Rule-constrained decision generation method of ai-based centralized procurement governance hub

Abstract

The application discloses a rule constraint decision generation method of an AI-based centralized procurement governance hub. The method comprises the following steps: receiving a procurement governance task, and obtaining a rule constraint set related to the procurement governance task; determining a text generation format based on the rule constraint set, and calling a decision content generation model to generate an initial decision text under the constraints of the rule constraint set and the text generation format; performing multi-dimensional information checking on the initial decision text based on the rule constraint set, and performing rollback and regeneration operations on the initial decision text that fails to pass the checking to obtain a corrected decision text; associating and storing the decision text that has passed the checking and correction with corresponding rule application paths, data source information and multi-dimensional checking information to form traceable decision records; and generating a procurement governance decision report based on the decision records. The above method can ensure the reliability, auditability and traceability of the procurement governance decision report, and reduce procurement governance risks.

Description

Technical Field

[0001] This application relates to the field of centralized procurement governance technology, and in particular to a rule-constraint decision generation method for a centralized procurement governance hub based on AI. Background Technology

[0002] In centralized procurement governance scenarios, with the continuous expansion of business scale and the increasing demands for governance, leveraging artificial intelligence technology to assist in the generation of governance decision reports has become an important technical means to improve report generation efficiency. Generally speaking, governance decision reports need to meet certain reporting formats, compliance requirements, and cross-departmental sharing requirements, and their generation quality directly affects the standardized execution and management efficiency of procurement activities.

[0003] In existing technologies, artificial intelligence can be used to generate text content in a specific format based on the procurement task description, which can then be manually adjusted to form a corresponding governance decision report. However, decision reports generated in this way often fail to match complex procurement governance rules, and the output is prone to deviating from compliance boundaries, requiring significant manual adjustments later. Furthermore, the report content often lacks supporting data or contains logically contradictory or false information, resulting in poor auditability and traceability, and easily leading to governance risks. Summary of the Invention

[0004] In view of this, the embodiments of this application provide a rule constraint decision generation method for a centralized procurement governance center based on AI. The main purpose is to solve the technical problems that existing governance decision report generation methods are prone to causing report content to deviate from compliance boundaries and have poor reliability, auditability and traceability of report content.

[0005] According to one aspect of this application, a rule-constraint decision generation method for an AI-based centralized procurement governance center is provided, the method being executed within the centralized procurement governance center, the method comprising: Receive procurement governance tasks and obtain a set of rules and constraints related to the procurement governance tasks, wherein the set of rules and constraints includes governance objective rules, compliance boundary rules, and business execution rules; The text generation format is determined based on the set of rules and constraints, and the decision content generation model is invoked to generate initial decision text under the constraints of the set of rules and constraints and the text generation format. Based on the set of rules and constraints, the initial decision text is verified in multiple dimensions, and the initial decision text that fails the verification is rolled back and regenerated to obtain the corrected decision text. The verified and corrected decision text is associated and stored with the corresponding rule application path, data source information, and multi-dimensional verification information to form a traceable decision record. A procurement governance decision report for the procurement governance task is generated based on the traceable decision records.

[0006] Optionally, the step of determining the text generation format based on the rule constraint set and calling the decision content generation model to generate initial decision text under the constraints of the rule constraint set and the text generation format includes: parsing the rule constraint set and extracting governance target rules, compliance boundary rules, and business execution rules from the rule constraint set; generating structured generation instructions for controlling text generation based on the governance target rules, the compliance boundary rules, and the business execution rules; and calling the decision content generation model to generate initial decision text whose text content and text format both conform to the rule requirements under the constraints of the structured generation instructions.

[0007] Optionally, the step of generating a structured generation instruction for controlling text generation based on the governance objective rules, the compliance boundary rules, and the business execution rules includes: performing intent recognition on the governance objective rules to obtain the core conclusions and evaluation dimensions that the procurement governance task needs to achieve; extracting conditions from the compliance boundary rules to obtain the compliance statements and expression restrictions that the decision text needs to follow; logically deconstructing the business execution rules to obtain the decision logic, paragraph structure, and paragraph order of the decision text; and parameterizing and encapsulating the core conclusions, evaluation dimensions, compliance statements, expression restrictions, decision logic, paragraph structure, and paragraph order to obtain the structured generation instruction.

