Artificial intelligence decision analysis method and device fusing model attribution and business rules

Abstract

This application discloses an AI decision analysis method and apparatus that integrates model attribution and business rules, belonging to the field of artificial intelligence. The method includes: calculating the contribution percentage and contribution characteristics of each input feature to the preliminary decision, and generating a model contribution report based on the contribution percentage and characteristics of each input feature; revising the preliminary decision using business rules, obtaining rule information corresponding to the decision points affected by the business rules, and generating a rule record report based on the rule information; fusing and structuring the model contribution report and the rule record report to obtain unified explanatory information; integrating the explanatory information into a preset structured prompt template according to the guidance of the preset structured prompt template, and inputting the preset structured prompt template filled with explanatory information into a large language model to generate an AI decision analysis report. This method can solve the problems of fragmented explanatory information and high technical barriers in existing technologies.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence, and in particular relates to an AI decision analysis method and apparatus that integrates model attribution and business rules. Background Technology

[0002] In the field of AI (Artificial Intelligence) assisted decision-making, machine learning models are widely used due to their powerful predictive capabilities. However, their "black box" nature makes the decision-making process opaque, giving rise to eXplainable Artificial Intelligence (XAI) technology. Currently, mainstream explanation techniques mainly include model attribution analysis (such as SHAP, which explains why the model makes the predictions as they are) and business rule enforcement (constraining and correcting model recommendations through deterministic rules).

[0003] Currently, model attribution analysis and business rule enforcement are separate outputs and presentations, leading to fragmented and incomplete decision explanations, high technicality, and poor business readability. For example, users receive separate model attribution reports and rule enforcement logs. When a rule corrects the model's recommendation (e.g., the model recommends a price increase, but the rule prohibits it due to the off-season), the two reports present contradictory information, making it difficult for users to understand the complete logical chain of the final decision, thus reducing their trust in the entire system. Furthermore, SHAP reports are too technical, while rule logs are too procedural; neither effectively "translates" technical findings into business insights, failing to clearly present the "game" process between the various factors influencing the decision. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in related technologies. To this end, this application proposes an AI decision analysis method and apparatus that integrates model attribution and rule-based execution, which automatically creates a unified, coherent, and easy-to-understand business decision summary, thereby solving the problems of fragmented information interpretation and high technical barriers in existing technologies, and improving the transparency and credibility of AI decision-making.

[0005] Firstly, this application provides an AI decision analysis method that integrates model attribution and business rules, the method comprising: Calculate the contribution percentage and contribution characteristics of each input feature to the model's initial decision, and generate a model contribution report based on the contribution percentage and contribution characteristics of each input feature. The initial decision is corrected using business rules, and rule information corresponding to the decision points affected by the business rules is obtained. A rule record report is generated based on the decision points and the rule information. The model contribution report and the rule record report are merged and structured to obtain unified interpretable information; Guided by the preset structured prompt template, the explanatory information is integrated into the preset structured prompt template, and the preset structured prompt template with the explanatory information is input into the large language model to generate an AI decision analysis report.

[0006] Secondly, this application provides an AI decision analysis device that integrates model attribution and business rules, the device comprising: The calculation module is used to calculate the contribution ratio and contribution characteristics of each input feature to the model's initial decision, and to generate a model contribution report based on the contribution ratio and contribution characteristics of each input feature. The correction module is used to correct the preliminary decision using business rules, obtain rule information corresponding to the decision points affected by the business rules, and generate a rule record report based on the rule information. The fusion module is used to fuse and structure the model contribution report and the rule record report to obtain unified interpretable information; The generation module is used to integrate the explanatory information into the preset structured prompt template according to the guidance of the preset structured prompt template, and input the preset structured prompt template filled with the explanatory information into the large language model to generate an AI decision analysis report.

[0007] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the AI ​​decision analysis method of fusion model attribution and business rules as described in the first aspect above.

[0008] Fourthly, this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the AI ​​decision analysis method for fusion model attribution and business rules as described in the first aspect above.

