Risk assessment method and related device based on rule engine and hybrid reasoning

Abstract

The application relates to the field of artificial intelligence and discloses a risk assessment method based on a rule engine and mixed reasoning and related equipment, which comprises the following steps: acquiring multidimensional data of risk points containing structured numerical data and unstructured text data; inputting the structured numerical data into a preset rule engine to output rule scores and hit information; determining the relevance of the two types of data according to the occurrence probability and the influence degree; constructing reasoning prompt input based on the unstructured text data and the associated rule scores and hit information, inputting the reasoning prompt input into a large language model to perform semantic reasoning and output semantic scores; performing an aggregation strategy on the scores of each dimension according to the dimension to obtain occurrence probability scores and influence degree scores; and finally fusing and determining to obtain a risk assessment level. The application can be applied to business scenarios such as financial technology, realizes the bidirectional coupling of the rule engine and the large language model through the constraint of the rule scores and the hit information on the semantic reasoning, and effectively improves the accuracy and interpretability of risk assessment.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology and can be applied to business areas such as fintech, particularly to risk assessment methods and related equipment based on rule engines and hybrid reasoning. Background Technology

[0002] Risk assessment is a core technical means for compliance and quality control in various fields. For example, in the financial sector, risk assessment is widely used in compliance risk management, operational risk rating, and regulatory event classification.

[0003] As risk data becomes increasingly diverse, risk assessment requires processing both structured numerical data and unstructured text data simultaneously. This makes traditional single-technology-architecture assessment solutions difficult to adapt to complex business assessment needs. Even when some solutions attempt to combine the two types of technologies, they are mostly simple splicing of parallel and independent scoring, which makes it difficult to ensure the reliability and accuracy of risk assessment. Summary of the Invention

[0004] The main objective of this invention is to provide a risk assessment method and related equipment based on rule engines and hybrid reasoning, aiming to solve the technical problem in the prior art that it is difficult to reliably assess risk data that is a mixture of structured and unstructured risk data, and to improve the reliability and accuracy of risk assessment.

[0005] The technical solution of the present invention is as follows: The first aspect of this invention provides a risk assessment method based on rule engines and hybrid reasoning, comprising: Obtain multidimensional data of the risk points to be assessed, including structured numerical data and unstructured text data; The structured numerical data in the multidimensional data is input into a preset rule engine, and at least one rule score and rule hit information are output through corresponding threshold matching and rule priority determination. Determine the correlation between various structured numerical data and unstructured text data based on the dimensions of probability of occurrence and degree of impact; Based on the rule scores and rule hit information of the unstructured text data, the associated structured numerical data, the corresponding reasoning prompts are constructed and input into the large language model for semantic reasoning, and at least one semantic score is output. The scoring results of each dimension are aggregated according to the probability of occurrence dimension and the degree of impact dimension to obtain the probability of occurrence score and the degree of impact score. The risk value is fused and the level is determined based on the probability score and the impact score to obtain the corresponding risk assessment level.

[0006] A second aspect of the present invention provides a risk assessment device based on a rule engine and hybrid reasoning, comprising: The data acquisition module is used to acquire multidimensional data of the risk points to be assessed, including structured numerical data and unstructured text data. The rule scoring module is used to input the structured numerical data in the multidimensional data into the preset rule engine, and output at least one rule score and rule hit information through corresponding threshold matching and rule priority determination; The association segmentation module is used to determine the associations between various structured numerical data and unstructured text data according to the dimensions of probability of occurrence and degree of impact. The semantic scoring module is used to construct corresponding reasoning prompts based on the rule scores and rule hit information of the unstructured text data, the associated structured numerical data, and input them into the large language model for semantic reasoning, and output at least one semantic score. The rating aggregation module is used to perform aggregation strategies on the rating results of each dimension according to the probability of occurrence dimension and the degree of impact dimension, to obtain the probability of occurrence rating and the degree of impact rating. The risk fusion and determination module is used to fuse risk values ​​and determine risk levels based on the probability of occurrence score and the degree of impact score, so as to obtain the corresponding risk assessment level.

[0007] A third aspect of the present invention provides a computer device including at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the aforementioned risk assessment method based on rule engine and hybrid reasoning.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the aforementioned risk assessment method based on rule engines and hybrid reasoning.

[0009] Beneficial Effects: This invention discloses a risk assessment method and related equipment based on a rule engine and hybrid reasoning. Compared with existing technologies, this invention acquires multi-dimensional risk point data including structured numerical data and unstructured text data; inputs the structured numerical data into a preset rule engine to output rule scores and hit information; determines the correlation between the two types of data according to the dimensions of probability of occurrence and degree of impact; constructs a reasoning prompt input large language model based on unstructured text data and associated rule scores and hit information for semantic reasoning and outputs semantic scores; performs aggregation strategies on the scores of each dimension to obtain probability of occurrence scores and degree of impact scores; and finally, fuses and determines the risk assessment level. This invention can be applied to business scenarios such as fintech. By constraining semantic reasoning through rule scores and hit information, it achieves bidirectional coupling between the rule engine and the large language model, effectively improving the accuracy and interpretability of risk assessment. Attached Figure Description

[0010] To more clearly illustrate the solutions in this invention, the accompanying drawings used in the description of the embodiments of this invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0011] Figure 1 A schematic diagram of an application environment for the risk assessment method based on rule engine and hybrid reasoning provided in an embodiment of the present invention; Figure 2 A flowchart of a risk assessment method based on rule engine and hybrid reasoning provided in an embodiment of the present invention; Figure 3 A flowchart of step S202 in the risk assessment method based on rule engine and hybrid reasoning provided in the embodiments of the present invention; Figure 4 A flowchart of step S204 in the risk assessment method based on rule engine and hybrid reasoning provided in the embodiments of the present invention; Figure 5 This is another flowchart of step S202 in the risk assessment method based on rule engine and hybrid reasoning provided in the embodiments of the present invention; Figure 6 This is another flowchart of step S204 in the risk assessment method based on rule engine and hybrid reasoning provided in the embodiments of the present invention; Figure 7 Another flowchart of the risk assessment method based on rule engine and hybrid reasoning provided in the embodiments of the present invention; Figure 8A flowchart of step S206 in the risk assessment method based on rule engine and hybrid reasoning provided in the embodiments of the present invention; Figure 9 A schematic diagram of the functional modules of the risk assessment device based on rule engine and hybrid reasoning provided in an embodiment of the present invention; Figure 10 A schematic diagram of the hardware structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention is further described in detail below. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The embodiments of the invention are described below in conjunction with the accompanying drawings.