[0008] Optionally, the step of performing multi-dimensional information verification on the initial decision text based on the rule constraint set includes: comparing and verifying the entity information, data information, and legal clauses referenced in the initial decision text with a preset knowledge database or a certified external data source to obtain an authenticity verification result; checking whether the core conclusions in the initial decision text are consistent with the governance objective rules in the rule constraint set to obtain a governance objective verification result; checking whether the analysis and reasoning content in the initial decision text are consistent with the business execution rules in the rule constraint set to obtain a logical consistency verification result; and determining whether the decision suggestions and conclusions in the initial decision text are within the compliance boundary defined by the compliance boundary rules in the rule constraint set to obtain a compliance verification result.

[0009] Optionally, the step of rolling back and regenerating the initial decision text that failed verification to obtain the corrected decision text includes: automatically triggering a text correction mechanism when the initial decision text fails multi-dimensional information verification, wherein the text correction mechanism includes at least one of the following: feeding back the text that failed verification and the reason for failure to the decision content generation model, so that the decision content generation model regenerates part or all of the text based on the rule constraint set; adjusting the constraint parameters in the generation instruction according to the verification result, and controlling the decision content generation model to regenerate part or all of the text based on the rule constraint set based on the adjusted generation instruction; generating a verification anomaly report and submitting the verification anomaly report to the manual review interface to receive the manually corrected text returned by the manual review interface; obtaining the corrected decision text based on the regenerated part or all of the text and the manually corrected text.

[0010] Optionally, obtaining the set of rule constraints related to the procurement governance task includes: retrieving a subset of rule constraints corresponding to the procurement governance task from the rule management module of the centralized procurement governance hub according to the business domain involved in the procurement governance task; obtaining the hierarchical path identifier corresponding to each rule constraint in the subset of rule constraints based on rule trees pre-constructed for multiple business modules, wherein the rule tree includes upper-level governance target nodes, middle-level rule boundary nodes, and lower-level business rule nodes; and attaching the corresponding hierarchical path identifier to each rule constraint in the subset of rule constraints to obtain the set of rule constraints related to the procurement governance task.

[0011] Optionally, obtaining the hierarchical path identifier corresponding to each rule constraint in the rule constraint subset based on rule trees pre-constructed for multiple business modules includes: determining at least one business module related to the procurement governance task according to the governance objectives and business scope of the procurement governance task; extracting a partial node set related to the procurement governance task from the rule tree corresponding to each business module, wherein the partial node set includes at least one upper-level governance target node and a lower-level business rule node associated with the upper-level governance target node; and determining the hierarchical path identifier of each rule constraint in the rule tree corresponding to the target business module based on the partial node set.

[0012] Optionally, the step of associating and storing the verified and corrected decision text with the corresponding rule application path, data source information, and multi-dimensional information verification results to form a traceable decision record includes: generating a rule application path corresponding to the decision text based on the hierarchical path identifier of each rule constraint called during the initial decision text generation and correction process and the rule tree in which the rule constraint is located; obtaining the data source information corresponding to the decision text based on the source data identifier and data generation time of each assertion and data basis in the initial decision text; obtaining the multi-dimensional verification information corresponding to the decision text based on the multi-dimensional verification results, multi-dimensional verification time, and verification execution unit identifier of the initial decision text; and associating and encapsulating the verified and corrected decision text, the rule application path, the data source information, and the multi-dimensional verification information to form the traceable decision record.

[0013] Optionally, generating the procurement governance decision report for the procurement governance task based on the traceable decision records includes: integrating the traceable decision records through a preset presentation format to generate the procurement governance decision report, wherein the procurement governance decision report includes formatted decision text and supplementary files containing the rule application path, the data source information, and the multi-dimensional verification information.

[0014] Optionally, the method further includes: parsing the rule application path corresponding to the decision text to identify the node position and level of the rule constraints applied by the decision text in the rule tree; generating a decision logic explanation text based on the node position and level, the decision logic explanation text being used to explain the rules on which the decision points in the decision text are based and the business logic positioning of the rules on which they are based; and outputting the decision logic explanation text as an independent chapter of the procurement governance decision report or as part of an appendix to the procurement governance decision report.