[0009] Fifthly, this application provides a chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the AI ​​decision analysis method of fusing model attribution and business rules as described in the first aspect.

[0010] Sixthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the AI ​​decision analysis method for fusing model attribution and business rules as described in the first aspect above.

[0011] The above-described one or more technical solutions in the embodiments of this application have at least one of the following technical effects: This application proposes an AI decision analysis method that integrates model attribution and business rules. It seamlessly integrates the originally separate model internal attribution (strategic level) and business rule execution log (tactical level) into a unified and coherent decision story chain, solving the technical pain point of fragmented or even contradictory explanatory information in the prior art.

[0012] Furthermore, this application proposes to transform complex technical data into intuitive business insights by processing the attribution values ​​of technologies such as SHAP and LIME through "quantification as contribution percentage" and "qualitative positive / negative game relationship". This allows users to instantly understand the key factors influencing decisions and their interactions, significantly lowering the understanding threshold of AI decision-making.

[0013] Furthermore, this application designs and applies a structured prompting module with a preset analysis framework to guide the large language model in generating high-quality, highly stable interpretations, automatically transforming them into AI decision analysis reports with clear hierarchical levels and business logic in natural language narrative. This transforms LLM from a general-purpose text tool into a professional AI decision interpretation engine, ensuring that the output business minutes always possess a clear logical hierarchy (strategy-tactics) and business insights, solving the problem that traditional methods struggle to effectively "translate" technical explanations into business language.

[0014] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0015] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is one of the flowcharts illustrating the AI ​​decision analysis method that integrates model attribution and business rules provided in the embodiments of this application; Figure 2 This is the second flowchart of the AI ​​decision analysis method that integrates model attribution and business rules provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the AI ​​decision analysis device for fusing model attribution and business rules provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0017] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0018] The following description, in conjunction with the accompanying drawings, details the AI ​​decision analysis method, device, electronic device, and readable storage medium for fusion model attribution and business rules provided in this application, through specific embodiments and application scenarios.

[0019] Among them, the AI ​​decision analysis method that integrates model attribution and business rules can be applied to the terminal, specifically executed by the hardware or software in the terminal.

[0020] The AI ​​decision analysis method for fusing model attribution and business rules provided in this application embodiment can be executed by an electronic device or a functional module or entity within an electronic device capable of implementing the AI ​​decision analysis method for fusing model attribution and business rules. The electronic devices mentioned in this application embodiment include, but are not limited to, mobile phones, tablets, computers, cameras, and wearable devices. The following description uses an electronic device as the execution subject to illustrate the AI ​​decision analysis method for fusing model attribution and business rules provided in this application embodiment.

[0021] like Figure 1 As shown, the AI ​​decision analysis method that integrates model attribution and business rules includes the following steps: S110, calculate the contribution ratio and contribution characteristics of each input feature to the model's preliminary decision, and generate a model contribution report based on the contribution ratio and contribution characteristics of each input feature.

[0022] S120: Modify the preliminary decision using business rules, obtain the rule information corresponding to the decision points affected by the business rules, and generate a rule record report based on the decision points and rule information.

[0023] S130 integrates and structures the model contribution report and rule record report to obtain unified interpretive information.

[0024] S140: Following the guidance of the preset structured prompt template, the explanatory information is integrated into the preset structured prompt template, and the preset structured prompt template with the explanatory information is input into the large language model to generate an AI decision analysis report.

[0025] The AI ​​decision analysis method that integrates model attribution and business rules provided in this application generates a model contribution report, including the contribution percentage and contribution characteristics of each input feature to the preliminary decision, and a rule record report, including the decision points affected by business rules when modifying the preliminary decision and the rule information corresponding to the decision points. Then, the model contribution report and the rule record report are fused and structured to obtain unified explanatory information. Finally, this unified explanatory information is integrated into a preset structured prompt template and input into a large language model, which outputs an AI decision analysis report. Because it integrates the model contribution report and the rule record report, and guides the input of unified explanatory information from both reports through a preset structured prompt module, and finally uses a large language model to analyze the preset structured prompt template containing the explanatory information, it can not only automatically create a business decision summary, but also solve the problems of fragmented explanatory information, high technical barriers, and poor business readability in existing technologies, greatly improving the transparency and credibility of AI decision-making.

