Data processing method, device and equipment

Abstract

The embodiment of the invention provides a data processing method, device and equipment, and the method comprises the steps: obtaining the project information of a historical project, and the risk information of the historical project; according to a large language model, based on the project information of the historical project, a first extraction result corresponding to the historical project is determined, and the first extraction result comprises summary information, risk factor information and project demand type information of the historical project; performing correction processing on the first extraction result according to a knowledge system corresponding to the project demand type information of the historical project and / or an association relationship between information in the first extraction result to obtain a corrected first extraction result; and according to the corrected first extraction result and the risk information of the historical project, constructing a knowledge base which is used for carrying out risk detection processing on a to-be-executed project.

Description

Technical Field

[0001] This document relates to the field of computer technology, and in particular to a data processing method, device and equipment. Background Art

[0002] As people pay more and more attention to their private data, in order to protect user privacy and ensure data security, it is necessary to conduct risk detection on pending projects to ensure data security during project execution. For example, a knowledge base can be built based on executed projects to conduct risk detection on pending projects based on the knowledge base.

[0003] In order to save data storage resources, feature extraction can be performed on the project information of executed projects, and a knowledge base can be constructed based on the extracted feature vectors. However, due to the increasing number of project types and the increasing complexity of project information, the accuracy of risk detection based only on the knowledge base constructed by feature vectors is low. To this end, the embodiments of this specification provide a better knowledge base construction solution for project risk detection. Summary of the invention

[0004] The purpose of the embodiments of this specification is to provide a better knowledge base construction solution for project risk detection.

[0005] In order to implement the above technical solution, the embodiments of this specification are implemented as follows: A data processing method provided in an embodiment of the present specification includes: obtaining project information of a historical project and risk information of the historical project; determining a first extraction result corresponding to the historical project based on the project information of the historical project according to a large language model, the first extraction result including summary information, risk factor information and project requirement type information of the historical project; correcting the first extraction result according to a knowledge system corresponding to the project requirement type information of the historical project and / or the correlation between each information in the first extraction result to obtain a corrected first extraction result; constructing a knowledge base according to the corrected first extraction result and the risk information of the historical project, the knowledge base being used to perform risk detection on the project to be executed.

[0006] A data processing device provided in an embodiment of the present specification comprises: a first acquisition module, used to acquire project information of a historical project and risk information of the historical project; a first determination module, used to determine a first extraction result corresponding to the historical project based on the project information of the historical project according to a large language model, wherein the first extraction result comprises summary information, risk factor information and project requirement type information of the historical project; a result correction module, used to correct the first extraction result according to a knowledge system corresponding to the project requirement type information of the historical project and / or the correlation between each information in the first extraction result, so as to obtain a corrected first extraction result; and a construction module, used to construct a knowledge base according to the corrected first extraction result and the risk information of the historical project, wherein the knowledge base is used to perform risk detection on a pending project.

[0007] A data processing device provided in an embodiment of the present specification comprises: a processor; and a memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to: obtain project information of a historical project, and risk information of the historical project; determine a first extraction result corresponding to the historical project based on the project information of the historical project according to a large language model, wherein the first extraction result comprises summary information, risk factor information, and project requirement type information of the historical project; perform correction processing on the first extraction result according to a knowledge system corresponding to the project requirement type information of the historical project, and / or the correlation between each information in the first extraction result, to obtain a corrected first extraction result; construct a knowledge base according to the corrected first extraction result and the risk information of the historical project, wherein the knowledge base is used to perform risk detection processing on the project to be executed.

[0008] The embodiments of the present specification also provide a storage medium, which is used to store computer executable instructions, and the executable instructions implement the following process when executed by a processor: obtaining project information of a historical project and risk information of the historical project; determining a first extraction result corresponding to the historical project based on the project information of the historical project according to a large language model, the first extraction result including summary information, risk factor information and project requirement type information of the historical project; correcting the first extraction result according to the knowledge system corresponding to the project requirement type information of the historical project and / or the correlation between each information in the first extraction result to obtain a corrected first extraction result; constructing a knowledge base according to the corrected first extraction result and the risk information of the historical project, the knowledge base being used to perform risk detection on the project to be executed.

[0009] The embodiments of the present specification also provide a computer program product, including a computer program, which implements the following process when executed by a processor: obtaining project information of a historical project and risk information of the historical project; determining a first extraction result corresponding to the historical project based on the project information of the historical project according to a large language model, the first extraction result including summary information, risk factor information and project requirement type information of the historical project; correcting the first extraction result according to a knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result to obtain a corrected first extraction result; constructing a knowledge base according to the corrected first extraction result and the risk information of the historical project, the knowledge base being used to perform risk detection processing on the project to be executed. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings required for use in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor. Figure 1 This is an embodiment of a data processing method of this specification; Figure 2 A schematic diagram of a data processing process of this specification; Figure 3 Another data processing method embodiment of this specification; Figure 4 Another data processing method embodiment of this specification; Figure 5 Another data processing method embodiment of this specification; Figure 6 Another data processing method embodiment of this specification; Figure 7 This is another data processing method embodiment of the present specification; Figure 8 This is another data processing method embodiment of the present specification; Fig. 9 A schematic diagram of a data processing process of this specification; Fig.10 This is an embodiment of a data processing device of the present specification; Fig.11 This is an embodiment of a data processing device in this specification. DETAILED DESCRIPTION

[0011] The embodiments of this specification provide a data processing method, device and equipment.

[0012] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0013] The embodiment of this specification provides a knowledge base construction mechanism for project risk detection. As people pay more and more attention to their own privacy data, in order to protect user privacy and ensure data security, it is necessary to perform risk detection on the pending projects to ensure data security during project execution. For example, a knowledge base can be constructed based on the executed projects to perform risk detection on the pending projects based on the knowledge base. In order to save data storage resources, feature extraction can be performed on the project information of the executed projects, and a knowledge base can be constructed based on the extracted feature vectors. However, due to the increasing number of project types and the increasing complexity of project information, the detection accuracy of risk detection based only on the knowledge base constructed by feature vectors is low. To this end, the embodiment of this specification provides a better knowledge base construction scheme for project risk detection. In this scheme, first, by using the computing power and understanding ability of a large language model, the first extraction result corresponding to the historical project can be quickly and accurately determined. Secondly, on the one hand, by correcting the first extraction result through the knowledge system corresponding to the project requirement type information of the historical project, the reasoning process of the large language model can be enhanced, and the accuracy and credibility of the knowledge base construction can be improved. On the other hand, by correcting the first extraction result through the association between the information in the first extraction result, the consistency and accuracy of the information in the corrected first extraction result can be improved. Then, based on the knowledge base constructed by the corrected first extraction result, the target to be executed can be subjected to risk detection processing to improve the accuracy of project risk detection. For specific processing, please refer to the specific content in the following embodiment.

[0014] like Figure 1 As shown, the embodiment of this specification provides a data processing method, the execution subject of the method can be a server, wherein the server can be an independent server, or a server cluster composed of multiple servers, etc. The server can be a background server for financial services or online shopping services, or a background server for an application, etc. In this embodiment, the execution subject is taken as an example for detailed description, and the method can specifically include the following steps: In step S102, project information of historical projects and risk information of historical projects are obtained.

[0015] Among them, the historical project can be any project that has been executed within the knowledge base creation cycle (or knowledge base update cycle). For example, the historical project can be a small program project that has been put into use in a certain application, etc. The project information of the historical project can include the demand information of the historical project. For example, the project information can include the project identification of the historical project, the project background information of the historical project department, and the specific content of the historical project, etc. The risk information of the historical project can include the risk type, risk severity and risk characteristics of the historical project.

[0016] In implementation, since the log data of risk review activities covers multiple aspects of structured and unstructured data, the server can determine historical projects based on the log data of risk review activities, and obtain project information and risk information of historical projects based on the log data.

[0017] For example, the log records for risk review activities may include the following data: 1) Requirement information: a)name: Record the identifying title of each risk review activity.

[0018] b) background: Provide detailed information about the project background, which can be used to understand the original intention of setting up the project.

[0019] c) description: Comprehensively record and explain the project content.

[0020] 2) Risk assessment content: a) risk_name: clearly indicate the type of risk identified.