[0015] By employing the above technical solutions, this application provides an AI-based rule-constrained decision generation method for centralized procurement governance. This method acquires a set of rule constraints related to procurement governance tasks, determines a text generation format based on the rule constraint set, and calls a model to generate initial decision text under the constraints of the rule constraint set and text generation format. The initial decision text is then subjected to multi-dimensional verification, and text that fails verification is rolled back and regenerated. Finally, a traceable decision record is generated based on the verified and corrected decision text, and these records are combined to form a procurement governance decision report for the procurement governance task. This method can effectively identify and correct potential errors, contradictions, and out-of-bounds risks in the report, thereby ensuring the reliability of the decision content. Furthermore, by associating and storing the decision text with the corresponding rule application path, data source information, and multi-dimensional verification information, the auditability and traceability of the procurement governance decision report can be ensured, thereby reducing procurement governance risks. By generating procurement governance decision reports that conform to relevant rule constraints, format constraints, and compliance boundaries, the above method can significantly improve the report output efficiency and quality of the centralized procurement governance center, and enhance the reliability, auditability, and traceability of the report content.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 The illustration shows a flowchart of a rule constraint decision generation method for a centralized procurement governance center based on AI, as provided in an embodiment of this application. Detailed Implementation

[0018] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0019] In one embodiment, such as Figure 1 As shown, an AI-based rule-constraint decision generation method for a centralized procurement governance center is provided. Taking the application of this method to a procurement governance center as an example, the method includes the following steps: Step 101: Receive the procurement governance task and obtain the set of rules and constraints related to the procurement governance task. The set of rules and constraints includes governance objective rules, compliance boundary rules, and business execution rules.

[0020] Among them, procurement governance tasks refer to specific business requests that need to be processed by the centralized procurement governance center, such as conducting risk assessments on a procurement application or conducting compliance reviews on a supplier list; the rule constraint set refers to a set of constraints extracted from the rule management module to ensure the quality of decision generation. Its core components include: governance objective rules, which are the core conclusions and evaluation criteria that the procurement governance tasks need to achieve; compliance boundary rules, which are the laws, regulations and internal mandatory standards that the procurement governance tasks must comply with; and business execution rules, which are the specific operational procedures and judgment logic criteria that guide the procurement business.

[0021] Specifically, the system can receive procurement governance tasks submitted from upstream processes or users through a pre-defined system interface. It then parses the business type of the task and initiates a query request to the rule management module of the centralized procurement governance hub based on the identified business type. The rule management module can retrieve all relevant rule nodes from its stored module-level rule tree based on the business type, extract specific rule clauses from these nodes, and then categorize them into three types: governance objective rules, compliance boundary rules, and business execution rules, thereby assembling a set of rule constraints related to the procurement governance task. Alternatively, in other optional implementations, the set of rule constraints can also be obtained and assembled through other methods such as calling external rule service interfaces or querying shared knowledge graphs. Obtaining the set of rule constraints provides a clear rule basis for the generation of decision texts and the verification of multi-dimensional information.

[0022] Step 102: Determine the text generation format based on the set of rule constraints, and call the decision content generation model to generate the initial decision text under the constraints of the set of rule constraints and the text generation format.

[0023] Among them, the text generation format refers to the mandatory requirements for the structure and expression of the output text determined by the set of rules and constraints. These mandatory requirements may include requirements for various dimensions such as text paragraphs, text length, logical order, and output granularity. The decision content generation model refers to an artificial intelligence model that can understand and integrate governance objective rules, compliance boundary rules, and business execution rules, and can simultaneously meet the requirements for content compliance and format specification during the text generation process, and directly output the initial decision text.

[0024] Specifically, upon receiving the procurement governance task and its corresponding set of rules and constraints, the set of rules and constraints is first parsed, and the governance objective rules, compliance boundary rules, and business execution rules are extracted. Next, the various rules in the set are merged and transformed into structured generation instructions. These instructions integrate content generation logic and format generation logic, driving the model to create content. Subsequently, the decision content generation model performs end-to-end text generation based on these instructions. The generation process is constrained in real-time by the rules embedded within the instructions to ensure that the output text conforms to preset format and content requirements in terms of paragraph structure, logical order, content granularity, and key expressions, without requiring subsequent formatting and content polishing, ultimately forming a rule-driven initial decision text. In this embodiment, the set of rules and constraints is the sole source of the generation control instructions and is directly embedded as a core constraint factor, dominating the text generation process. In optional implementations, the decision content generation model can adopt a single-agent structure or a multi-agent structure where multiple dedicated models collaborate under a unified generation control instruction schedule.