[0026] The model can be a machine learning model, such as CatBoost (a powerful machine learning tool that can automatically and efficiently process data and is fast and easy to use).

[0027] In some embodiments, steps S110 includes: S111, obtain the initial contribution value of each input feature to the preliminary decision, and sum the absolute values ​​of each initial contribution value to obtain the total contribution value.

[0028] S112, calculate the ratio of each initial contribution value to the total contribution value to obtain the contribution percentage of each input feature.

[0029] S113, determine whether each input feature is a gain feature or a loss feature for the preliminary decision based on the positive or negative polarity of each initial contribution value.

[0030] S114, Sort the contribution ratio of each input feature according to the gain characteristics and / or loss characteristics, and generate a model contribution report based on the sorted input features.

[0031] Understandably, the initial contribution value is a numerical value with positive and negative polarity. Therefore, based on the positive and negative polarity of the initial contribution value, it can be determined whether the initial decision corresponding to each input feature is a gain characteristic (the contribution direction is the same as the initial decision, which is a "supporting force") or a detrimental characteristic (the contribution direction is opposite to the initial decision, which is a "balancing force"). The total contribution value is obtained by calculating the sum of the absolute values ​​of each initial contribution value. Then, the ratio of each initial contribution value to the total contribution value is calculated to obtain the corresponding contribution percentage. Finally, the contribution percentages of each input feature are sorted according to the aforementioned gain and / or detrimental characteristics, and a model contribution report is generated based on the sorted input features.

[0032] The initial contribution value can be obtained by calling the SHapley Additive ex Planations (SHAP) algorithm, the Local Interpretable Model-agnostic Explanations (LIME) algorithm, or other model attribution algorithms such as ensemble gradients.

[0033] The initial contribution value can be a SHAP value, LIME value, etc. It should be noted that the contribution percentage can be either a score or a percentage.

[0034] Furthermore, in some embodiments, step S114 includes: The input features are categorized according to the gain features to obtain the supporting input features.

[0035] The input features are categorized according to the aforementioned loss characteristics to obtain the balancing input features.

[0036] The contribution percentages of the supporting input features and the balancing input features are sorted in descending order, and the sorted supporting input features and the balancing input features are written into the first preset report template to form the model contribution report.

[0037] In this embodiment, the input features are first classified according to their gain and loss characteristics, and then sorted in descending order according to their contribution percentage within each category. This yields the supporting and balancing input features that have a significant impact on the initial decision. The supporting and balancing input features are then input into the first preset report template to obtain a structured model contribution report that reflects the "power game" relationship, providing a reliable basis for the subsequent AI decision analysis report.

[0038] For example, the first preset report template may include a first feature column for recording supporting input features and a first percentage column for recording the contribution percentage of each supporting input feature, as well as a second feature column for recording balancing input features and a second percentage column for recording the contribution percentage of each balancing input feature.

[0039] In some embodiments, step S120 includes: The initial decision is input into the business rule engine. The business rule engine corrects the initial decision according to the preset business logic and records each decision point affected by the business rule and the final decision, as well as the rule identifier triggered by each decision point, the rule content, the content before and after correction, and the core driving factors of the final decision.

[0040] Each decision point, the rule identifier triggered by that decision point, the rule content, the content before and after the correction, and the final decision are written into the second preset report template to form the rule record report.

[0041] In this embodiment, the preliminary decision is input into the business rule engine, and the business rule engine corrects the preliminary decision to obtain each decision point affected by the business rule, the final decision, the rule identifier triggered by each decision point, the rule content, the content before and after correction, and the core driving factors of the final decision. These contents are written into the second preset report template to obtain a structured rule record report.