[0021] b) risk_grade: reveals the severity of the assessed risk, categorized as high, medium, or low.

[0022] c) risk_desc: Detailed description of risk characteristics and possible impacts.

[0023] 3) Risk assessment recommendations: a) advice_type: describes the type of advice given.

[0024] b) advice_content: State the specific content of the advice.

[0025] The server can obtain the log record data of the existing risk review activities, and perform data preprocessing on the obtained log record data to determine the historical projects based on the preprocessed log record data, and obtain the project information and risk information of the historical projects based on the log record data.

[0026] Among them, data preprocessing can include standardizing the file format and removing noise data (such as removing HTML tags, extra spaces, and footer information in the text). In this way, the project information and risk information of historical projects can be determined through the log record data obtained through preprocessing, which can ensure the accuracy and efficiency of subsequent analysis.

[0027] In addition, the method for determining the project information and risk information of the above-mentioned historical projects is an optional and feasible determination method. There may be a variety of different determination methods in actual application scenarios. Different determination methods can be selected according to different actual application scenarios. The embodiments of this specification do not make specific limitations on this.

[0028] In step S104, a first extraction result corresponding to the historical item is determined according to the large language model and based on the item information of the historical item.

[0029] Among them, the first extraction result may include summary information, risk factor information and project requirement type information of historical projects. The summary information can be used to characterize key information of historical projects. The risk factor information can include factor information that may pose risks during project execution. Project requirement types may include internal requirement types, external requirement types, new requirement types that appear for the first time, persistent requirement types to be resolved, and regular requirement types that are reviewed periodically, etc. Large language models (LLMs) can be advanced natural language processing (NLP) models built on the basis of deep learning technology. Large language models have a large number of parameters and a wide range of training data sets, and have strong capabilities in language understanding and generation tasks. They can be effectively applied to text generation, machine translation, content summarization, question-answering systems and a variety of complex dialogue scenarios. Large language models can capture the richness and subtle differences of language expressions, and can display the captured information through powerful computing capabilities of billions to trillions of parameters.

[0030] In implementation, the server may determine the first extraction result corresponding to the historical item according to the large language model, based on a preset text prompt and item information of the historical item.

[0031] Taking the example of determining summary information of a historical project based on a large language model, a large language model can be used to extract and summarize key information and viewpoints in a long text. For example, a summary generation algorithm can be deployed in the large language model, and the large language model can extract important content from the text through the summary generation algorithm.

[0032] Taking a large language model built on the Transformer architecture and using a large-scale pre-trained dataset (such as a variety of data types including online text, academic resources, and code) as an example, the corresponding text prompt can be determined for the summary generation task. For example, the determined text prompt can be: "Document Summary Request: ------------------- Document title: {title} Document background: {background} Document description: {description} ------------------- Based on the above information, develop a concise summary of the document, no more than 300 words. The summary should cover the following key elements: - Product core needs and goals - Key operations that users can perform through the product - The potential impact of the product on user experience Please ensure that the information points extracted in the abstract are clear and accurate, and show in-depth understanding and comprehensive analysis as much as possible. " In this way, the large language model can be guided to focus on extracting and condensing the key content of the project information of the historical projects, while complying with the word limit and outputting summary information that is concise and contains all important information points. In this way, the efficiency of knowledge extraction can be improved through the above-mentioned refined prompting method.

[0033] The above is an example of determining the summary information of historical projects based on a large language model. In addition to this, the risk factor information and project requirement type information of historical projects can also be determined based on the large language model, based on text prompts corresponding to the risk factor information and text prompts corresponding to the project requirement type information.

[0034] For example, taking the risk factor information as action factor information, a task may be generated for the action factor information, and a corresponding text prompt may be determined. For example, the determined text prompt may be: "This is a document that details the product requirements. Title: "{title}". Here is a summary of the document and the analysis tasks that need to be performed: ------------------- Document title: {title} Document background: {background} Document description: {description} -------------------

Demand Overview

[0035] -------------------

Analysis Task

Logic Instructions

[0036] 2. If the document does not mention any key actions, your output should be: {{"reason": "No explanation for the action involved", "label": "No action"}}.

[0037] 3. During the evaluation process, please carefully consider the actual performer of each action, ensure that the identified actions are performed by the user group specified in the document, rather than developers or other participants, and ensure that the output action types are accurate. " In addition, based on the log data of risk review activities, a hierarchical approximation method can be used to construct a project requirement type system. For example, project requirement types can be divided into three major categories, and each major category can be subdivided into multiple subcategories. The project requirement type system constructed in this way can specifically display the unique attributes of each requirement and the background information it contains.

[0038] Specifically, project requirements can be divided into the following three categories, and each category contains subcategories: 1) Internal demand category, including internal platforms, technical transformation, defect repair, data algorithms, etc.

[0039] 2) External demand category, including types determined according to the specific source of demand, such as household, merchant capital flow, personal account opening, management, merchant entry, and application submission.

[0040] 3) New and old demand categories, including new demands that appear for the first time, pending demands that continue to exist and need to be resolved, and regular demands that are reviewed periodically.

[0041] In this way, by making a refined classification of project requirement types, we can ensure a complete interpretation of the requirements and improve the accuracy of subsequent risk detection.

[0042] In step S106, the first extraction result is corrected according to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result to obtain a corrected first extraction result.

[0043] Among them, the knowledge system can be the sum of knowledge corresponding to the project requirement type information obtained from a preset acquisition source. For example, the knowledge system can be the sum of knowledge obtained from sources such as industry standards, academic papers, and experience knowledge bases. The association relationship between each information in the first extraction result may include dependency relationships, collaborative relationships, and other relationships between the information.

[0044] During implementation, the server can correct the first extraction result according to a pre-trained correction model, based on the knowledge system corresponding to the project requirement type information of historical projects, and / or the correlation between the information in the first extraction result, to obtain a corrected first extraction result. The correction model can be a model built based on a machine learning algorithm.

[0045] There are many ways to correct the first extraction result. Different processing methods can be selected according to different actual application scenarios. This specification does not specifically limit this.

[0046] In step S108, a knowledge base is constructed based on the corrected first extraction result and the risk information of the historical projects.

[0047] Among them, the knowledge base can be used to perform risk detection and processing on the projects to be executed. For example, the knowledge base can be a risk knowledge base (RKB), that is, the knowledge base can refer to an information storage structure customized for the field of risk assessment and management, which aims to integrate data, institutional frameworks, case analysis and other resources related to business risk control to support rapid identification, accurate assessment and effective response to risks.

[0048] During implementation, in addition to the corrected first extraction results and risk information of historical projects, the server can also build a knowledge base based on review rules and expert knowledge. In this way, the knowledge base can be dynamically expanded according to actual application needs to achieve self-update and optimized layout of knowledge resources. The main task of knowledge base construction is to make a structured combing of all historical review records with risk needs.

[0049] The constructed knowledge base can have the following functions: 1) Accumulated professional knowledge: The knowledge base stores historical data and expert knowledge related to risk management. The corrected first extraction results and risk information of historical projects stored in the knowledge base can be used to perform risk detection on the projects to be executed. That is, for new risk review activities, the knowledge base can provide reference past cases; 2) Improve review speed: The intelligent review system can quickly compare the projects to be executed with historical projects by accessing the knowledge base to improve the risk detection and processing speed; 3) Ensure consistency: The knowledge base can ensure the consistency of review criteria and decision-making logic, and avoid the impact of human differences on review results; 4) Support complex decision-making: For complex decision-making problems, the knowledge base provides the necessary background information and reference standards to assist risk reviewers in making more accurate judgments.

[0050] The embodiment of this specification provides a data processing method, by obtaining the project information of the historical project and the risk information of the historical project, according to the large language model, based on the project information of the historical project, determining the first extraction result corresponding to the historical project, the first extraction result includes the summary information of the historical project, the risk factor information and the project requirement type information, according to the knowledge system corresponding to the project requirement type information of the historical project, and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain the corrected first extraction result, according to the corrected first extraction result and the risk information of the historical project, a knowledge base is constructed, and the knowledge base is used to perform risk detection processing on the project to be executed, so that, firstly, by using the computing power and understanding ability of the large language model, the first extraction result corresponding to the historical project can be quickly and accurately determined. Secondly, on the one hand, by correcting the first extraction result through the knowledge system corresponding to the project requirement type information of the historical project, the reasoning process of the large language model can be enhanced, and the accuracy and credibility of the knowledge base construction can be improved. On the other hand, by correcting the first extraction result through the association relationship between each information in the first extraction result, the consistency and accuracy of each information in the corrected first extraction result can be improved. Furthermore, risk detection processing can be performed on the target to be executed based on the knowledge base constructed by the corrected first extraction result, thereby improving the accuracy of project risk detection.