[0025] Step 103: Based on the rule constraint set, perform multi-dimensional information verification on the initial decision text, and perform rollback and regeneration operations on the initial decision text that fails the verification to obtain the corrected decision text.

[0026] Specifically, after generating the initial decision text, it can be verified in multiple dimensions, such as authenticity verification, governance objective verification, logical consistency verification, and compliance verification, based on the set of rule constraints. For authenticity verification, key entity information, specific numerical information, and cited policy and regulatory clauses mentioned in the initial decision text can be extracted and automatically compared with authoritative internal and external databases, certified data sources, and the latest regulatory text library to verify their authenticity, accuracy, and timeliness. For governance objective verification, the core conclusions in the initial decision text can be extracted and verified using the core objectives and evaluation criteria defined in the governance objective rules to verify whether the core conclusions in the initial decision text have achieved the established core objectives. For logical consistency verification, the core arguments and reasoning processes in the initial decision text can be parsed and formally matched with the business execution rules in the rule constraint set to check for any statements that violate or contradict the rules. In addition, it can be checked whether the text structure of the initial decision text matches the predefined text structure in the business execution rules, whether the order of each paragraph is correct, and whether anything is missing. For compliance verification, the final decision recommendations or action plans in the initial decision text can be evaluated to determine whether all their elements fall within the various compliance boundary rules defined in the rule constraint set, ensuring that there is no risk of exceeding the boundaries.

[0027] Furthermore, if any of the above-mentioned dimension verifications fails, a rollback and regeneration mechanism is automatically triggered. This rollback and regeneration mechanism can include at least one of the following processing methods: feeding back the problematic text fragment and related contextual information to the decision content generation model for partial rewriting; adjusting the constraint parameters in the generation instruction in reverse according to the verification results, and then driving the decision content generation model to regenerate part or the entire text; when key compliance risks are involved, submitting the problematic text for manual review and receiving the manually corrected text. By performing multi-dimensional verification on the initial decision text and rolling back and regenerating texts that fail verification, it is possible to effectively screen and correct potential false information, logical errors, and compliance risks in the decision text, thereby effectively improving the credibility and reliability of the decision content and enhancing the auditability of the report content.

[0028] Step 104: The verified and / or corrected decision text is associated and stored with the corresponding rule application path, data source information and multi-dimensional verification information to form a traceable decision record.

[0029] Specifically, after multi-dimensional information verification and correction of the initial decision text, the unique identifiers of each rule constraint called by the decision content generation model when generating the initial decision text can be extracted from the system logs. These identifiers can then be used to query the rule management module to obtain the complete hierarchical path of these rules within their respective module-level rule tree, thus generating a clear rule application path. Next, the decision text can be parsed, and key conclusions and data references can be annotated, linked to the unique identifiers of the original datasets upon which these conclusions depend, as well as the data acquisition time, thus forming data source information. Furthermore, detailed results of the initial decision text's authenticity verification, governance objective verification, logical consistency verification, and compliance verification can be recorded, including whether the verification passed, the verification time, the identifier of the service unit responsible for the verification, the type of rollback operation triggered, and the number of regenerations, integrating this information into multi-dimensional verification information. Finally, the verified and modified decision text, rule application path, data source information, and multi-dimensional verification information can be bound together and stored using a unique decision record identifier. In other alternative implementations, the associated storage may also employ blockchain technology to enhance the immutability of records, or use knowledge graphs for association to support more complex traceability queries.

[0030] Step 105: Generate a procurement governance decision report for the procurement governance task based on traceable decision records.

[0031] Specifically, after generating traceable decision records, the decision text portion of the records can be formatted into a formal document body according to a predetermined format, ensuring that its font, paragraphs, headings, etc., conform to relevant standards. Then, the rule application paths, data source information, and multi-dimensional verification information from the decision records can be extracted and organized, and appended to the document body or embedded in relevant locations as appendices, footnotes, or independent attachments. Finally, the formatted document body and traceability information attachments can be merged to output a complete procurement governance decision report, which can be used for approval, archiving, or distribution to relevant stakeholders. In an optional implementation, the report can also be presented in an interactive webpage format, allowing users to click on decision points to view the underlying rules and data sources, or generating report versions with varying levels of detail based on the recipient's role.