[0013] The risk assessment method based on rule engines and hybrid reasoning provided in this invention can be applied to, for example... Figure 1 In the application environment, it includes a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0014] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as financial clients, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (for example only).

[0015] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0016] Server 105 can be a server providing various services, such as a backend server supporting the content browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend server can analyze and process received user requests and other data, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices. Server 105 can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the shortcomings of traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS"), such as high management difficulty and weak business scalability. Server 105 can also be a server for a distributed system or a server combined with blockchain.

[0017] It should be noted that the risk assessment method based on rule engine and hybrid reasoning provided in this application embodiment can generally be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103. Correspondingly, the risk assessment device based on rule engine and hybrid reasoning provided in this embodiment can also be located in the first terminal device 101, the second terminal device 102, or the third terminal device 103. Alternatively, the risk assessment method based on rule engine and hybrid reasoning provided in this embodiment can generally also be executed by the server 105. Correspondingly, the risk assessment device based on rule engine and hybrid reasoning provided in this embodiment can generally be located in the server 105.

[0018] It should be understood that the number of terminal devices, networks, and servers listed above is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be used.

[0019] like Figure 2 As shown, the risk assessment method based on rule engine and hybrid reasoning provided in this embodiment of the invention specifically includes the following steps: S201. Obtain multidimensional data of the risk points to be assessed, wherein the multidimensional data includes structured numerical data and unstructured text data.

[0020] In this embodiment, complete multidimensional data of the risk point to be assessed is obtained from the upstream business system or data acquisition terminal, and divided into two categories according to data form: structured numerical data and unstructured text data. Structured numerical data consists of quantifiable statistical and count indicators, stored in integer or numerical form, and can be used to reflect the quantifiable attributes of risk. Examples include historical frequency data (covering the number of risk occurrences in the past year, past two years, and past three consecutive years); impact scope data (covering the number of affected financial institutions, critical infrastructure units, and business units); and risk item scale data (covering the total number of risk items associated with the risk point). Unstructured text data is risk-related information described in natural language, containing semantic information and contextual logic, and can be used to reflect the qualitative attributes of risk. Examples include risk point description text, risk factor explanation text, regulatory background and compliance requirement text, and historical event details text. The structured and unstructured data together constitute the complete input set for risk assessment, providing a data foundation for subsequent modal reasoning.

[0021] Preferably, after acquiring the multidimensional data of the risk points to be assessed, the system further performs legality and integrity checks on the acquired multidimensional data. The checks include verifying that the risk point description field is not empty, the risk type field is not empty, and all structured numerical fields are non-negative integers. If any check fails, the scoring process is terminated directly, and the corresponding error message is returned. Only after all fields pass the checks does the system proceed to the subsequent rule engine scoring stage. This pre-emptive data integrity check intercepts illegal input data in advance, preventing invalid data from entering the subsequent reasoning process and causing wasted computational resources and abnormal scoring results.

[0022] For example, in a bank's internal control operation risk assessment scenario, if the risk point to be assessed is the risk of customer information leakage in the core business system, the obtained structured numerical data includes quantitative indicators such as the number of information leakage incidents in the past one to three years, the number of financial companies affected, the number of critical infrastructures, the number of business departments, and the total number of related risk items; the unstructured text data includes text content such as risk point descriptions, explanations of information leakage risk factors, relevant regulatory requirements for personal information protection, and details of historical data leakage incidents.

[0023] This embodiment, by uniformly collecting both structured and unstructured risk data, can comprehensively cover the quantifiable and semantic attributes of risk points, providing complete risk data for subsequent bimodal hybrid reasoning, avoiding the lack of risk assessment dimensions caused by a single data format, and ensuring the completeness and reliability of risk assessment.

[0024] S202. Input the structured numerical data in the multidimensional data into the preset rule engine, and output at least one rule score and rule hit information through corresponding threshold matching and rule priority determination.

[0025] In this embodiment, the preset rule engine has a pre-configured scoring rule library. Each rule in the scoring rule library includes a trigger condition, a corresponding score, and a rule description. The rule engine performs scoring processing on the structured data in the multidimensional data. This involves inputting all structured numerical data into the rule engine, and then calling the corresponding scoring rules for matching the structured numerical data of different dimensions. Specifically, it matches each numerical indicator contained in the structured numerical data one by one according to the judgment priority and threshold in the trigger condition. After a successful match, it outputs the rule score for each dimension of the structured numerical data, and simultaneously records the description of the currently matched rule condition as rule hit information. Preferably, the rule scores can adopt a unified score range to ensure that the rule scores of different dimensions are aggregable, facilitating subsequent fusion processing.

[0026] For example, in the risk assessment of customer information leakage in commercial banks, the rule engine performs threshold matching on numerical indicators such as the number of historical events and the number of departments involved according to their respective priorities, outputs the rule score of the corresponding dimension, and records the specific threshold conditions that are hit.

[0027] This embodiment uses a rule engine to perform deterministic scoring on structured numerical data, ensuring the accuracy and efficiency of quantifiable risk scoring results. At the same time, the output of rule scores also provides specific rule hit information, providing a constraint basis for subsequent semantic reasoning, thereby improving the reliability of risk assessment.

