiment of the specification provides a computer program product, receives a risk analysis request for a target transaction object, acquires risk analysis data corresponding to the target transaction object in response to the risk analysis request, performs error identification processing on the risk analysis data by using a pre-trained first analysis model, determines an error category to which the risk analysis data belongs, and acquires a plurality of error subcategories corresponding to the error category, performs error diagnosis processing on the risk analysis data based on the plurality of error subcategories by using a pre-trained second analysis model, and obtains diagnosis data corresponding to the risk analysis data, wherein the parameter magnitude of the second analysis model can be greater than the parameter magnitude of the first analysis model, and the risk analysis data is corrected based on the diagnosis data, and whether the target transaction object has a transaction risk is determined based on the corrected risk analysis data. In this way, a two-stage quality inspection system can be constructed by the first analysis model and the second analysis model to realize rapid error identification and classification by the first analysis model with smaller parameter magnitude, and then realize fine diagnosis by the second analysis model with larger parameter magnitude, realize the combination of rapid screening and deep diagnosis, and efficiently balance the quality inspection efficiency and the quality inspection depth, so as to quickly and accurately judge whether the target transaction object has a transaction risk based on the corrected risk analysis data, and improve the transaction risk detection efficiency and accuracy.

[0126] The above describes specific embodiments of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different than the order in the embodiments and still achieve the desired result. In addition, the processes depicted in the figures do not necessarily require the particular order shown or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be utilized or can be advantageous.

[0127] In the 1990s, it was possible to distinguish whether an improvement in a technology was a hardware improvement (e.g., an improvement in the circuit structure of a diode, transistor, switch, etc.) or a software improvement (an improvement in a method flow). However, as technology has advanced, many improvements in method flows today can be considered as direct improvements in hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into a hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is an integrated circuit whose logic function is determined by the user programming the device. A digital system is "integrated" on a PLD by the designer programming it himself, without having to ask a chip manufacturer to design and manufacture a special integrated circuit chip. Moreover, instead of manually manufacturing an integrated circuit chip, this programming is now mostly implemented using "logic compiler" software, which is similar to the software compiler used when developing a program, and the original code before compilation must also be written in a specific programming language, which is called a hardware description language (HDL), and there are many types of HDL, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., and the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that it is only necessary to logically program the method flow using the above-mentioned hardware description languages and program it into an integrated circuit to easily obtain a hardware circuit that implements the logical method flow.

[0128] The controller can be implemented in any suitable way, for example, the controller can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code, such as software or firmware, executable by the microprocessor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that, in addition to implementing the controller in pure computer readable program code, it is possible to implement the controller in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers, etc. to perform the same functions by logically programming the method steps. Such a controller can therefore be considered to be a hardware component, and the means included therein for performing various functions can also be considered to be structures within the hardware component. Alternatively, the means for performing various functions can even be considered to be both a software module implementing the method and a structure within the hardware component.

[0129] The systems, apparatuses, modules or units illustrated by the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0130] For the sake of description, the above apparatuses are described in functional division and are described respectively. Of course, the functions of each unit can be implemented in the same or more software and / or hardware when implementing one or more embodiments of the present specification.

[0131] Those skilled in the art will understand that the embodiments of the present specification can be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of the present specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the present specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] The embodiments of the present specification are described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to embodiments of the present specification. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus generate a device implemented in the flowcharts Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks

[0133] These computer program instructions can also be stored in a computer readable memory capable of directing the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices, which implement the functions specified in the flowcharts Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks

[0134] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide steps for implementing the functions specified in the flowcharts Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks

[0135] In a typical configuration, the computing device includes one or more processors (CPU), input / output interface, network interface and memory.

[0136] The memory can include non-persistent memory in the computer readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer readable media.

[0137] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0138] It should also be noted that the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed or inherent to such processes, methods, articles or devices. Without more limitations, it does not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0139] It should also be noted that the terms "one", "a" and "the" do not necessarily refer to the singular, but can also include the plural. The ordinal terms first, second, etc. do not necessarily indicate order, but are often used to distinguish objects. For example, the first server, the second server, usually refers to two servers. In order to distinguish the two servers, they are expressed as the first server and the second server. Of course, the two servers may also be the same server at times.

[0140] In this specification, unless specifically stated, "receiving, sending of data" is not necessarily direct receiving and sending, but can be indirect receiving and sending. For example, A receives data sent by B, which can be understood as A directly receiving data sent by B, or A indirectly receiving data sent by B through C and other subjects. Similarly, B sends data to A, which can be understood as B directly sending data to A, or B indirectly sending data to A through C and other subjects. Here C can be a subject, or two or more subjects.

[0141] The description uses specific terminology including the following. The use of "an" or "one" embodiment or implementation of the description is intended to mean that a particular feature, structure, or characteristic described is included in at least one embodiment of the description. Therefore, use of the terms in the description includes instances of both the singular and the plural unless otherwise specifically stated or clearly contradicted by context. So, for example, it means that a feature, structure, or characteristic is included in one embodiment or implementation of the description or in multiple different embodiments or implementations of the description. The applications described herein are not limited to the embodiments and implementations described and shown, but can be used in conjunction with numerous other arrangements.

[0142] Although the description one or more embodiments provide method steps as recited in the embodiments or flow charts, it is understood that the order of the steps recited in the embodiments or flow charts is merely one way of executing the steps, and is not meant to be the only way of executing the steps. Therefore, the order of the steps recited in the claims is not meant to be the only way of executing the steps, and the order of the steps recited in the claims can be adjusted, or the steps can be executed in parallel, and still be within the scope of the claims.