[0032] The above methodology will now be illustrated with a specific example. Suppose the procurement governance task involves a risk assessment of server procurement for a new data center. In this scenario, the relevant set of rules and constraints can first be obtained from the rule management module of the centralized procurement governance hub. For example, governance objective rules may include "outputting a clear risk level conclusion and its main basis," compliance boundary rules may include "the conclusion must not violate the mandatory clauses in the data center energy efficiency standards," and business execution rules may include "the report must include three core paragraphs: technical risk, supplier risk, and cost risk, with technical risk being analyzed first." Subsequently, the text generation format can be determined based on these rules, and the decision content generation model can be invoked. Guided by structured instructions incorporating the aforementioned rule constraints, the model generates an initial decision text containing the paragraphs required by the business execution rules and the core conclusions. Next, the initial decision text can be validated using multiple dimensions based on the set of rule constraints. For example, after authenticity verification, it is found that the power consumption data of a certain server model cited in the text deviates from the latest value in an authoritative database; after logical consistency verification, it is found that the "high-risk" conclusion in the text conflicts with the statement "the supplier has excellent qualifications" in the analysis content. For text that fails validation, a rollback mechanism can be automatically triggered, and the specific issues will be fed back to the decision content generation model for regeneration, resulting in a corrected decision text. Then, the corrected decision text, along with the paths of the rules invoked during its generation (i.e., rule application paths), the referenced power consumption data source identifiers (i.e., data source information), and the issues discovered and correction records (i.e., multi-dimensional validation information), can be stored together to form a traceable decision record. Finally, based on this decision record containing generation, validation, and traceability information, a reliable, auditable, and traceable procurement governance decision report can be generated.

[0033] The technical solution of this embodiment obtains a set of rule constraints related to the procurement governance task, determines the text generation format based on the set of rule constraints, and calls a model to generate initial decision text under the constraints of the set of rule constraints and the text generation format. The initial decision text is then subjected to multi-dimensional verification, and text that fails verification is rolled back and regenerated. Based on the verified and corrected decision text, a traceable decision record is generated, and finally, a procurement governance decision report for the procurement governance task is formed. This effectively identifies and corrects potential errors, contradictions, and risks of exceeding boundaries in the report, thereby ensuring the reliability of the decision content. Furthermore, by associating and storing the decision text with the corresponding rule application path, data source information, and multi-dimensional verification information, the auditability and traceability of the procurement governance decision report can be ensured, thereby reducing procurement governance risks. The above method, by generating procurement governance decision reports that conform to relevant rule constraints, format constraints, and compliance boundaries, can significantly improve the report output efficiency and quality of the centralized procurement governance center, and enhance the reliability, auditability, and traceability of the report content.

[0034] In one embodiment, step 102 can be implemented by the following method: First, the rule constraint set is parsed, and governance target rules, compliance boundary rules, and business execution rules are extracted from the rule constraint set. Then, based on the governance target rules, compliance boundary rules, and business execution rules, a structured generation instruction for controlling text generation is generated. Finally, the decision content generation model is called to generate initial decision text whose text content and text format both meet the rule requirements under the constraints of the generation instruction.

[0035] In this embodiment, after receiving the rule constraint set corresponding to the procurement governance task and the procurement instruction task, the rule constraint set is first parsed to distinguish various types of rules, including governance objective rules, compliance boundary rules, and business execution rules. Then, these rules are merged and transformed into a unified, structured generation instruction. Subsequently, the decision content generation model is invoked, and this structured generation instruction is input into the model. This allows the generation instruction to comprehensively constrain the text generation process from macro-structure to micro-expression, ultimately driving the model to output a formatted and compliant initial decision text without requiring independent formatting.

[0036] This embodiment drives the text generation process based on a set of rule constraints, which allows for simultaneous control of the generated text content and format during the text generation process. This ensures that the generated initial decision text meets the pre-set rule requirements, thereby improving the reliability and standardization of the decision report.

[0037] In one embodiment, in step 102, a structured generation instruction can be generated using the following method: First, the governance objective rules are identified to obtain the core conclusions and evaluation dimensions that the procurement governance task needs to achieve; then, the compliance boundary rules are extracted to obtain the compliance statements and expression restrictions that the decision text needs to follow; next, the business execution rules are logically deconstructed to obtain the decision logic, paragraph structure, and paragraph order of the decision text; finally, the core conclusions, evaluation dimensions, compliance statements, expression restrictions, decision logic, paragraph structure, and paragraph order are parameterized and encapsulated to form a structured generation instruction.