[0028] S203. Determine the correlation between various structured numerical data and unstructured text data according to the dimensions of probability of occurrence and degree of impact.

[0029] In this embodiment, a mapping relationship between dimensions and data is pre-established, dividing all dimensional data into two major categories: probability of occurrence and degree of impact. For each of these two categories, corresponding structured numerical indicators and unstructured text indicators are matched. This determines the correlation between various structured numerical and unstructured text data according to the probability of occurrence and degree of impact dimensions, clarifying which structured numerical indicators and which text descriptions jointly support the risk assessment of the probability of occurrence or degree of impact dimensions. After determining the correlation between different forms of data, structured numerical data and unstructured text data under the same dimension (i.e., the probability of occurrence or degree of impact dimension) are paired, providing a data pairing basis for subsequent constrained semantic reasoning.

[0030] Specifically, structured numerical data related to historical occurrence frequency and unstructured text data such as risk factor text and risk description text are categorized into the occurrence probability dimension to establish the correlation between historical occurrence frequency and risk factors and risk descriptions; structured numerical data related to the scope of impact, structured numerical data related to the size of risk items, and unstructured text data related to regulatory background information and historical event information are categorized into the degree of impact dimension to establish the correlation between the value of the scope of impact and regulatory background information, as well as the correlation between the value of the size of risk items and historical event information.

[0031] This embodiment establishes the correlation between structured and unstructured data by determining the correlation between structured and unstructured data according to the dimensions of probability of occurrence and degree of influence. This enables subsequent large language model inference to perform semantic enhancement based on the deterministic results of rule scoring, rather than performing completely independent unconstrained text inference, which is beneficial to improving the accuracy of inference.

[0032] S204. Based on the unstructured text data, the rule scores and rule hit information of the associated structured numerical data, construct corresponding reasoning prompts and input them into the large language model for semantic reasoning, and output at least one semantic score.

[0033] In this embodiment, based on the aforementioned correlations between various structured numerical data and unstructured text data, a structured reasoning prompt is constructed by combining the rule scores and rule hit information of the unstructured text data and the corresponding related structured data. Specifically, at least one reasoning prompt is constructed according to both the probability of occurrence and the degree of impact dimensions. This prompt includes role settings, scoring criteria, constraints, and the text to be evaluated. Rule scores and hit information are injected as pre-constraints into the reasoning prompt to guide the large language model in semantic reasoning based on objectively quantified scoring results. After the constructed reasoning prompt is input into the large language model, the model performs semantic understanding, context analysis, and risk scoring, outputting the semantic scoring results of the corresponding dimension of unstructured text data.

[0034] For example, in the risk assessment of customer information leakage in commercial banks, the rule scores of the number of historical events and the hit rules can be used as constraints. Combined with risk factors and risk description text, reasoning prompts can be constructed. The input can be used to infer the probability of future occurrence in a large language model, and the risk semantic score under the dimensions of risk factors and risk description text can be obtained.

[0035] S205. Perform aggregation strategies on the scoring results of each dimension according to the probability of occurrence dimension and the degree of impact dimension to obtain the probability of occurrence score and the degree of impact score.

[0036] In this embodiment, all rule scores and semantic scores are grouped according to two main dimensions: probability of occurrence and degree of impact. Each group contains scores for multiple sub-dimensions. For example, historical frequency scores and future probability scores are grouped into the probability of occurrence dimension score group, while regulatory dimension scores, scope of impact scores, number of risk items scores, and historical event scores are grouped into the degree of impact dimension score group. An aggregation strategy is applied to the multiple sub-dimension scores within each group. Specifically, a maximum value aggregation strategy is applied, which takes the highest score among all sub-dimension scores under the same major category as the comprehensive score for that major category. This results in the output of the comprehensive probability of occurrence score and the comprehensive degree of impact score, ensuring that a high-risk result in any sub-dimension is not diluted by other low-scoring dimensions.

[0037] This embodiment ensures that the significance of a single high-risk dimension is not diluted by executing the maximum value aggregation strategy according to the two major dimension categories, and achieves effective integration of multi-dimensional scores. It retains the deterministic advantages of the rule engine and integrates the semantic understanding depth of the large language model, thereby improving the reliability of risk classification.

[0038] S206. Based on the probability score and the impact score, risk values ​​are integrated and level is determined to obtain the corresponding risk assessment level.

[0039] In this embodiment, the probability score and the impact score are numerically fused to calculate the final total risk value, which comprehensively reflects the probability of the risk occurring and the degree of its impact. Simultaneously, multiple risk level threshold ranges are pre-set, and the calculated total risk value is matched with these threshold ranges to determine the risk assessment level corresponding to the risk point.

[0040] For example, in the risk assessment of customer information leakage in commercial banks, the probability score of 4 points and the impact score of 5 points are combined to obtain a total risk score of 20 points; after level comparison, it is confirmed that 20 points belong to the high-risk range [16,25], and the risk assessment level is determined to be high risk.

[0041] Furthermore, after obtaining the risk assessment level of the current risk point, the system automatically integrates all data and results from the entire scoring process to generate a structured comprehensive scoring explanation report. The report includes the calculation formula and detailed calculation process for the total risk value, the final risk assessment level and matching control recommendations, detailed scoring for each sub-dimension (including dimension name, corresponding score, data source type, rule hit information or semantic reasoning basis), dimension comparison and value explanation for the maximum value aggregation process, etc. This explanation report is output in a standardized text format and can be directly used for audit traceability and regulatory reporting, meeting the compliance management requirements of the financial sector.

[0042] This embodiment combines the scoring of probability of occurrence and degree of impact with a level determination to form a complete risk assessment output, which can intuitively reflect the overall severity of the risk and provide a clear decision-making basis for subsequent risk management.