[0143] It is noted that the user data obtained by the description is authorized by the user and does not involve the user's privacy.

[0144] It is understood by those skilled in the art that the embodiments of the description can be provided as a method, a system or a computer program product. Therefore, one or more embodiments of the description can be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the description can be in the form of a computer program product embodied on one or more computer usable storage media including, but not limited to, magnetic disk storage, CD-ROM, optical storage and the like.

[0145] One or more embodiments of the description can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like, which perform particular tasks or implement particular abstract data types. One or more embodiments of the description can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communication network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including memory storage devices.

[0146] The various embodiments described in this specification are intended to be exemplary only. The same features and aspects of the various embodiments can be combined in any combination. The various embodiments are described in the specification with reference to the following drawings, in which:

[0147] The above description is intended to be illustrative and not restrictive. Many other changes and modifications can occur to those skilled in the art once advised of the principles of the present document. Any and all such changes and modifications within the scope and spirit of the document are intended to be included.

Claims

1. A transaction risk detection method, comprising: receiving a risk analysis request for a target transaction object; in response to the risk analysis request, obtaining risk analysis data corresponding to the target transaction object; performing error identification processing on the risk analysis data using a pre-trained first analysis model, determining an error category to which the risk analysis data belongs, and obtaining a plurality of error sub-categories corresponding to the error category; performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model, wherein the parameter magnitude of the second analysis model is greater than the parameter magnitude of the first analysis model, to obtain diagnosis data corresponding to the risk analysis data; based on the diagnosis data, performing correction processing on the risk analysis data, and based on the corrected risk analysis data, determining whether the target transaction object has a transaction risk.

2. The method of claim 1, wherein the performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data comprises: obtaining business domain knowledge corresponding to a business domain to which the transaction business data belongs; performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories and the business domain knowledge using the pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data.

3. The method of claim 2, wherein the diagnosis data comprises one or more of an error sub-category to which the risk analysis data belongs, an error location, error interpretation data, error impact analysis data, and correction suggestion data.

4. The method of claim 1, wherein the obtaining risk analysis data corresponding to the target transaction object comprises: obtaining risk detection requirements corresponding to the target transaction object and transaction business data of the target transaction object; generating risk analysis data corresponding to the target transaction object based on the risk detection requirements and the transaction business data using a pre-set large language model.

5. The method of claim 4, further comprising: updating the pre-set large language model based on the corrected risk analysis data and the risk analysis data to obtain an updated large language model.

6. The method of claim 1, wherein the performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data comprises: obtaining target analysis data corresponding to each of the error sub-categories in historical risk analysis data; performing error diagnosis processing on the risk analysis data based on the plurality of error sub-categories and the target analysis data corresponding to each of the error sub-categories using the pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data.

7. The method of claim 6, wherein the error diagnosis processing of the risk analysis data based on the plurality of error sub-categories and the target analysis data corresponding to each of the error sub-categories using the pre-trained second analysis model to obtain the diagnosis data corresponding to the risk analysis data comprises: generating prompt information corresponding to each of the error sub-categories based on the target analysis data corresponding to the error sub-categories, the prompt information including an error reasoning process corresponding to the error sub-category; and performing error diagnosis processing of the risk analysis data based on the plurality of error sub-categories and the target analysis data and the prompt information corresponding to each of the error sub-categories using the pre-trained second analysis model to obtain the diagnosis data corresponding to the risk analysis data.

8. The method of claim 1, wherein the error categories include an accuracy error category, a professionalism error category, and a completeness error category, the accuracy error category includes one or more error sub-categories of factitious delusion, data reference error, and logical contradiction error, the professionalism error category includes one or more error sub-categories of term misuse, analysis depth deficiency, and conclusion support deficiency, and the completeness error category includes one or more error sub-categories of objectivity and neutrality, and readability and structure.

9. A transaction risk detection apparatus comprising: a request receiving module configured to receive a risk analysis request for a target transaction object; a data obtaining module configured to obtain risk analysis data corresponding to the target transaction object in response to the risk analysis request; a first analysis module configured to perform error identification processing of the risk analysis data using a pre-trained first analysis model, determine an error category to which the risk analysis data belongs, and obtain a plurality of error sub-categories corresponding to the error category; a second analysis module configured to perform error diagnosis processing of the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model; and a risk detection module configured to perform correction processing of the risk analysis data based on the diagnosis data, and determine whether the target transaction object has a transaction risk based on the corrected risk analysis data.

10. A transaction risk detection device comprising: a processor; and a memory arranged to store computer-executable instructions that, when executed, cause the processor to: receive a risk analysis request for a target transaction object; obtain risk analysis data corresponding to the target transaction object in response to the risk analysis request; perform error identification processing of the risk analysis data using a pre-trained first analysis model, determine an error category to which the risk analysis data belongs, and obtain a plurality of error sub-categories corresponding to the error category; and perform error diagnosis processing of the risk analysis data based on the plurality of error sub-categories using a pre-trained second analysis model to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model. ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ Perform error diagnosis processing on the risk analysis data based on the plurality of error subcategories by using a pre-trained second analysis model, to obtain diagnosis data corresponding to the risk analysis data, wherein a parameter magnitude of the second analysis model is greater than a parameter magnitude of the first analysis model; Perform correction processing on the risk analysis data based on the diagnosis data, and determine whether the target transaction object has a transaction risk based on the corrected risk analysis data.

Images