[0038] In this embodiment, after receiving the procurement governance task and the corresponding set of rule constraints, the intent of the governance objective rules in the rule constraint set can be identified first to determine the core conclusions required by the task for the final presentation of the decision text and the evaluation dimensions that need to be covered. Next, the compliance boundary rules can be parsed and conditions extracted, and the mandatory compliance statements that must be included in the decision text can be extracted, while simultaneously identifying the forbidden expressions that must be strictly avoided during text generation. Then, the business execution rules can be logically deconstructed and transformed into specific requirements for the macro-organizational form of the decision text, i.e., forming a clear decision logic, paragraph structure, and logical arrangement order between paragraphs. Finally, all key elements obtained in the above steps, such as the core conclusions, evaluation dimensions, compliance statements, forbidden expressions, paragraph structure, and arrangement order, can be standardized and parameterized, thereby integrating them into a structured generation instruction that can be understood and executed by the decision content generation model.

[0039] This embodiment transforms different types of rules into text elements corresponding to the decision text, and converts each text element into structured generation instructions. This allows the generated decision text to be subject to various rule constraints, format constraints, and compliance boundary controls, thereby ensuring the reliability of the report content.

[0040] In one embodiment, step 103 can be implemented as follows: First, the entity information, data information, and regulatory clauses referenced in the initial decision text are compared and verified with a preset knowledge database or a certified external data source to obtain an authenticity verification result. Then, it is checked whether the core conclusions in the initial decision text are consistent with the governance objective rules in the rule constraint set to obtain a governance objective verification result. Next, it is checked whether the analysis and reasoning content in the initial decision text are consistent with the business execution rules in the rule constraint set to obtain a logical consistency verification result. Finally, it is determined whether the decision suggestions and conclusions in the initial decision text are within the compliance boundary defined by the compliance boundary rules in the rule constraint set to obtain a compliance verification result.

[0041] In this embodiment, during authenticity verification, all explicitly cited entity names, specific numerical data, and relevant legal and policy clauses in the initial decision text can be extracted. This information is then automatically compared and verified against a pre-set authoritative knowledge database within the system or a securely certified external data service interface to obtain a verification result regarding the authenticity and accuracy of the information. During governance objective verification, the core conclusions in the initial decision text can be extracted and verified using the core objectives and evaluation criteria defined in the governance objective rules. This verifies whether the core conclusions in the initial decision text have achieved the predetermined core objectives, thereby obtaining the governance objectives. Verification results: During logical consistency verification, the argumentative structure of the initial decision text can be analyzed, including the logical chain of analysis and deduction, as well as the final conclusion. This content is then matched and checked against the business execution rules contained in the rule constraint set item by item to determine whether the internal logic of the text is self-consistent and conforms to the business rules, and a logical consistency verification result is generated. During compliance verification, all decision suggestions and key conclusions proposed in the initial decision text can be verified one by one to see whether they are completely within the compliance boundary range defined by the various compliance boundary rules explicitly specified in the rule constraint set, and the final compliance verification result is output.

[0042] For example, when verifying the authenticity of the initial text of a server procurement risk assessment report, the supplier name "Company A" and its quotation "1.2 million yuan" mentioned in the text can be extracted and compared with the enterprise credit information disclosure system and the recent market bidding price database. When performing logical consistency verification, the reasoning in the text that "it is recommended to negotiate the price because the quotation exceeds the historical average price by 15%" can be checked to see if it is consistent with the business logic rule in the rule set that "when the price deviates by more than 10%, the negotiation process must be initiated". When performing compliance verification, it can be determined whether the conclusion in the text that "it is recommended to purchase model B" meets the compliance boundary condition in the rule set that "the energy efficiency of the equipment must meet the national first-class standard".

[0043] This embodiment verifies the authenticity, reliability, and compliance of the initial decision text through multiple dimensions, including authenticity, governance objectives, logical consistency, and compliance. This ensures the authenticity, reliability, and compliance of the text content, thereby effectively identifying and intercepting potential inaccuracies, contradictions, and violations in the decision text and improving text quality.