[0043] In the above embodiments, this invention discloses a risk assessment method based on a rule engine and hybrid reasoning. The method acquires multi-dimensional risk point data including structured numerical data and unstructured text data; inputs the structured numerical data into a preset rule engine to output rule scores and hit information; determines the correlation between the two types of data according to the dimensions of probability of occurrence and degree of impact; constructs a reasoning prompt input large language model based on the unstructured text data and the associated rule scores and hit information for semantic reasoning and outputs a semantic score; performs aggregation strategies on each dimension of the score to obtain the probability of occurrence score and degree of impact score; and finally, fuses and determines the risk assessment level. This invention can be applied to business scenarios such as fintech. By constraining semantic reasoning through rule scores and hit information, it achieves bidirectional coupling between the rule engine and the large language model, effectively improving the accuracy and interpretability of risk assessment.

[0044] In one embodiment, such as Figure 3 As shown, step S202 includes: S301. Extract historical occurrence frequency data for multiple time periods from the structured numerical data; S302. Input the historical occurrence frequency data into the preset rule engine, and output the historical frequency score and the corresponding first hit rule according to the priority of near and far time.

[0045] In this embodiment, multi-time-period fields related to historical occurrence frequency are extracted from structured numerical data, including the number of occurrences in the past year, the number of occurrences in the past two years, and the number of occurrences in the past three years. These fields record the historical trigger counts of risk points within different time windows, constituting multi-time-period historical occurrence frequency data, which is used to reflect the historical activity level and time trend of risk points.

[0046] Historical frequency data is input into a preset rule engine. Preset rules are matched based on their time priority; that is, the rule engine contains historical frequency scoring rules, with rules prioritized from most recent to oldest time period. The frequency scoring rules are sorted according to their time priority, with the highest priority for the past year's occurrences, followed by the past two years, and the lowest for the past three years. The values ​​of each time period field are compared sequentially with the thresholds in the corresponding priority frequency scoring rules. When a priority rule condition is met, matching of subsequent lower-priority rules is terminated, and the rule that meets the condition is designated as the first hit rule. The scoring outputter outputs the score corresponding to this first hit rule as the historical frequency score.

[0047] In practice, a five-level matching rule is set according to time, from most recent to oldest, with the following priorities from highest to lowest: If the occurrence frequency in the past year is greater than or equal to 3, it corresponds to 5 points, indicating high frequency occurrence in the past year; if the occurrence frequency in the past year is greater than or equal to 1, it corresponds to 4 points, indicating an occurrence record in the past year; if the occurrence frequency in the past two years is greater than or equal to 1 and the occurrence frequency in the past year is 0, it corresponds to 3 points, indicating an occurrence in the past two years but not in the past year; if the occurrence frequency in the past three years is greater than or equal to 1 and the occurrence frequency in the past two years is 0, it corresponds to 2 points, indicating an occurrence in the past three years but not in the past two years; if none of the above conditions are met, it corresponds to 1 point, indicating no occurrence record in the past three years. Matching is performed in the above order. Once the highest priority rule is hit, matching of subsequent lower priority rules is terminated, and the corresponding score is output as a historical frequency score. The description of the currently hit rule condition is recorded as the first hit rule.

[0048] This embodiment matches historical frequency rules according to the priority of recent events, ensuring that the impact of recent risk events on the score is given higher weight. This makes the rule score more accurately reflect the timeliness of the risk point, better conforms to the time decay law of risk assessment, and improves the rationality and accuracy of the historical frequency dimension score.

[0049] In one embodiment, such as Figure 4 As shown, step S204 includes: S401. Obtain the risk description and risk factor text associated with the historical frequency data; S402. Using the historical frequency score and the first hit rule as the constraint context, and combining the risk description and risk factor text to construct reasoning prompts, the prompts are input into the large language model to perform semantic reasoning on the probability of future occurrence, and output the future probability score.

[0050] In this embodiment, based on the dimensional correlation determination results, risk descriptions and risk factor texts belonging to the same occurrence probability dimension as historical occurrence frequency data are extracted as the core evaluation objects for semantic reasoning under the occurrence probability dimension. When constructing reasoning prompts, the role settings of risk assessment experts and multi-level scoring standards are first established. Then, historical frequency scores and the first hit rule are written into the prompts as pre-constraint contexts, clearly indicating the existing objective historical scoring results of the large language model. This guides the large language model to analyze future occurrence trends based on risk factor text. The risk descriptions and risk factor texts are then filled into the prompt template and input into the large language model for semantic reasoning. The large language model comprehensively considers risk triggers, environmental changes, and historical patterns to determine the probability level of the future occurrence of the risk point and assign a corresponding score. A score of 5 represents a very high probability of occurrence, 4 represents a relatively high probability, 3 represents a certain probability, 2 represents a low probability, and 1 represents a very low probability. After obtaining the historical frequency score and the future probability score, the maximum value aggregation strategy is executed to obtain the occurrence probability score.

[0051] Preferably, after the large language model returns the results, it undergoes a three-level fault-tolerant parsing process before outputting the final future probability score. The first level is format cleaning, removing unstructured formatting elements such as Markdown formatting tags and code block identifiers from the returned content, extracting the plain text inference results. The second level is structured parsing, parsing the cleaned content in JSON format to extract the scoring fields and inference basis fields. The third level is score range validation, limiting the extracted score value to a preset range of 1-5 points, and correcting it by taking the extreme values ​​of the score and the upper and lower limits of the range to ensure that the output score is within a valid range. If any level of parsing fails or an anomaly occurs, a degradation strategy is triggered, and a preset default score of 1 point is output as the future probability score.

[0052] This embodiment injects objective scores of historical frequencies as constraints into semantic reasoning of future possibilities, enabling the reasoning of the large language model to be based on verified historical data. It not only utilizes the deep understanding of text semantics of the large language model, but also constrains the fluctuation range of semantic reasoning through the deterministic results of rule scoring, effectively reducing the subjective bias of pure semantic reasoning and improving the stability and accuracy of future possibility scoring.