[0051] In practical applications, risk factor information can include product factor information, subject factor information, platform factor information, and action factor information. The above four risk factor information are basic and key components in the project execution process. Therefore, these four risk factor information can be used to jointly define the main components of a project, providing a structured perspective for understanding and analyzing project risks.

[0052] These risk factors enable risk managers to analyze specifically which specific products the project involves, what actions are performed, which entities are involved, and on which platform. Any business activity can be described and understood through these four risk factor information, and comprehensive coverage of risk points can be achieved through these four risk factor information.

[0053] In this way, through risk factor information, potential risk points in the project execution process can be more effectively identified, quantified and mitigated, providing a clear and unified language for complex business operations and ensuring the integrity and accuracy of the risk review process. Through the detailed description of the four risk factor information of products, actions, subjects and platforms, the specific scenarios of risk occurrence can be intuitively presented, thus providing a solid foundation for the formulation of effective risk management strategies.

[0054] In practical applications, after step S108, risk detection processing can be performed based on the constructed knowledge base. The specific processing methods of risk detection processing can be various. The following is an optional processing method, such as Figure 2 As shown, the processing may specifically include the following steps S202 to S210.

[0055] In step S202, a risk detection request for a target project to be executed is received.

[0056] In step S204, in response to the risk detection request, project information of the target project is obtained.

[0057] During implementation, the server can obtain the Project Requirements Document (PRD) of the target project. PRD can be used to describe the project to be executed. The PRD can record the requirement information of the target project, including the requirement name, background, goal, scope, characteristics and detailed description of the target project. The server can determine the project information of the target project based on the PRD of the target project.

[0058] Among them, the PRD requirements may include the industry type, activity type, operational details involved (for example, sensitive information such as amount, user, commodity, card, equipment, etc.), potential risks and other information corresponding to the target project. This information can be used to conduct an accurate risk assessment of the execution of the target project.

[0059] In step S206, a second extraction result corresponding to the target project is determined according to the large language model and based on the project information of the target project.

[0060] The second extraction result may include a summary of the target project, summary information, risk factor information, and project requirement type information.

[0061] In implementation, the processing method of the second extraction result corresponding to the target project can refer to the processing method of the first extraction result corresponding to the historical project, which will not be repeated here.

[0062] In step S208, a first historical item corresponding to the target item among the historical items included in the knowledge base is determined according to the second extraction result and the corrected first extraction result corresponding to the historical item.

[0063] In implementation, in the above step S208, based on the second extraction result and the corrected first extraction result corresponding to the historical item, the specific processing method of determining the first historical item corresponding to the target item among the historical items included in the knowledge base can be varied. An optional processing method is provided below, which can specifically include the processing of the following steps A1 and A2.

[0064] In step A1, the similarity between the target item and each historical item is determined according to a preset similarity algorithm based on the second extraction result and the corrected first extraction result corresponding to the historical item.

[0065] The preset similarity algorithm may include a Euclidean distance algorithm, a Jaccard similarity algorithm, a cosine algorithm, and the like.

[0066] In implementation, the server performs feature extraction processing on the second extraction result and the corrected first extraction result according to the pre-trained feature extraction model, respectively, to obtain a first embedding vector corresponding to the second extraction result, and a second embedding vector corresponding to the corrected first extraction result. Then, the server can determine the similarity between the first embedding vector and the second embedding vector according to a preset similarity algorithm, and determine the similarity as the similarity between the target item and each historical item.

[0067] In this way, deep learning technology can be used to convert text content into vector representation, and then combined with a preset similarity algorithm, accurate calculation of the similarity between projects can be achieved.

[0068] Among them, the feature extraction model can be a model built based on a preset deep learning algorithm for converting text content into vector representation.

[0069] In step A2, a first historical item in the historical items corresponding to the target item is determined according to the similarity between the target item and each historical item.

[0070] In implementation, the server may determine the history item whose similarity is greater than a preset similarity threshold as the first history item.

[0071] In addition, the server can also obtain a second historical project in the historical project that matches the project requirement type information of the target project based on the project requirement type information of the target project and the project department requirement type information of the historical project in the knowledge base. Then, the server can determine the similarity between the target project and each second historical project based on the second extraction result and the corrected first extraction result corresponding to the second historical project according to a preset similarity algorithm. Finally, the server can determine the first historical project corresponding to the target project in the second historical project based on the similarity between the target project and each second historical project.

[0072] In step S210, it is determined whether there is a risk in executing the target project according to the risk information of the first historical project.

[0073] In implementation, the server may determine whether there is a risk in executing the target project based on the risk type, risk severity, and risk characteristics in the risk information of the first historical project.

[0074] In practical applications, there may be various specific processing methods for determining whether there is a risk in executing the target project based on the risk information of the first historical project in the above step S210. An optional processing method is provided below, which may specifically include the following steps B1 to B4.

[0075] In step B1, first risk information corresponding to the target project is determined according to the risk information of the first historical project.

[0076] In step B2, a risk detection model corresponding to the target project is determined according to the business detection rules corresponding to the target project.

[0077] Among them, the risk detection model can be a model built based on a preset machine learning algorithm.

[0078] During implementation, the server can build different risk detection models based on different business detection rules. For example, for business detection rules with simpler rules, the server can build a risk detection model with a simpler structure (such as a risk detection model containing 3 convolutional layers); for business detection rules with more complex rules, the server can build a risk detection model with a more complex structure (such as a risk detection model containing 10 convolutional layers), etc.

[0079] In step B3, second risk information corresponding to the target project is determined according to the risk detection model based on the second extraction result corresponding to the target project.

[0080] In implementation, by loading pre-defined business detection rules and corresponding risk detection models, combined with the refined second extraction results corresponding to the target project, the business process of the target project can be accurately evaluated to ensure that the decision logic is consistent with the business detection requirements of the target project.

[0081] In step B4, it is determined whether there is a risk in executing the target project based on the first risk information and the second risk information.

[0082] In practical applications, before step S104, the large language model may be trained. The specific processing methods of the training may be various. An optional processing method is provided below, such as Figure 3 As shown, the processing may specifically include the following steps S302 to S308.

[0083] In step S302, a pre-trained large language model is obtained.

[0084] Among them, the pre-trained large language model can be obtained by preliminary training of the large language model based on a large amount of sample data. The pre-trained large language model has powerful processing capabilities in language understanding and generation tasks.

[0085] In step S304, historical sample data and a third extraction result corresponding to the historical sample data are obtained.

[0086] The third extraction result may include summary information of the historical sample data, risk factor information, and project requirement type information.

[0087] In step S306, according to the pre-trained large language model, based on the preset prompt information and the historical sample data, a fourth extraction result corresponding to the historical sample data is obtained.

[0088] In step S308, the pre-trained large language model is fine-tuned according to the third extraction result and the fourth extraction result to obtain a trained large language model.

[0089] In implementation, the server can fine-tune the pre-trained large language model through parameter-efficient fine-tuning (PEFT) technology (such as QLoRA) when it is determined that the pre-trained large language model has not converged. In this way, through fine-tuning, the memory requirement of the model parameters can be reduced and the efficiency of fine-tuning can be maintained.

[0090] In practical applications, in step S104, there are various specific processing methods for determining the first extraction result corresponding to the historical project based on the large language model and the project information of the historical project. An optional processing method is provided below, such as Figure 4 As shown, the processing may specifically include the following steps S1042 to S1048.

[0091] In step S1042, according to the detection requirements corresponding to the historical items, the first prompt information corresponding to the historical items is determined, and the first prompt information is extended and enhanced to obtain the target prompt information.