[0042] The business rules engine is a system component that extracts specific business logic (such as if-then statements) from the code, enabling independent configuration, dynamic management, and efficient execution. Compared to code, it is more flexible and efficient.

[0043] For example, the second preset report template may include entry boxes corresponding to "decision points and final decisions, as well as rule identifiers triggered by each decision point, rule content, content before and after correction, and core driving factors of the final decision," to guide the input of this content in the corresponding locations.

[0044] It should be noted that a rule-based model engine is used because the initial decision output by the model may not follow industry rules, thus requiring correction and constraints. For example, for a certain business model, the initial decision output by the model is to continuously increase the price to 3,000 yuan, but business rules stipulate that the price should not be increased during the off-season and the price cannot exceed 2,000 yuan. Therefore, the initial decision output by the model needs to be corrected to be below 2,000 yuan.

[0045] In some embodiments, step S130 includes: The same feature in the model contribution report and the rule record report is standardized and structured, and explanatory information of the same feature is obtained to obtain the unified explanatory information.

[0046] The model contribution report and rule record report are understandable reports output based on different rules. Therefore, the output expression of the same content may be different. It is necessary to standardize and structure the content expressing the same feature in the two reports, and then combine the content of the two reports to obtain the explanatory information of the same feature, so as to obtain unified explanatory information, so as to input the preset structured prompt template with unified expression content in the future.

[0047] For example, if the target data object recorded in the model contribution report is named "Data 001" and its corresponding explanatory description is "first data, commodity value", and the target data object recorded in the rule record report is named "Data 1" and its corresponding explanatory description is "first data, representing the initial price of commodity X", then the target data can be standardized and structured as "Data 1, first data, trademark value, initial price of commodity X".

[0048] In some embodiments, the preset structured prompt template includes at least: Core decision prompts are used to guide the input of the final decision and the core driving factors of that final decision; Market insight prompts are used to guide input based on the model's contribution report to obtain macroeconomic market trends and judgment criteria; The rule execution prompt instruction is used to guide the input of processing rules for key decision points obtained from the rule record report; Summary suggestions and prompts are used to guide the input of comprehensive conclusions and risk monitoring points.

[0049] Correspondingly, step S140 includes: The final decision and its core driving factors are written into a preset structured prompt template based on the core decision prompt instructions.

[0050] Based on market insight prompts, the contribution percentage and contribution characteristics of each input feature are written into a preset structured prompt template.

[0051] Based on the rule execution prompt instructions, each decision point and the rule identifier, rule content, and the content before and after correction are written into a preset structured prompt template.

[0052] Based on the summary and suggestion prompts, comprehensive conclusions and risk monitoring points are written into a preset structured prompt template.

[0053] Input the pre-defined structured prompt template with explanatory information into the large language model to generate an AI decision analysis report, which is then formatted as natural language text in a lightweight markup language format.

[0054] In this embodiment, the preset structured prompt template is not simply a collection of information, but rather a mandatory analysis logic from strategy to tactics. Each prompt instruction is used to guide the input of corresponding content in the corresponding position to obtain the preset structured prompt template with explanatory information, thereby guiding the subsequent large language model to perform high-quality and high-stability interpretation according to this logic and output an AI decision analysis report.

[0055] It should be noted that explanatory information includes all the content entered based on the aforementioned prompts and instructions.

[0056] Among them, the text content of the AI ​​decision analysis report, which is formatted in a lightweight markup language, can be highlighted by bolding or underlining to make it easier for users to read.

[0057] The AI ​​decision analysis method that integrates model attribution and business rules provided in this application can be executed by an AI decision analysis device that integrates model attribution and business rules. This application example illustrates the AI ​​decision analysis device that integrates model attribution and business rules provided in this application by using the AI ​​decision analysis device that integrates model attribution and business rules to execute the AI ​​decision analysis method.