[0053] In one embodiment, such as Figure 5 As shown, step S202 further includes: S501. Extract the number of entities affected and the scale of risk items from the structured numerical data; S502. Input the number of entities in the scope of influence into the preset rule engine, and output the scope of influence score and the corresponding second hit rule according to the priority of the regulatory level. S503. Input the risk item size data into the preset rule engine, and output the risk item quantity score and the corresponding third hit rule according to the quantity threshold and the preset rule.

[0054] In this embodiment, data on the number of entities related to the scope of impact is extracted from structured numerical data, such as data on the number of financial companies, key infrastructure, and business units; and data on the number of risk items related to the scale of risk items is also extracted. The data on the number of entities affected reflects the organizational and critical facility scope impacted by the risk point, while the data on the scale of risk items reflects the total number of monitoring items or assets involved in the risk point.

[0055] Furthermore, the rule engine employs two types of structured scoring rules for the impact level category, corresponding to the impact scope and risk item size dimensions, respectively. For the impact scope dimension, the scoring rules are prioritized according to the regulatory level of the entity, with higher-level entities having higher priority rules. First, the count values ​​of different types of entities within the impact scope are extracted as the number of entities within the impact scope. This data is then input into the rule engine and matched according to regulatory level priority, outputting the impact scope score and the corresponding second hit rule. For the risk item size dimension, the scoring rules are matched according to the quantity threshold, from high to low. The total number of risk items is extracted as the risk item size data. This data is then input into the rule engine and matched against multi-level quantity thresholds, outputting the risk item quantity score and the corresponding third hit rule.

[0056] In practice, the impact scope scoring rules can be set into five levels of matching rules from high to low according to the regulatory level. The priorities from high to low are as follows: if the number of affected financial companies reaches the full coverage threshold, it corresponds to 5 points; if the number of affected financial companies is greater than 0 and the number of critical infrastructure units is greater than or equal to 2, it corresponds to 4 points; if the number of affected financial companies is greater than 0 and the number of critical infrastructure units is 0, it corresponds to 3 points; if the number of affected financial companies is 0 and the number of business units is greater than 1, it corresponds to 2 points; if the number of affected financial companies is 0 and the number of business units is equal to 1, it corresponds to 1 point. The rule engine matches according to the above priorities in sequence. After hitting the highest priority rule, it outputs the impact scope score corresponding to the score and records the corresponding rule condition as the second hit rule.

[0057] The risk item quantity scoring rule sets five matching levels based on the quantity threshold, from high to low: 5 points for 1000 or more risk items, 4 points for 800 or more, 3 points for 500 or more, 2 points for 200 or more, and 1 point for less than 200. The rule engine compares the risk item size data with the above thresholds one by one. When the highest threshold is hit, it outputs the corresponding risk item quantity score and records the corresponding threshold condition as the third hit rule.

[0058] This embodiment clarifies the scoring criteria for different regulatory levels by matching the impact scope score with the regulatory level priority, ensuring that the impact weight of higher regulatory levels can be reflected first; and achieves a quantitative assessment of the risk volume by matching the risk item scale score with the quantity threshold. Together, these two aspects provide a reliable structured scoring basis for the impact degree dimension.

[0059] In one embodiment, such as Figure 6 As shown, step S204 further includes: S601. Obtain regulatory background text and historical event description text associated with the number of entities and the scale of risk items within the scope of influence; S602. Using the influence range score and the second hit rule as the constraint context, and combining the regulatory background text to construct reasoning prompts, input them into the large language model to perform semantic reasoning of the severity of regulation, and output the regulatory dimension score. S603. Using the risk item quantity score and the third hit rule as the constraint context, and combining the historical event description text to construct reasoning prompts, input them into the large language model to perform semantic reasoning of the historical event level, and output the historical event score.

[0060] In this embodiment, based on the dimensional correlation determination results, regulatory background information fields that are correlated with the number of entities affected are extracted from unstructured text data, and historical event information fields that are correlated with the scale of risk items are extracted. The regulatory background text includes descriptions of laws, regulations, regulatory requirements, or industry standards related to the risk point, while the historical event description text includes records and impact descriptions of past triggering events for this risk point or similar risk points.

[0061] Next, for the regulatory dimension, the impact range score and the second hit rule are used as constraint contexts. Combined with the regulatory background text, reasoning prompts are constructed to guide the large language model to analyze the requirement level and consequences of violations of regulatory provisions based on the objective impact range, and output a regulatory dimension score. Specifically, constraint statements are inserted into the reasoning prompt template. These statements include the numerical value of the impact range score and a description of the second hit rule upon which the score is based. For example, "According to the impact range rule, this risk point involves N entities, and the impact range score is X points. Based on this, and combined with the following regulatory background information, assess the severity of the regulation. The output score should refer to the organizational impact range reflected by X points." The regulatory background text is then embedded into the template as the assessment object, forming a regulatory assessment reasoning prompt. After receiving this reasoning prompt, the large language model performs semantic reasoning of the regulatory severity under the guidance of the constraint context, and parses and outputs the corresponding regulatory dimension score.

[0062] For the historical event dimension, the risk item quantity score and the third hit rule are used as constraint context. Combined with the historical event description text, reasoning prompts are constructed to guide the large language model in analyzing the actual impact level of historical events based on the objective risk volume, and outputting a historical event score. Similarly, constraint statements are inserted into the reasoning prompt template. These constraints include the numerical value of the risk item quantity score and the description of the third hit rule on which the score is based. For example, "According to the risk item size rule, this risk point involves N risk items, and the risk item quantity score is X points. Based on this, and combined with the following historical event information, assess the severity level of the historical event. The output score should refer to the risk exposure scale reflected by X points." After embedding the historical event description text into the template, a historical assessment reasoning prompt is formed and input into the large language model. Guided by the constraint context, the large language model performs semantic reasoning on the historical event level and outputs the corresponding historical event score. After obtaining the impact scope score, risk item quantity score, future probability score, regulatory dimension score, and historical event score, the maximum value aggregation strategy (max) is executed to obtain the impact degree score.