[0092] In implementation, the server may construct a new thought chain prompt (ie, target prompt information) based on the first prompt information, so as to encourage the large language model to explore the possibility of different extraction results contained in the project information through the target prompt information.

[0093] Taking the identification of action element information in risk element information as an example, the first prompt information can be expanded and enhanced through a user experience-oriented action deduction method to obtain target prompt information. For example, the target prompt information can be "Imagine a typical user using this product in daily life. Consider the value of the product to them and how it may improve their lifestyle. Based on these considerations, explore what actions users may perform to achieve their goals and how these actions are reflected in the requirements document."

[0094] Alternatively, the first prompt information can be expanded and enhanced by summarizing actions through scenario analysis to obtain target prompt information. For example, the target prompt information can be "Imagine the interaction that users may have with the product in a specific usage scenario, such as travel, work, or family life. Try to point out which new or changed needs may cause users to take new actions based on such scenarios, and describe in detail how these actions may affect the relationship between users and products."

[0095] In addition, the server can also perform expansion and enhancement processing on the first prompt information through a pre-trained enhancement model to obtain target prompt information, wherein the enhancement model can be a model built based on a preset machine learning algorithm.

[0096] In addition, there are many methods for determining the target prompt information, and different determination methods can be selected according to different actual application scenarios. The embodiments of this specification do not make specific limitations on this.

[0097] In step S1044, a plurality of different candidate extraction results are obtained according to the large language model, based on the target prompt information, and the project information of the historical project.

[0098] In implementation, the server can repeat the above steps multiple times using different target prompt information to iteratively generate a series of candidate extraction results. This iterative process can enhance the multi-angle thinking ability of the large language model and explore different logical paths.

[0099] In step S1046, logical analysis is performed on the candidate extraction results according to the reasoning paths corresponding to the candidate extraction results to obtain a logical analysis result for each candidate extraction result.

[0100] In implementation, the server can perform a logical analysis on the candidate extraction results to obtain a logical analysis result for each candidate extraction result, that is, the server can use the large language model to perform a detailed analysis on logically conflicting or contradictory answers to determine which reasoning paths are feasible and which reasoning paths are logically unreasonable. For example, if one reasoning path suggests that the user will take a certain operation to solve a problem, while another reasoning path suggests that the problem does not exist at all, then the large language model needs to further judge the two conflicting reasoning paths to confirm which reasoning path is more in line with the actual situation.

[0101] In step S1048, according to the occurrence frequency of each candidate extraction result and the result of the logic analysis, a first extraction result corresponding to the historical item is selected from the candidate extraction results.

[0102] In implementation, the server may integrate the candidate extraction results and perform logic conflict analysis to screen out the first extraction result.

[0103] For example, the server may integrate and analyze the generated multiple candidate extraction results to determine the candidate extraction result with the highest frequency of occurrence, and then determine whether the candidate extraction result can be determined as the first extraction result based on the logical analysis result of the candidate extraction result.

[0104] In addition, the server can also construct a first extraction result that meets the requirements according to the preset result output specification. For example, the first extraction result can be {{"reason": "After multiple iterations of reasoning, it is found that {action} is the action that the user is most likely to perform. Although there are some logical contradictions, after analysis, the correspondence between this action and user needs and product functions is the most reasonable.", "label": "{action_type}"}}, where {action} can be action element information, and {action_type} can be the category to which the action element information belongs.

[0105] In this way, the server can further enhance the quality of the output data of the large language model through the reflective capacity of large language models (RCLLM), ensuring the logical coherence and information accuracy in continuous text generation tasks.

[0106] Among them, the reflective ability of the large language model refers to an advanced function of self-evaluation and correction of the large language model. Through this reflective ability, the large language model can not only generate text, but also conduct in-depth reflection with reference to historical generated content to identify and correct potential deviations or errors in the output text.

[0107] The server can use the self-consistency in the reflection ability of the large language model. When reasoning, it does not rely on a single solution path. Instead, it generates multiple reasoning paths and answers through the large language model, and then uses majority voting to select the final answer. Self-consistency refers to a method that can improve the reasoning ability of the large language model. Self-consistency is based on an assumption: every complex problem can be answered through multiple ideas.

[0108] In specific implementation, it can be achieved through the following steps: 1. Use Chain of Thought (CoT) prompts to stimulate the large language model and enable it to perform deep reasoning.

[0109] 2. Generate multiple answers from the language model decoder using random sampling instead of a single greedy decoding.

[0110] 3. Among all the generated answers, the answer with the highest frequency is selected as the final result, that is, the aggregated answer obtained through self-consistency.

[0111] The self-consistency method can effectively assist in identifying and summarizing various types of information in the project information. By stimulating the language model multiple times, different outputs (i.e., candidate extraction results) can be generated. The sub-consistency method can then screen out the first extraction result that meets the review logic from the candidate extraction results.

[0112] In practical applications, in step S104, there are various specific processing methods for determining the first extraction result corresponding to the historical project based on the large language model and the project information of the historical project. An optional processing method is provided below, such as Figure 5 As shown, the processing may specifically include the following steps S10410 to S10416.

[0113] In step S10410, first prompt information corresponding to the historical item is obtained according to the detection requirement corresponding to the historical item.

[0114] In step S10412, a plurality of different candidate extraction results are obtained according to the large language model based on the first prompt information, the project information of the historical project, and a plurality of different sampling parameters.

[0115] In implementation, random sampling can be used to obtain multiple different candidate extraction results. For example, when performing sampling, the random sampling parameters of the decoder (such as the temperature coefficient temperature) can be adjusted to produce variations in the reasoning process. Different temperature values ​​can lead to diversity in text generation, thereby providing candidate extraction results from different angles.

[0116] The server can repeat the above steps multiple times using different sampling parameters to iteratively generate a series of candidate extraction results. This iterative process can enhance the multi-angle thinking ability of the large language model and explore different logical paths.

[0117] In step S10414, logical analysis is performed on the candidate extraction results according to the reasoning paths corresponding to the candidate extraction results to obtain a logical analysis result for each candidate extraction result.

[0118] In step S10416, according to the occurrence frequency of each candidate extraction result and the result of the logic analysis, a first extraction result corresponding to the historical item is selected from the candidate extraction results.

[0119] In implementation, the processing procedures of the above steps S10414 to S10416 can refer to the above processing procedures and will not be repeated here.

[0120] In addition, the server can also obtain multiple different candidate extraction results based on the large language model, based on the target prompt information, the project information of the historical project, and multiple different sampling parameters, and then select the first extraction result corresponding to the historical project from the candidate extraction results based on the frequency of occurrence of each candidate extraction result and the logical analysis result.

[0121] In practical applications, in step S106, the first extraction result is corrected according to the knowledge system corresponding to the project requirement type information of the historical project. The specific processing method for obtaining the corrected first extraction result can be various. The following is an optional processing method, such as Figure 6 As shown, the processing may specifically include the following steps S1062 to S1064.

[0122] In step S1062, a match check is performed between the knowledge system and the first extraction result.

[0123] In implementation, the server can build an initial knowledge base (which can be recorded as Knowledge Base-Init) based on the first extraction result determined in the above steps. Then, the server can use the knowledge system and the reflective ability of the large language model to make self-corrections to correct the initial knowledge base and improve the accuracy of the knowledge base.

[0124] For example, the server may utilize the knowledge-based self-correction in the reflection capability of the large language model to perform correction processing on the first extraction result to obtain a corrected first extraction result.

[0125] The server can pre-set a knowledge system for risk factor information and project requirement type information, and maintain cross-task consistency for risk factor information and project requirement type information based on the knowledge system. For example, taking project requirement type information as an example, the following knowledge system can be pre-configured for project requirement type information: a) Direct beneficiary rule of demand: If the direct beneficiary of the demand is an external entity such as an end user or a cooperative merchant, it belongs to the external demand type; if the direct beneficiary is an internal department or employee of the enterprise, it belongs to the internal demand type. For example, a user statement project can be an external demand type, and a technology upgrade project can be an internal demand type.

[0126] b) Demand triggering factor rule: If the triggering factor of the demand is market changes, user feedback, new demands of partners, or updates to laws and regulations, then this demand can be an external demand type; if the triggering factor is internal audit, internal efficiency assessment, technical debt, or problems found by internal quality control, then this demand can be an internal demand type. New demands are often triggered by external driving factors, while the iteration of old demands may be triggered by both internal and external factors.