[0058] It's worth noting that in scenarios where a large language model isn't used, a rule-based and template-based natural language generation system can be employed as an alternative. This system predefines multiple narrative templates corresponding to different decision-making scenarios (such as price increases, price decreases, significant rule intervention, etc.), and then fills the fused explanatory information into the best-matching template to generate a relatively fixed-format interpretive text. This alternative solution can also achieve the core objectives of information fusion and narrative output.

[0059] This application also provides an AI decision analysis device that integrates model attribution and business rules, such as... Figure 3 As shown, the AI ​​decision analysis device 100 that integrates model attribution and business rules includes: The calculation module 110 is used to calculate the contribution ratio and contribution characteristics of each input feature to the model's preliminary decision, and to generate a model contribution report based on the contribution ratio and contribution characteristics of each input feature.

[0060] The correction module 120 is used to correct the preliminary decision using business rules, obtain rule information corresponding to the decision points affected by the business rules, and generate a rule record report based on the rule information.

[0061] The fusion module 130 is used to fuse and structure the model contribution report and the rule record report to obtain unified interpretable information.

[0062] The generation module 140 is used to integrate the explanatory information into the preset structured prompt template according to the guidance of the preset structured prompt template, and input the preset structured prompt template filled with the explanatory information into the large language model to generate an AI decision analysis report.

[0063] The AI ​​decision analysis device for fusing model attribution and business rules provided in this application generates a model contribution report including the contribution ratio and contribution characteristics of each input feature to the preliminary decision, and a rule record report including the decision points affected by business rules when the preliminary decision is modified by business rules and the rule information corresponding to the decision points. Then, the model contribution report and the rule record report are fused and structured to obtain unified explanatory information. Finally, the unified explanatory information is integrated into a preset structured prompt template and input into a large language model, which outputs an AI decision analysis report. Because it integrates the model contribution report and the rule record report, and guides the input of unified explanatory information from the two reports through a preset structured prompt module, and finally uses the large language model to analyze and process the preset structured prompt template with the explanatory information, it can not only automatically create a business decision summary, but also solve the problems of fragmented explanatory information, high technical threshold and poor business readability in the prior art, greatly improving the transparency and credibility of AI decision-making.

[0064] In some embodiments, the calculation module 110 is further configured to obtain the initial contribution value of each input feature to the preliminary decision, and sum the absolute values ​​of each initial contribution value to obtain the total contribution value; calculate the ratio of each initial contribution value to the total contribution value to obtain the contribution ratio of each input feature; determine whether each input feature is a gain feature or a loss feature to the preliminary decision based on the positive or negative polarity of each initial contribution value; sort the contribution ratios of each input feature according to the gain feature and / or loss feature, and generate a model contribution report based on the sorted input features.

[0065] In some embodiments, the calculation module 110 is further configured to classify each input feature according to the gain feature to obtain supporting input features; classify each input feature according to the loss feature to obtain balancing input features; sort the contribution ratios corresponding to the supporting input features and the balancing input features in descending order, and write the sorted supporting input features and the balancing input features into a first preset report template to form the model contribution report.

[0066] In some embodiments, the correction module 120 is further configured to input the preliminary decision into a business rule engine, the business rule engine to correct the preliminary decision according to a preset business logic, and record each decision point affected by the business rule and the final decision, as well as the rule identifier, rule content, and content before and after correction triggered by each decision point and the core driving factors of the final decision; and write each decision point, the rule identifier, rule content, and content before and after correction triggered by the decision point, and the final decision into a second preset report template to form the rule record report.

[0067] In some embodiments, the fusion module 130 is further configured to standardize and structure the same feature in the model contribution report and the rule record report, and obtain explanatory information of the same feature to obtain the unified explanatory information.

[0068] In some embodiments, the preset structured prompt template includes at least: Core decision prompts are used to guide the input of the final decision and the core driving factors of that final decision; Market insight prompts are used to guide input based on the model's contribution report to obtain macroeconomic market trends and judgment criteria; The rule execution prompt instruction is used to guide the input of processing rules for key decision points obtained from the rule record report; Summary suggestions and prompts are used to guide the input of comprehensive conclusions and risk monitoring points.