[0063] Similarly, the output results of the two semantic scores, regulatory dimension score and historical event score, also undergo a three-level fault-tolerant parsing process. That is, first, the format marks in the returned content are removed, then the structured parsing is performed to extract the scoring and reasoning basis, and finally the score is verified to be within the valid range of 1-5 points. When the parsing fails, a downgrade strategy is triggered. The specific three-level fault-tolerant parsing process is the same as that of the future possibility score, and will not be elaborated here.

[0064] This embodiment injects the corresponding structured scores into the semantic reasoning of regulatory and historical events, respectively, so that both the assessment of regulatory severity and the assessment of historical event level are subject to the prior constraints of the rule scores. This achieves bidirectional guidance of structured numerical scores on the semantic reasoning of unstructured text, and improves the objectivity and consistency between regulatory dimension scores and historical event scores.

[0065] In one embodiment, such as Figure 7 As shown, after step S205, the method further includes: S701. Perform a logical consistency check on the probability of occurrence score and the degree of impact score to confirm whether the probability of occurrence score and the degree of impact score satisfy the preset logical constraint relationship between dimensions. S702. When the logical constraint relationship between the dimensions is not satisfied, a review instruction is triggered and a logical exception prompt is output.

[0066] In this embodiment, a logical constraint relationship is pre-set between two dimensions: the probability of occurrence and the degree of impact. This defines a reasonable corresponding range between the probability of risk occurrence and the degree of impact, avoiding logical contradictions such as abnormally high or low scores for a single dimension. After obtaining the comprehensive score for both dimensions, the combined score is compared with the pre-set logical constraint relationship to verify whether it falls within a reasonable corresponding range.

[0067] Specifically, the logical constraints between dimensions can include the following: when the probability score is greater than or equal to 4 (extremely high probability), the impact score should not be less than 2 (at least affecting the management requirements of the professional company); otherwise, it is considered a logical anomaly. When the impact score is greater than or equal to 4 (may trigger group penalties or regulatory sanctions), the probability score should not be less than 2 (at least the probability of future occurrence is low); otherwise, it is considered a logical anomaly. Substituting the current probability score and impact score into the above constraints, a Boolean judgment is performed, and the corresponding judgment result is output.

[0068] If the score combination meets the constraint relationship, the logic is considered consistent, and the subsequent risk value fusion step continues; if it does not meet the constraint relationship, it is considered a logical anomaly, and a review instruction is immediately triggered while outputting a logical anomaly prompt. This review instruction can be sent to the risk control review platform through the message queue to prompt human risk control experts to review the data quality or scoring logic of the current risk point.

[0069] This embodiment verifies the logical consistency of the probability score and the impact score by performing logical consistency constraint checks. It promptly identifies and marks abnormal score combinations that do not conform to business logic, avoiding score distortion caused by data errors or model misjudgments, and ensuring the logical consistency and reliability of risk assessment results.

[0070] In one embodiment, such as Figure 8 As shown, step S206 includes: S801. The probability score and the impact score are fused together to obtain the corresponding total risk value; S802. Compare the total risk value with multi-level preset thresholds, confirm the risk assessment level based on the comparison results, and generate control recommendations for the current risk assessment level.

[0071] In this embodiment, when the probability score and the impact score are fused, a product fusion method can be used to calculate the probability score and the impact score to obtain the total risk value. Of course, in other embodiments, the total risk value can also be calculated by other fusion methods such as weighted fusion, but this embodiment does not limit this.

[0072] The total risk value comprehensively reflects the probability of risk occurrence and the intensity of its impact. It is compared with multiple preset thresholds, i.e., pre-defined threshold ranges corresponding to multiple risk levels (e.g., high, medium, and low risk), each corresponding to a continuous score range. The calculated total risk value is compared with each threshold to determine its corresponding score range, thus matching the appropriate risk assessment level. Simultaneously, preset control strategies are automatically matched based on the risk level, generating corresponding risk management recommendations and clarifying the handling requirements and priorities for different risk levels.

[0073] In practice, the total risk value ranges from 1 to 25. The corresponding risk level thresholds are as follows: when the total risk value is between 16 and 25, it is considered high risk, and the corresponding control recommendation is "immediate measures need to be taken to control it"; when the total risk value is between 6 and 15, it is considered medium risk, and the corresponding control recommendation is "measures need to be developed to control it gradually"; when the total risk value is between 1 and 5, it is considered low risk, and the corresponding control recommendation is "acceptable but requires continuous monitoring".

[0074] This embodiment achieves a quantitative coupling of the probability of occurrence and the degree of impact by merging and calculating the total risk value. At the same time, by matching and generating control recommendations through multi-level thresholds, it can output assessment results that can be directly used for business decisions, thereby improving the practicality of risk assessment results.

[0075] It should be noted that there is no necessary order between the above steps. Those skilled in the art will understand from the description of the embodiments of the present invention that the above steps may have different execution orders in different embodiments, that is, they may be executed in parallel or in turn, etc.

[0076] Further reference Figure 9 As a response to the above Figure 2The present invention provides an embodiment of a risk assessment device based on a rule engine and hybrid reasoning, which implements the method shown. Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0077] like Figure 9 As shown, the risk assessment device 90 based on rule engine and hybrid reasoning described in this embodiment includes: The data acquisition module 901 is used to acquire multidimensional data of the risk points to be assessed, including structured numerical data and unstructured text data. The rule scoring module 902 is used to input the structured numerical data in the multidimensional data into a preset rule engine, and output at least one rule score and rule hit information through corresponding threshold matching and rule priority determination; The association segmentation module 903 is used to determine the association between various structured numerical data and unstructured text data according to the dimensions of probability of occurrence and degree of influence. The semantic scoring module 904 is used to construct corresponding reasoning prompts based on the rule scores and rule hit information of the unstructured text data, the associated structured numerical data, and input them into the large language model for semantic reasoning, and output at least one semantic score. The rating aggregation module 905 is used to perform aggregation strategies on the rating results of each dimension according to the probability of occurrence dimension and the degree of impact dimension, to obtain the probability of occurrence rating and the degree of impact rating. The risk fusion and determination module 906 is used to fuse risk values ​​and determine risk levels based on the probability of occurrence score and the degree of impact score to obtain the corresponding risk assessment level.