[0127] c) Degree of innovation and iterative rules of demand: New products, new business models, and new functions usually indicate a high degree of innovation, so they can be classified as new demand types, while the expansion of existing products and functions, rule adjustments, experience optimization, etc. are usually iterative improvement needs, so they can be classified as old demand types. Both internal and external needs can be new or iterative needs of old needs. The key is whether they are introducing new concepts and technologies or optimizing existing frameworks.

[0128] In step S1064, when the knowledge system and the first extraction result do not match, the first extraction result is corrected according to the large language model based on the knowledge system and the project information of the historical project to obtain a corrected first extraction result.

[0129] In implementation, the server may perform correction processing on the first extraction result in the initial knowledge base based on the knowledge system through the following operations to obtain a corrected first extraction result: 1. Knowledge integration: Extract relevant knowledge from knowledge systems (such as industry standards, academic papers, and empirical knowledge bases) and integrate it into the reasoning framework of the large language model as background knowledge for analysis tasks.

[0130] 2. Content alignment: Analyze the output generated by the large language model to check whether it is consistent with the integrated knowledge system in terms of content. By comparing the predefined knowledge items with the information output by the large language model, it can ensure that the reasoning results are consistent with external known facts and logic.

[0131] 3. Knowledge-driven adjustment: If it is determined that the content output by the large language model (i.e., the first extraction result) is inconsistent with the integrated knowledge system, a knowledge-driven adjustment process can be executed, that is, for each deviation point, appropriate argumentation or deduction steps are introduced to correct the reasoning path of the large language model.

[0132] 4. Knowledge verification and validation: Verify the output after knowledge correction (i.e., the corrected first extraction result) to ensure that the changes brought about by the correction process have indeed increased the quality of the output and are aligned with the knowledge system of the knowledge system.

[0133] In practical applications, in step S106, the first extraction result is corrected according to the correlation between the information in the first extraction result. The specific processing method for obtaining the corrected first extraction result can be various. The following is an optional processing method, such as Figure 7 As shown, the processing may specifically include the following steps S1066 to S1068.

[0134] In step S1066, a sampling strategy and second prompt information corresponding to the historical item are determined according to the association relationship between each information in the first extraction result.

[0135] During implementation, the server may also utilize cross-task consistency in the large language reflection capability to correct the first extraction result to obtain a corrected first extraction result.

[0136] In step S1068, the first extraction result is corrected according to the large language model based on the sampling strategy, the second prompt information, and the project information of the historical project to obtain a corrected first extraction result.

[0137] Since there is often an association between tasks (i.e., the generation tasks of each information in the first extraction result), there may be logical conflicts between the results output by the large language model, thus affecting the consistency of the entire system. Therefore, the cross-task system consistency can be used to correct the output results of the large language model. The specific steps are as follows: a) Construction of cross-task thought chain: Based on the thought chain (CoT) of related tasks, a comprehensive prompt (i.e., second prompt information) containing multiple tasks can be created to reflect the interrelationship and dependency between different tasks based on the second prompt information. For each task, it is necessary not only to guide the large language model to explain the possible actions under the current task, but also to guide the large language model to consider the relationship between the current task and other tasks.

[0138] b) Multi-task random sampling coordination: In the random sampling process, the sampling strategies of different tasks are coordinated to prevent the randomness of one task from affecting the output quality of another task. This requires the establishment of an inter-task coordination mechanism in the sampling process to produce a coordinated and consistent reasoning path.

[0139] c) Repeated cross-task iteration: Perform multiple rounds of task output iterations to enhance the reasoning quality of each task while ensuring that different tasks can support and verify each other. During the collaborative iteration process, special attention is paid to the results that may conflict with the information obtained in one task and other tasks.

[0140] d) Comprehensive results integration and conflict identification: Combine the outputs of all tasks to build a cross-task result set. For logical conflicts or inconsistencies between tasks, conduct centralized analysis, such as using contextual and task-related logical frameworks to identify the source of the conflict.

[0141] e) Final normalization of cross-task outputs: Create a unified output format for the results across tasks, including JSON or other structures that are easy to interact with and interpret.

[0142] In practical applications, in step S106, the first extraction result is corrected according to the knowledge system corresponding to the project requirement type information of the historical project and the association relationship between the information in the first extraction result. The specific processing method for obtaining the corrected first extraction result can be various. The following is an optional processing method, such as Figure 8 As shown, the processing may specifically include the following steps S10610~S10614.

[0143] In step S10610, a sampling strategy and second prompt information corresponding to the historical item are determined according to the association relationship between each information in the first extraction result.

[0144] In step S10612, a match check is performed between the knowledge system and the first extraction result.

[0145] In step S10614, when the knowledge system and the first extraction result do not match, the first extraction result is corrected according to the large language model based on the sampling strategy, the second prompt information, the knowledge system and the project information of the historical project to obtain a corrected first extraction result.

[0146] The processing process of the above steps S10610~S10614 can refer to the above processing process and will not be repeated here.

[0147] The server can build an intelligent review system based on a large language model combined with agent technology. Among them, agent technology (AI Agent Technology) is a composite technology that uses a series of artificial intelligence algorithms to simulate the actions of intelligent subjects. Artificial intelligence algorithms can give agents the ability to perceive the environment, perform logical reasoning and perform complex tasks.

[0148] Among them, by utilizing the powerful computing and understanding capabilities of large language models, a large number of knowledge base construction tasks can be completed quickly, providing accurate risk identification and assessment-related knowledge. Agent technology can continuously iterate its own knowledge and understanding, thereby reducing dependence on large-scale manual labeling work. In this way, the intelligent review system built based on large language models and agent technology has the ability to self-learn and adapt to changes, and can update its knowledge base in an evolving business environment to achieve long-term and effective risk control. In addition, the intelligent review system can also generate explainable reasoning paths and decision logic to improve the transparency of the knowledge base construction process.

[0149] In this way, during the knowledge base construction process, the intelligent agent can assume the role of coordinating different computing modules within the system, effectively integrating and optimizing the process of knowledge base construction, maintenance and real-time updating, so as to conduct Risk Intelligent Review (RIR) through the constructed knowledge base.

[0150] Among them, risk intelligence review refers to the process of evaluating potential security risks in business proposals through an automated platform.

[0151] The intelligent review system based on large language models combined with intelligent agent technology can overcome the limitations of traditional manual review and machine learning methods, and achieve efficient and accurate review of large-scale data. In addition, the application of large language models, the combination of retrieval-augmented generation (RAG) and deep learning technology, and the integration of intelligent agent technology can improve the scalability, timeliness, and generalization of the risk review process, and reduce maintenance costs.

[0152] like Fig. 9 As shown, when conducting risk review, the intelligent agent can be initialized first, which can specifically include starting the intelligent agent of the intelligent review system and configuring corresponding computing resources and permissions for the intelligent agent. Since the large language model has a large scale and contains more parameters, it has higher requirements for hardware resources.

[0153] For example, the hardware model for running large language models can be NVIDIA A100 or V100. These GPUs can be used for high-performance computing (HPC) and deep learning training tasks, and have a large amount of parallel processing capabilities. Among them, the A100 GPU can provide 40GB or 80GB HBM2 video memory, and the V100 can provide 16GB or 32GB HBM2 video memory. Large language models may require multiple such GPUs to work in parallel during training and application.

[0154] In this way, by initializing and configuring high-performance GPU resources, a large language model with large parameters (such as 7.2 billion) can be started. This technical architecture can provide powerful computing power and efficient data processing capabilities for handling complex risk review tasks. Then, the risk review system can define receiving parameters and introduce key input materials such as product requirement documents (PRDs) as the basic information for risk review.

[0155] After the initialization of the agent is completed, the input material can be preprocessed to obtain the project information and risk information of the historical project, and then the content can be extracted and summarized according to the large language model, and the risk factor information and the demand type information can be initially screened to obtain the first extraction result. For example, the summary generation algorithm in the large language model can be used to extract and summarize the core ideas and information (i.e., obtain summary information) from the long text provided. Among them, the agent can combine the open source LM framework and customized logic to accelerate the processing and analysis of text content.