[0069] In some embodiments, the generation module 140 is further configured to: write the final decision and its core driving factors into the preset structured prompt template based on the core decision prompt instruction; write the contribution ratio and contribution characteristics of each input feature into the preset structured prompt template based on the market insight prompt instruction; write each decision point and its triggered rule identifier, rule content, and content before and after correction into the preset structured prompt template based on the rule execution prompt instruction; write the comprehensive conclusion and risk monitoring points into the preset structured prompt template based on the summary suggestion prompt instruction; input the preset structured prompt template filled with the explanatory information into a large language model to generate the AI ​​decision analysis report, wherein the generated AI decision analysis report is natural language text formatted in a lightweight markup language.

[0070] The AI ​​decision analysis based on the fusion model attribution and business rules provided in this application aims to solve the technical problems of incomplete decision explanation information, high technicality, and poor business readability in existing AI-assisted decision-making processes. Specifically, the technical problems to be solved by this invention are as follows: 1) Existing technologies, when explaining AI decisions, typically present the attribution analysis of machine learning models or the execution records of business rules in isolation, lacking a method that can organically integrate these two into a unified and coherent decision-making chain, resulting in gaps in the explanation.

[0071] 2) The raw numerical output of existing model attribution methods has a high technical threshold, making it difficult for non-professional users to understand. In particular, it cannot intuitively present the "game relationship" of support and checks and balances among various influencing factors, resulting in low efficiency in information interpretation.

[0072] 3) Existing technologies lack a method to automatically transform multi-source and complex technical explanation information into a clear and business-logical natural language narrative report, which causes the explanation process of AI decision-making to become disconnected from the actual needs of business users, affecting the credibility of the decision.

[0073] The AI ​​decision analysis device for fusing model attribution and business rules in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc. This application embodiment does not specifically limit the specific devices.

[0074] The AI ​​decision analysis device that integrates model attribution and business rules in this application embodiment can be a device with an operating system. This operating system can be a Microsoft (Windows) operating system, an Android operating system, an iOS operating system, or other possible operating systems; this application embodiment does not specifically limit it.

[0075] The AI ​​decision analysis device that integrates model attribution and business rules provided in this application embodiment can achieve... Figures 1 to 2The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0076] In some embodiments, such as Figure 4 As shown, this application embodiment also provides an electronic device 800, including a processor 801, a memory 802, and a computer program stored on the memory 802 and executable on the processor 801. When the program is executed by the processor 801, it implements the various processes of the above-described AI decision analysis method embodiment of fusion model attribution and business rules, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0077] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0078] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described AI decision analysis method embodiment of fusion model attribution and business rules, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0079] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0080] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described AI decision analysis method for fusion model attribution and business rules.

[0081] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0082] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described embodiment of the AI ​​decision analysis method for fusing model attribution and business rules, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0083] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0084] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0085] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0086] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0087] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0088] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. An AI decision analysis method that integrates model attribution and business rules, characterized in that, include: Calculate the contribution percentage and contribution characteristics of each input feature to the model's initial decision, and generate a model contribution report based on the contribution percentage and contribution characteristics of each input feature. The initial decision is corrected using business rules, and rule information corresponding to the decision points affected by the business rules is obtained. A rule record report is generated based on the decision points and the rule information. The model contribution report and the rule record report are merged and structured to obtain unified interpretable information; Guided by the preset structured prompt template, the explanatory information is integrated into the preset structured prompt template, and the preset structured prompt template with the explanatory information is input into the large language model to generate an AI decision analysis report.