[0078] The modules referred to in this invention are a series of computer program instruction segments capable of performing specific functions. They are more suitable than programs for describing the risk assessment execution process based on rule engines and hybrid reasoning. For specific implementation methods of each module, please refer to the corresponding method embodiments described above, which will not be repeated here.

[0079] In one embodiment, the rule scoring module 902 includes: The first data extraction unit is used to extract historical occurrence frequency data for multiple time periods from the structured numerical data; The first rule matching unit is used to input the historical occurrence frequency data into the preset rule engine, match the preset rules according to the priority of near and far time, and output the historical frequency score and the corresponding first hit rule.

[0080] In one embodiment, the semantic scoring module 904 includes: The first association acquisition unit is used to acquire risk descriptions and risk factor texts associated with the historical occurrence frequency data; The first semantic reasoning unit is used to take the historical frequency score and the first hit rule as a constraint context, combine the risk description and risk factor text to construct reasoning prompts, and then input them into the large language model to perform semantic reasoning on the future probability of occurrence, and output a future possibility score.

[0081] In one embodiment, the rule scoring module 902 includes: The second data extraction unit is used to extract data on the number of entities affected and the scale of risk items from the structured numerical data. The second rule matching unit is used to input the number of subjects in the scope of influence into the preset rule engine, match preset rules according to the priority of the regulatory level, and output the scope of influence score and the corresponding second hit rule; The third rule matching unit is used to input the risk item size data into the preset rule engine, match the preset rules according to the quantity threshold, and output the risk item quantity score and the corresponding third hit rule.

[0082] In one embodiment, the semantic scoring module 904 includes: The second association acquisition unit is used to acquire regulatory background text and historical event description text associated with the number of entities and the scale of risk items in the scope of influence. The second semantic reasoning unit is used to take the influence range score and the second hit rule as a constraint context, combine them with the regulatory background text to construct reasoning prompts, and then input them into the large language model to perform semantic reasoning of the severity of regulation, and output the regulatory dimension score. The third semantic reasoning unit is used to take the risk item quantity score and the third hit rule as the constraint context, combine them with the historical event description text to construct reasoning prompts, and then input them into the large language model to perform semantic reasoning of the historical event level and output the historical event score.

[0083] In one embodiment, the device 90 further includes: The consistency verification module is used to perform logical consistency verification on the probability of occurrence score and the degree of impact score to confirm whether the probability of occurrence score and the degree of impact score meet the preset logical constraint relationship between dimensions. The review prompt module is used to trigger a review instruction and output a logical exception prompt when the logical constraint relationship between the dimensions is not satisfied.

[0084] In one embodiment, the risk fusion determination module 906 includes: The fusion calculation unit is used to fuse the probability of occurrence score and the degree of impact score to obtain the corresponding total risk value; The risk classification unit is used to compare the total risk value with multiple preset thresholds, confirm the risk assessment level based on the comparison results, and generate control recommendations for the current risk assessment level.

[0085] In the above embodiments, this invention discloses a risk assessment device based on a rule engine and hybrid reasoning. It acquires multi-dimensional risk point data including structured numerical data and unstructured text data; inputs the structured numerical data into a preset rule engine to output rule scores and hit information; determines the correlation between the two types of data according to the dimensions of probability of occurrence and degree of impact; constructs a reasoning prompt input large language model based on the unstructured text data and the associated rule scores and hit information for semantic reasoning and outputs a semantic score; performs an aggregation strategy on each dimension score to obtain a probability of occurrence score and a degree of impact score; and finally, fuses and determines the risk assessment level. This invention can be applied to business scenarios such as fintech. By constraining semantic reasoning through rule scores and hit information, it achieves bidirectional coupling between the rule engine and the large language model, effectively improving the accuracy and interpretability of risk assessment.

[0086] Specific limitations regarding the risk assessment device based on rule engines and hybrid inference can be found in the limitations of the risk assessment method based on rule engines and hybrid inference mentioned above, and will not be repeated here. Each module in the aforementioned risk assessment device based on rule engines and hybrid inference can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0087] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0088] Another embodiment of the present invention provides a computer device, such as... Figure 10 As shown, the computer device 100 includes: One or more processors 1001 and memory 1002, Figure 10 The following description uses a processor 1001 as an example. The processor 1001 and the memory 1002 can be connected via a bus or other means. Figure 10 Taking the example of a connection between China and Israel via a bus.

[0089] The processor 1001 is used to perform various control logics of the computer device 100. It can be any conventional processor, microprocessor, state machine, general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), microcontroller, ARM (Acorn RISC Machine) or other programmable logic device, discrete gate or transistor logic, discrete hardware components or any combination of these components.

[0090] The memory 1002, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions corresponding to the risk assessment method based on rule engines and hybrid reasoning in the embodiments of the present invention. The processor 1001 executes various functional applications and data processing of the computer device 100 by running the non-volatile software programs, instructions, and units stored in the memory 1002, thereby implementing the risk assessment method based on rule engines and hybrid reasoning in the above method embodiments.

[0091] Another embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by one or more processors, perform the steps of the risk assessment method based on rule engine and hybrid inference in any of the above method embodiments.