[0156] After obtaining the first extraction result, the intelligent agent can build an initial knowledge base based on the first extraction result, and then make self-corrections through the reflection ability of the large language model to improve the accuracy of the knowledge base.

[0157] Through the self-consistency, knowledge-based self-correction, and cross-task consistency of the reflective ability of the large language model, the automation, accuracy, and real-time updating capabilities of risk knowledge summary can be improved, which can effectively improve the efficiency and quality of knowledge base construction.

[0158] By leveraging the reflective capabilities of large language models combined with intelligent agent technology, a closed-loop feedback mechanism can be formed for self-correction. In addition, cross-task coordination can ensure the consistency and accuracy of the intelligent review system, allowing the intelligent review system to process large-scale data while ensuring the reliability and timely updating of the results.

[0159] Among them, through self-consistency, multiple reasoning paths can be automatically generated and the most consistent answer can be selected, which no longer relies on manual input from experts, improving scalability and efficiency. Through self-correction based on the knowledge system, combined with cross-task consistency, new information can be continuously monitored and integrated to ensure that the knowledge base can be updated synchronously with the latest business practices.

[0160] After completing the initial self-correction, the loop feedback process can be started. During this process, the correction process can be iterated in a loop to continuously improve the accuracy and coverage of the knowledge base. The correction process is repeated until the change in the knowledge base results is minimized, or reaches a predetermined optimization threshold, and the corrected first extraction result is obtained, and the knowledge base is constructed based on the corrected first extraction result.

[0161] Through closed-loop information feedback, a self-perfecting information feedback mechanism can be established to ensure that the knowledge base can keep pace with the times. The results of each round of iterative correction can be recorded in detail and incorporated into the continued training of the large language model. In this way, the large language model can continue to learn with new data points to improve its reasoning performance and classification accuracy when facing similar scenarios. In addition, this process also provides a basis for the integration of new knowledge, ensuring that the knowledge base can flexibly adapt to changes in the external environment.

[0162] Finally, the knowledge base results can be archived. The results of the knowledge base built by the risk review agent need to be archived to ensure the continuous update and implementation of the policy. The archived content includes but is not limited to the decision logic, revision history and final reasoning results. The execution task of the agent will be completed in this link, and then the cache data can be cleaned up to release execution resources, and the knowledge base itself will enter the daily update and maintenance stage.

[0163] The knowledge base constructed can be incorporated into a wider decision support system and become a powerful tool for risk assessment and management. Based on the knowledge base, continuous risk detection of pending projects and regular evaluation based on actual application feedback can continuously improve the knowledge base and provide a more reliable reference for future decision making.

[0164] After building the knowledge base, the target project to be executed can be tested for risks. The intelligent review system can search for similar cases for the target project through the knowledge base, that is, the intelligent review system can use the efficient database management system and full-text search engine to match the historical projects and target projects in the knowledge base, which can enhance the accuracy and reference value of the review. In addition, the text content to be matched can be converted into a vector representation through deep learning technology, and combined with the Jaccard similarity algorithm, the accurate calculation of the similarity between the historical project and the target project can be achieved.

[0165] In addition, the intelligent review system can also obtain the risk detection model corresponding to the pre-defined business rules through the rule engine link, and combine the extracted information to accurately evaluate the business process of the target project to ensure that the decision logic is consistent with current business needs and regulations.

[0166] In addition, the intelligent review system can also use manual review to verify the risk detection results of the intelligent agent through manual sampling and alignment stages, thereby ensuring the accuracy of the risk detection results.

[0167] Finally, in the knowledge sedimentation stage, the intelligent review system can transform the review results and experience into retrievable structured knowledge and integrate it into the knowledge base, providing richer reference information and decision support for future reviews.

[0168] In summary, by combining large-scale language models, intelligent agent technology, deep learning and other advanced technologies, we can build a comprehensive, efficient and accurate intelligent review system. This intelligent review system can not only handle a large number of review requests, but also provide accurate risk identification and detection, meeting the needs of high-risk industries for efficiency, accuracy and scalability of the review process.

[0169] Among them, the process of building the knowledge base can be effectively managed by using intelligent agents to implement system initialization, data preprocessing, content summary generation, large language model reflection, and loop iteration until convergence.

[0170] By using the reflective capabilities of large models and intelligent agent technology to build a knowledge base, the accuracy, adaptability and timeliness of the knowledge base can be improved.

[0171] Among them, accuracy can refer to: the reflective ability of the large language model allows the intelligent review system to self-evaluate and correct the outputs it generates to avoid errors, while continuously integrating new information through deep learning technology to improve the understanding of risk assessment logic.

[0172] Adaptability means that by executing automated processes such as data collection and analysis, and coordinating update tasks through intelligent agent technology, the consistency and real-time nature of the content in the knowledge base can be ensured, and scattered information can be integrated into a unified knowledge system.

[0173] Timeliness means that through the flexibility and multi-tasking capabilities of intelligent agents, the risk assessment process can be optimized at each stage, reducing manpower requirements while improving the quality and speed of review.

[0174] The embodiment of this specification provides a data processing method, by obtaining the project information of the historical project and the risk information of the historical project, according to the large language model, based on the project information of the historical project, determining the first extraction result corresponding to the historical project, the first extraction result includes the summary information of the historical project, the risk factor information and the project requirement type information, according to the knowledge system corresponding to the project requirement type information of the historical project, and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain the corrected first extraction result, according to the corrected first extraction result and the risk information of the historical project, a knowledge base is constructed, and the knowledge base is used to perform risk detection processing on the project to be executed, so that, firstly, by using the computing power and understanding ability of the large language model, the first extraction result corresponding to the historical project can be quickly and accurately determined. Secondly, on the one hand, by correcting the first extraction result through the knowledge system corresponding to the project requirement type information of the historical project, the reasoning process of the large language model can be enhanced, and the accuracy and credibility of the knowledge base construction can be improved. On the other hand, by correcting the first extraction result through the association relationship between each information in the first extraction result, the consistency and accuracy of each information in the corrected first extraction result can be improved. Furthermore, risk detection processing can be performed on the target to be executed based on the knowledge base constructed by the corrected first extraction result, thereby improving the accuracy of project risk detection.

[0175] The above is a data processing method provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a data processing device, such as Fig.10 shown.

[0176] The data processing device comprises: a first acquisition module 1001, a first determination module 1002, a result correction module 1003 and a construction module 1004, wherein: The first acquisition module 1001 is used to acquire project information of historical projects and risk information of the historical projects; A first determination module 1002 is used to determine a first extraction result corresponding to the historical project based on the project information of the historical project according to the large language model, wherein the first extraction result includes summary information, risk factor information, and project requirement type information of the historical project; A result correction module 1003, configured to correct the first extraction result according to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result, to obtain a corrected first extraction result; The construction module 1004 is used to construct a knowledge base according to the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed.

[0177] In the embodiment of this specification, the risk factor information includes product factor information, subject factor information, platform factor information and action factor information.

[0178] In the embodiment of this specification, the device further includes: A request receiving module, used to receive a risk detection request for a target project to be executed; A second acquisition module, configured to acquire project information of the target project in response to the risk detection request; A second determination module, configured to determine a second extraction result corresponding to the target project based on the project information of the target project according to the large language model; A project determination module, configured to determine, according to the second extraction result and the corrected first extraction result corresponding to the historical project, a first historical project among the historical projects included in the knowledge base that corresponds to the target project; The risk detection module is used to determine whether there is a risk in executing the target project based on the risk information of the first historical project.

[0179] In the embodiment of this specification, the project determination module is used to: Determine the similarity between the target project and each of the historical projects based on the second extraction result and the corrected first extraction result corresponding to the historical project according to a preset similarity algorithm; According to the similarity between the target item and each of the historical items, a first historical item among the historical items corresponding to the target item is determined.

[0180] In the embodiment of this specification, the device includes: Model acquisition module, used to obtain pre-trained large language models; A third acquisition module, used to acquire historical sample data and a third extraction result corresponding to the historical sample data; A third determination module, configured to obtain a fourth extraction result corresponding to the historical sample data according to the pre-trained large language model, based on preset prompt information, and the historical sample data; A model training module is used to pre-train a large language model based on the third extraction result and the fourth extraction result.