2. The method according to claim 1, characterized in that, Calculate the contribution percentage and contribution characteristics of each input feature to the preliminary decision, and generate a model contribution report based on the contribution percentage and contribution characteristics of each input feature, including: Obtain the initial contribution value of each input feature to the preliminary decision, and sum the absolute values ​​of each initial contribution value to obtain the total contribution value; Calculate the ratio of each initial contribution value to the total contribution value to obtain the contribution percentage of each input feature; Based on the positive or negative polarity of each initial contribution value, determine whether each input feature contributes to the preliminary decision as a gain feature or a loss feature. The contribution percentages of each input feature are sorted according to the gain characteristics and / or the loss characteristics, and a model contribution report is generated based on the sorted input features.

3. The method according to claim 2, characterized in that, The contribution ratio of each input feature is sorted according to the gain feature and / or the loss feature, and a model contribution report is generated based on the sorted input features, including: The input features are categorized according to the gain features to obtain the supporting input features; The input features are classified according to the aforementioned loss characteristics to obtain the balancing input features; The contribution percentages of the supporting input features and the balancing input features are sorted in descending order, and the sorted supporting input features and balancing input features are written into the first preset report template to form the model contribution report.

4. The method according to claim 1, characterized in that, The step of revising the initial decision using business rules, obtaining rule information corresponding to the decision points affected by the business rules, and generating a rule record report based on the decision points and the rule information includes: The preliminary decision is input into the business rule engine, which corrects the preliminary decision according to the preset business logic, and records each decision point affected by the business rule and the final decision, as well as the rule identifier, rule content, content before and after correction, and the core driving factors of the final decision triggered by each decision point. Each decision point, the rule identifier triggered by that decision point, the rule content, the content before and after the correction, and the final decision are written into the second preset report template to form the rule record report.

5. The method according to claim 1, characterized in that, The model contribution report and the rule record report are fused and structured to obtain unified interpretive information, including... The same feature in the model contribution report and the rule record report is standardized and structured, and explanatory information of the same feature is obtained to obtain the unified explanatory information.

6. The method according to any one of claims 1-5, characterized in that, The preset structured prompt template includes at least: Core decision prompts are used to guide the input of the final decision and the core driving factors of that final decision; Market insight prompts are used to guide input based on the model's contribution report to obtain macroeconomic market trends and judgment criteria; The rule execution prompt instruction is used to guide the input of processing rules for key decision points obtained from the rule record report; Summary suggestions and prompts are used to guide the input of comprehensive conclusions and risk monitoring points.

7. The method according to claim 6, characterized in that, The process involves integrating the explanatory information into a pre-defined structured prompt template, and then inputting the pre-defined structured prompt template containing the explanatory information into a large language model to generate an AI decision analysis report, including: Based on the core decision prompt instructions, the final decision and its core driving factors are written into the preset structured prompt template; Based on the market insight prompt instruction, the contribution percentage and contribution characteristics of each input feature are written into the preset structured prompt template; Based on the rules, the execution prompt instruction writes each decision point and its triggered rule identifier, rule content, and the content before and after correction into the preset structured prompt template; Based on the summary and suggestion prompts, the comprehensive conclusions and risk monitoring points are written into the preset structured prompt template; The preset structured prompt template containing the explanatory information is input into the large language model to generate the AI ​​decision analysis report, which is a natural language text formatted in a lightweight markup language.

8. An AI decision analysis device that integrates model attribution and business rules, characterized in that, The device includes: The calculation module is used to calculate the contribution ratio and contribution characteristics of each input feature to the model's initial decision, and to generate a model contribution report based on the contribution ratio and contribution characteristics of each input feature. The correction module is used to correct the preliminary decision using business rules, obtain rule information corresponding to the decision points affected by the business rules, and generate a rule record report based on the rule information. The fusion module is used to fuse and structure the model contribution report and the rule record report to obtain unified interpretable information; The generation module is used to integrate the explanatory information into the preset structured prompt template according to the guidance of the preset structured prompt template, and input the preset structured prompt template filled with the explanatory information into the large language model to generate an AI decision analysis report.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the AI ​​decision analysis method of fusion model attribution and business rules as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the AI ​​decision analysis method for fusing model attribution and business rules as described in any one of claims 1-7.

Images