[0092] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0093] Based on the above description of the embodiments, those skilled in the art will understand that the methods described in the embodiments can be implemented using software plus necessary general-purpose hardware platforms. Of course, they can also be implemented using hardware, but in many cases, the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a 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 device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0094] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The computer program can be stored in a non-volatile, computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The storage medium can be a memory, magnetic disk, floppy disk, flash memory, optical storage, etc.

[0095] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0096] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A risk assessment method based on rule engine and hybrid reasoning, characterized in that, include: Obtain multidimensional data of the risk points to be assessed, including structured numerical data and unstructured text data; The structured numerical data in the multidimensional data is input into a preset rule engine, and at least one rule score and rule hit information are output through corresponding threshold matching and rule priority determination. Determine the correlation between various structured numerical data and unstructured text data based on the dimensions of probability of occurrence and degree of impact; Based on the rule scores and rule hit information of the unstructured text data, the associated structured numerical data, the corresponding reasoning prompts are constructed and input into the large language model for semantic reasoning, and at least one semantic score is output. The scoring results of each dimension are aggregated according to the probability of occurrence dimension and the degree of impact dimension to obtain the probability of occurrence score and the degree of impact score. The risk value is fused and the level is determined based on the probability score and the impact score to obtain the corresponding risk assessment level.

2. The risk assessment method based on rule engine and hybrid reasoning according to claim 1, characterized in that, The step of inputting structured numerical data from the multidimensional data into a preset rule engine, and outputting at least one rule score and rule hit information through corresponding threshold matching and rule priority determination, includes: Extract historical occurrence frequency data for multiple time periods from the structured numerical data; The historical frequency data is input into the preset rule engine, and the preset rules are matched according to the priority of near and far time to output the historical frequency score and the corresponding first hit rule.

3. The risk assessment method based on rule engine and hybrid reasoning according to claim 2, characterized in that, The process involves constructing corresponding reasoning prompts based on the unstructured text data, rule scores of associated structured numerical data, and rule hit information, and inputting these prompts into a large language model for semantic reasoning, outputting at least one semantic score, including: Obtain the risk description and risk factor text associated with the historical frequency data; The historical frequency score and the first hit rule are used as the constraint context. After combining the risk description and risk factor text to construct reasoning prompts, they are input into the large language model to perform semantic reasoning on the probability of future occurrence and output a future probability score.

4. The risk assessment method based on rule engine and hybrid reasoning according to claim 1, characterized in that, The step of inputting structured numerical data from the multidimensional data into a preset rule engine, and outputting at least one rule score and rule hit information through corresponding threshold matching and rule priority determination, includes: Extract the number of entities affected and the scale of risk items from the structured numerical data; The number of entities within the scope of influence is input into the preset rule engine, and the preset rules are matched according to the priority of the regulatory level to output the scope of influence score and the corresponding second hit rule; The risk item size data is input into the preset rule engine, and the preset rule is matched according to the quantity threshold to output the risk item quantity score and the corresponding third hit rule.

5. The risk assessment method based on rule engine and hybrid reasoning according to claim 4, characterized in that, The process involves constructing corresponding reasoning prompts based on the unstructured text data, rule scores of associated structured numerical data, and rule hit information, and inputting these prompts into a large language model for semantic reasoning, outputting at least one semantic score, including: Obtain regulatory background text and historical event description text associated with the number of entities affected and the scale of risk items within the scope of influence; The impact range score and the second hit rule are used as the constraint context. After combining the regulatory background text to construct reasoning prompts, they are input into the large language model to perform semantic reasoning of the severity of regulation and output the regulatory dimension score. The risk item quantity score and the third hit rule are used as the constraint context. After combining the historical event description text to construct reasoning prompts, they are input into the large language model to perform semantic reasoning of the historical event level and output the historical event score.

6. The risk assessment method based on rule engine and hybrid reasoning according to claim 1, characterized in that, After performing aggregation strategies on the scoring results of each dimension according to the probability of occurrence dimension and the degree of impact dimension to obtain the probability of occurrence score and the degree of impact score, the method further includes: A logical consistency check is performed on the probability of occurrence score and the degree of impact score to confirm whether the probability of occurrence score and the degree of impact score satisfy the preset logical constraint relationship between dimensions. When the logical constraints between the dimensions are not met, a review instruction is triggered and a logical exception message is output.

7. The risk assessment method based on rule engine and hybrid reasoning according to claim 1, characterized in that, The process of fusing and determining the risk level based on the probability score and the impact score to obtain the corresponding risk assessment level includes: The probability score and the impact score are fused together to obtain the corresponding total risk value; The total risk value is compared with multiple preset thresholds. The risk assessment level is confirmed based on the comparison results, and control recommendations for the current risk assessment level are generated accordingly.

8. A risk assessment device based on a rule engine and hybrid reasoning, characterized in that, include: The data acquisition module is used to acquire multidimensional data of the risk points to be assessed, including structured numerical data and unstructured text data. The rule scoring module is used to input the structured numerical data in the multidimensional data into the preset rule engine, and output at least one rule score and rule hit information through corresponding threshold matching and rule priority determination; The association segmentation module is used to determine the associations between various structured numerical data and unstructured text data according to the dimensions of probability of occurrence and degree of impact. The semantic scoring module is used to construct corresponding reasoning prompts based on the rule scores and rule hit information of the unstructured text data, the associated structured numerical data, and input them into the large language model for semantic reasoning, and output at least one semantic score. The rating aggregation module is used to perform aggregation strategies on the rating results of each dimension according to the probability of occurrence dimension and the degree of impact dimension, to obtain the probability of occurrence rating and the degree of impact rating. The risk fusion and determination module is used to fuse risk values ​​and determine risk levels based on the probability of occurrence score and the degree of impact score, so as to obtain the corresponding risk assessment level.

9. A computer device, characterized in that, Includes at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the risk assessment method based on rule engine and hybrid reasoning as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the risk assessment method based on rule engine and hybrid reasoning as described in any one of claims 1-7.

Images