[0181] In the embodiment of this specification, the first determining module 1002 is used to: Determine, according to the detection requirement corresponding to the historical item, first prompt information corresponding to the historical item, and perform expansion and enhancement processing on the first prompt information to obtain target prompt information; Obtaining a plurality of different candidate extraction results based on the large language model, the target prompt information, and the project information of the historical project; According to the reasoning path corresponding to the candidate extraction result, the candidate extraction result is logically analyzed to obtain a logical analysis result for each candidate extraction result; According to the occurrence frequency of each of the candidate extraction results and the logic analysis result, a first extraction result corresponding to the historical item is selected from the candidate extraction results.

[0182] In the embodiment of this specification, the first determining module 1002 is used to: According to the detection requirement corresponding to the historical item, obtaining first prompt information corresponding to the historical item; Obtaining a plurality of different candidate extraction results according to the large language model, based on the first prompt information, the project information of the historical project, and a plurality of different sampling parameters; According to the reasoning path corresponding to the candidate extraction result, the candidate extraction result is logically analyzed to obtain a logical analysis result for each candidate extraction result; According to the occurrence frequency of each of the candidate extraction results and the logic analysis result, a first extraction result corresponding to the historical item is selected from the candidate extraction results.

[0183] In the embodiment of this specification, the result correction module 1003 is used to: Performing a match detection on the knowledge system and the first extraction result; In the case that the knowledge system and the first extraction result do not match, the first extraction result is corrected according to the large language model based on the knowledge system and the project information of the historical project to obtain the corrected first extraction result.

[0184] In the embodiment of this specification, the result correction module 1003 is used to: Determine a sampling strategy and second prompt information corresponding to the historical item according to the association relationship between each information in the first extraction result; According to the large language model, based on the sampling strategy, the second prompt information, and the project information of the historical project, the first extraction result is corrected to obtain the corrected first extraction result.

[0185] In the embodiment of this specification, the result correction module 1003 is used to: Determine a sampling strategy and second prompt information corresponding to the historical item according to the association relationship between each information in the first extraction result; Performing a match detection on the knowledge system and the first extraction result; In the case that the knowledge system and the first extraction result do not match, the first extraction result is corrected according to the large language model, based on the sampling strategy, the second prompt information, the knowledge system and the project information of the historical project to obtain the corrected first extraction result.

[0186] In the embodiment of this specification, the risk detection module is used to: Determining first risk information corresponding to the target project according to the risk information of the first historical project; Determine a risk detection model corresponding to the target project according to the business detection rules corresponding to the target project, wherein the risk detection model is a model constructed based on a preset machine learning algorithm; Determining, according to the risk detection model, second risk information corresponding to the target project based on the second extraction result corresponding to the target project; It is determined whether there is a risk in executing the target project according to the first risk information and the second risk information.

[0187] The embodiment of this specification provides a data processing device, which obtains the project information of the historical project and the risk information of the historical project, determines the first extraction result corresponding to the historical project according to the large language model and the project information of the historical project, and the first extraction result includes the summary information of the historical project, the risk factor information and the project requirement type information, and corrects the first extraction result according to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result to obtain the corrected first extraction result, and constructs a knowledge base according to the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed, so that, firstly, the computing power and understanding ability of the large language model can be used to quickly and accurately determine the first extraction result corresponding to the historical project. Secondly, on the one hand, by correcting the first extraction result through the knowledge system corresponding to the project requirement type information of the historical project, the reasoning process of the large language model can be enhanced, and the accuracy and credibility of the knowledge base construction can be improved. On the other hand, by correcting the first extraction result through the association relationship between each information in the first extraction result, the consistency and accuracy of each information in the corrected first extraction result can be improved. Furthermore, risk detection processing can be performed on the target to be executed based on the knowledge base constructed by the corrected first extraction result, thereby improving the accuracy of project risk detection.

[0188] The above is a data processing device provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a data processing device, such as Fig.11 shown.

[0189] The data processing device may provide a terminal device or a server, etc. for the above-mentioned embodiments.

[0190] The data processing device may have relatively large differences due to different configurations or performances, and may include one or more processors 1101 and memory 1102, and the memory 1102 may store one or more storage applications or data. Among them, the memory 1102 may be a short-term storage or a persistent storage. The application stored in the memory 1102 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the data processing device. Furthermore, the processor 1101 may be configured to communicate with the memory 1102 and execute a series of computer executable instructions in the memory 1102 on the data processing device. The data processing device may also include one or more power supplies 1103, one or more wired or wireless network interfaces 1104, one or more input and output interfaces 1105, and one or more keyboards 1106.

[0191] Specifically in this embodiment, the data processing 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 may include one or more modules, and each module may include a series of computer executable instructions in the data processing device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following: Obtaining project information of historical projects and risk information of the historical projects; Determine, according to the large language model, a first extraction result corresponding to the historical project based on the project information of the historical project, the first extraction result including summary information, risk factor information, and project requirement type information of the historical project; According to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain a corrected first extraction result; A knowledge base is constructed based on the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed.

[0192] Each embodiment in this 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 differences from other embodiments. In particular, for the data processing 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 partial description of the method embodiment.

[0193] The embodiment of this specification provides a data processing device, which obtains the project information of the historical project and the risk information of the historical project, determines the first extraction result corresponding to the historical project according to the large language model and the project information of the historical project, and the first extraction result includes the summary information of the historical project, the risk factor information and the project requirement type information, and corrects the first extraction result according to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result to obtain the corrected first extraction result, and builds a knowledge base according to the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed, so that, firstly, the computing power and understanding ability of the large language model can be used to quickly and accurately determine the first extraction result corresponding to the historical project. Secondly, on the one hand, by correcting the first extraction result through the knowledge system corresponding to the project requirement type information of the historical project, the reasoning process of the large language model can be enhanced, and the accuracy and credibility of the knowledge base construction can be improved. On the other hand, by correcting the first extraction result through the association relationship between each information in the first extraction result, the consistency and accuracy of each information in the corrected first extraction result can be improved. Furthermore, risk detection processing can be performed on the target to be executed based on the knowledge base constructed by the corrected first extraction result, thereby improving the accuracy of project risk detection.

[0194] Furthermore, based on the above Figures 1 to 9 In one embodiment, the present specification further provides a storage medium for storing computer executable instruction information. In a specific embodiment, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer executable instruction information stored in the storage medium is executed by the processor, the following process can be implemented: Obtaining project information of historical projects and risk information of the historical projects; Determine, according to the large language model, a first extraction result corresponding to the historical project based on the project information of the historical project, the first extraction result including summary information, risk factor information, and project requirement type information of the historical project; According to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain a corrected first extraction result; A knowledge base is constructed based on the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed.

[0195] Each embodiment in this 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 differences 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 partial description of the method embodiment.

[0196] The embodiment of this specification provides a storage medium, by obtaining the project information of the historical project and the risk information of the historical project, according to the large language model, based on the project information of the historical project, determining the first extraction result corresponding to the historical project, the first extraction result includes the summary information of the historical project, the risk factor information and the project requirement type information, according to the knowledge system corresponding to the project requirement type information of the historical project, and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain the corrected first extraction result, according to the corrected first extraction result and the risk information of the historical project, a knowledge base is constructed, and the knowledge base is used to perform risk detection processing on the project to be executed, so that, firstly, by using the computing power and understanding ability of the large language model, the first extraction result corresponding to the historical project can be quickly and accurately determined. Secondly, on the one hand, by correcting the first extraction result through the knowledge system corresponding to the project requirement type information of the historical project, the reasoning process of the large language model can be enhanced, and the accuracy and credibility of the knowledge base construction can be improved. On the other hand, by correcting the first extraction result through the association relationship between each information in the first extraction result, the consistency and accuracy of each information in the corrected first extraction result can be improved. Furthermore, risk detection processing can be performed on the target to be executed based on the knowledge base constructed by the corrected first extraction result, thereby improving the accuracy of project risk detection.

[0197] Furthermore, based on the above Figures 1 to 9 In one or more embodiments of the present specification, a computer program product is provided, including a computer program. When the computer program in the computer program product is executed by a processor, the following process can be implemented: Obtaining project information of historical projects and risk information of the historical projects; Determine, according to the large language model, a first extraction result corresponding to the historical project based on the project information of the historical project, the first extraction result including summary information, risk factor information, and project requirement type information of the historical project; According to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain a corrected first extraction result; A knowledge base is constructed based on the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed.

[0198] Each embodiment in this 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 differences 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 partial description of the method embodiment.

[0199] The embodiment of this specification provides a computer program product, which obtains the project information of the historical project and the risk information of the historical project, determines the first extraction result corresponding to the historical project according to the large language model and the project information of the historical project, and the first extraction result includes the summary information of the historical project, the risk factor information and the project requirement type information, and corrects the first extraction result according to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result to obtain the corrected first extraction result, and constructs a knowledge base according to the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed, so that, firstly, the computing power and understanding ability of the large language model can be used to quickly and accurately determine the first extraction result corresponding to the historical project. Secondly, on the one hand, by correcting the first extraction result through the knowledge system corresponding to the project requirement type information of the historical project, the reasoning process of the large language model can be enhanced, and the accuracy and credibility of the knowledge base construction can be improved. On the other hand, by correcting the first extraction result through the association relationship between each information in the first extraction result, the consistency and accuracy of each information in the corrected first extraction result can be improved. Furthermore, risk detection processing can be performed on the target to be executed based on the knowledge base constructed by the corrected first extraction result, thereby improving the accuracy of project risk detection.

[0200] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0201] In the 1990s, it was very clear whether the improvement of a technology was hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, 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. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0202] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, 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, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0203] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may 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.

[0204] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0205] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0206] The embodiments of this specification are described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, 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 fraud case serial and parallel device to produce a machine, so that the instructions executed by the processor of the computer or other programmable fraud case serial and parallel device generate instructions for implementing the processes in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0207] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable fraud case serial and parallel device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0208] These computer program instructions may also be loaded onto a computer or other programmable device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0209] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0210] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0211] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. 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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0212] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0213] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, one or more embodiments of this specification may be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Furthermore, one or more embodiments of this specification may be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0214] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0215] Each embodiment in this 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 differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0216] The above description is only an embodiment of this specification and is not intended to limit this document. For those skilled in the art, this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification should be included in the scope of the claims of this specification.

Claims

1. A data processing method, comprising: Obtaining project information of historical projects and risk information of the historical projects; Determine, according to the large language model, a first extraction result corresponding to the historical project based on the project information of the historical project, the first extraction result including summary information, risk factor information, and project requirement type information of the historical project; According to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain a corrected first extraction result; A knowledge base is constructed based on the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed.

2. According to the method of claim 1, the risk factor information includes product factor information, subject factor information, platform factor information and action factor information.

3. The method according to claim 2, further comprising: Receive risk detection requests for target projects to be executed; In response to the risk detection request, obtaining project information of the target project; Determining, according to the large language model and based on the project information of the target project, a second extraction result corresponding to the target project; Determine, according to the second extraction result and the corrected first extraction result corresponding to the historical item, a first historical item corresponding to the target item among the historical items included in the knowledge base; It is determined whether there is a risk in executing the target project according to the risk information of the first historical project.

4. The method according to claim 2, wherein determining, based on the second extraction result and the corrected first extraction result corresponding to the historical item, a first historical item corresponding to the target item in the historical items included in the knowledge base comprises: Determine the similarity between the target item and each of the historical items according to a preset similarity algorithm, based on the second extraction result and the corrected first extraction result corresponding to the historical items; According to the similarity between the target item and each of the historical items, a first historical item among the historical items corresponding to the target item is determined.

5. The method according to claim 2, before determining the first extraction result for the historical item based on the item information of the historical item according to the large language model, further comprising: Get a pre-trained large language model; Acquire historical sample data and a third extraction result corresponding to the historical sample data; According to the pre-trained large language model, based on preset prompt information, and the historical sample data, obtaining a fourth extraction result corresponding to the historical sample data; The pre-trained large language model is fine-tuned according to the third extraction result and the fourth extraction result to obtain a trained large language model.

6. The method according to claim 5, wherein determining the first extraction result corresponding to the historical item based on the large language model and the item information of the historical item comprises: Determine, according to the detection requirement corresponding to the historical item, first prompt information corresponding to the historical item, and perform expansion and enhancement processing on the first prompt information to obtain target prompt information; Obtaining a plurality of different candidate extraction results based on the large language model, the target prompt information, and the project information of the historical project; According to the reasoning path corresponding to the candidate extraction result, the candidate extraction result is logically analyzed to obtain a logical analysis result for each candidate extraction result; According to the occurrence frequency of each of the candidate extraction results and the logic analysis result, a first extraction result corresponding to the historical item is selected from the candidate extraction results.

7. The method according to claim 5, wherein determining the first extraction result corresponding to the historical item based on the large language model and the item information of the historical item comprises: According to the detection requirement corresponding to the historical item, obtaining first prompt information corresponding to the historical item; Obtaining a plurality of different candidate extraction results according to the large language model, based on the first prompt information, the project information of the historical project, and a plurality of different sampling parameters; According to the reasoning path corresponding to the candidate extraction result, the candidate extraction result is logically analyzed to obtain a logical analysis result for each candidate extraction result; According to the occurrence frequency of each of the candidate extraction results and the logic analysis result, a first extraction result corresponding to the historical item is selected from the candidate extraction results.

8. The method according to claim 5, wherein the first extraction result is corrected according to the knowledge system corresponding to the project requirement type information of the historical project to obtain the corrected first extraction result, comprising: Performing a match detection on the knowledge system and the first extraction result; In the case that the knowledge system and the first extraction result do not match, the first extraction result is corrected according to the large language model based on the knowledge system and the project information of the historical project to obtain the corrected first extraction result.

9. The method according to claim 5, wherein the first extraction result is corrected according to the correlation between each information in the first extraction result to obtain the corrected first extraction result, comprising: Determine a sampling strategy and second prompt information corresponding to the historical item according to the association relationship between each information in the first extraction result; According to the large language model, based on the sampling strategy, the second prompt information, and the project information of the historical project, the first extraction result is corrected to obtain the corrected first extraction result.

10. The method according to claim 5, wherein the first extraction result is corrected according to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result to obtain the corrected first extraction result, comprising: Determine a sampling strategy and second prompt information corresponding to the historical item according to the association relationship between each information in the first extraction result; Performing a match detection on the knowledge system and the first extraction result; In the case that the knowledge system and the first extraction result do not match, the first extraction result is corrected according to the large language model, based on the sampling strategy, the second prompt information, the knowledge system and the project information of the historical project to obtain the corrected first extraction result.

11. The method according to claim 3, wherein determining whether there is a risk in executing the target project based on the risk information of the first historical project comprises: Determining first risk information corresponding to the target project according to the risk information of the first historical project; Determine a risk detection model corresponding to the target project according to the business detection rules corresponding to the target project, wherein the risk detection model is a model constructed based on a preset machine learning algorithm; Determining, according to the risk detection model, second risk information corresponding to the target project based on the second extraction result corresponding to the target project; It is determined whether there is a risk in executing the target project according to the first risk information and the second risk information.

12. A data processing device, comprising: A first acquisition module is used to acquire project information of historical projects and risk information of the historical projects; A first determination module is used to determine a first extraction result corresponding to the historical project based on the project information of the historical project according to the large language model, wherein the first extraction result includes summary information, risk factor information, and project requirement type information of the historical project; A result correction module, used for correcting the first extraction result according to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result to obtain a corrected first extraction result; A construction module is used to construct a knowledge base based on the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed.

13. A data processing device, comprising: processor; as well as a memory arranged to store computer executable instructions which, when executed, cause the processor to: Obtaining project information of historical projects and risk information of the historical projects; Determine, according to the large language model, a first extraction result corresponding to the historical project based on the project information of the historical project, the first extraction result including summary information, risk factor information, and project requirement type information of the historical project; According to the knowledge system corresponding to the project requirement type information of the historical project and / or the association relationship between each information in the first extraction result, the first extraction result is corrected to obtain a corrected first extraction result; A knowledge base is constructed based on the corrected first extraction result and the risk information of the historical project, and the knowledge base is used to perform risk detection processing on the project to be executed.

Images