[0044] In one embodiment, when the initial decision text fails the multi-dimensional information verification, a text correction mechanism can be automatically triggered. The text correction mechanism includes at least one of the following: feeding back the text that failed verification and the reason for the failure to the decision content generation model, so that the decision content generation model can regenerate part or all of the text based on the rule constraint set; adjusting the constraint parameters in the generation instruction according to the verification result, and controlling the decision content generation model to regenerate part or all of the text based on the rule constraint set based on the adjusted generation instruction; generating a verification anomaly report and submitting the verification anomaly report to the manual review interface to receive the manually corrected text returned by the manual review interface; finally, obtaining the corrected decision text based on the regenerated part or all of the text and the manually corrected text.

[0045] In this embodiment, if the initial decision text fails to pass any of the following checks—authenticity verification, governance objective verification, logical consistency verification, or compliance verification—a text correction mechanism can be automatically triggered to handle this anomaly. During text correction, the specific type and location of the verification failure can be analyzed, and then, based on a preset strategy configuration, one or more correction methods can be selected and combined. The first method involves feeding back the failed text fragments, along with a detailed description of the failure reason (e.g., a data verification failed or a conclusion violated a rule), to the decision content generation model, requiring the model to regenerate that portion of the text content under the guidance of the original rule constraint set. The second method involves adjusting the constraint parameters in the generation instruction based on the multi-dimensional verification results, and then controlling the decision content generation model to regenerate part or all of the text under the constraints of the rule constraint set based on the adjusted generation instruction. The third method involves generating a structured verification anomaly report for complex conflicts that cannot be resolved automatically, detailing the problem, and submitting this report to a dedicated manual review terminal via the system interface, awaiting review and manual text correction by authorized personnel.

[0046] This embodiment automatically triggers a text correction mechanism with multiple strategies when a verification failure is detected, which can quickly process related erroneous content, thereby ensuring the accuracy and reliability of decision text.

[0047] In one embodiment, in step 101, the set of rule constraints related to the procurement governance task can be obtained by the following method: First, based on the business domain involved in the procurement governance task, a subset of rule constraints corresponding to the procurement governance task is retrieved in the rule management module of the centralized procurement governance hub. Then, based on the rule trees pre-built for multiple business modules, the hierarchical path identifier corresponding to each rule constraint in the subset of rule constraints is obtained. The rule tree includes, but is not limited to, upper-level governance target nodes, middle-level rule boundary nodes, and lower-level business rule nodes. Finally, the corresponding hierarchical path identifier is attached to each rule constraint in the subset of rule constraints to obtain the set of rule constraints related to the procurement governance task.

[0048] In this embodiment, when obtaining the set of rule constraints related to the procurement governance task, the received procurement governance task can first be parsed, and the business domain involved in the procurement governance task can be identified, such as cost control or supplier compliance. Then, based on this domain information, a query can be initiated to the rule management module of the centralized procurement governance hub to retrieve and obtain a subset of rule constraints that initially corresponds to the task. Subsequently, the relevant business modules can be determined according to the scope of the procurement governance task, and the rule tree structure pre-built for these modules can be accessed. In each relevant rule tree, the specific node corresponding to each rule in the rule constraint subset can be located, and its complete hierarchical path in the entire rule tree structure can be traced to obtain the hierarchical path identifier containing the upper-level governance target node, the middle-level rule boundary node, and the lower-level business rule node. Finally, each retrieved rule constraint can be appended with its unique hierarchical path identifier and assembled into a set of rule constraints with clear traceability information.

[0049] For example, when acquiring the set of rule constraints related to a server procurement risk assessment task, the business areas involving cost control and compliance checks are first identified. Then, a preliminary subset of rule constraints can be retrieved from the rule management module, including rules such as "procurement price must be compared to a threshold" and "equipment energy efficiency must meet national standards." Next, the rule tree of the cost control module can be accessed to find the lower-level business rule node corresponding to the price threshold rule, and traced upwards to its corresponding middle-level rule boundary node and upper-level governance target node, forming a path identifier such as "cost control module - target node cost optimization - boundary node price control - business rule node threshold comparison." Similarly, a similar operation can be performed on energy efficiency rules. Finally, each rule is appended with its source path, thus forming a complete set of rule constraints.

[0050] This embodiment ensures the relevance and specificity of rules within the rule constraint set by retrieving the corresponding subset of rules based on the domains involved in the procurement governance task. Furthermore, by using a pre-built rule tree structure to attach a corresponding hierarchical path identifier to each rule, each rule constraint can be traced back to its corresponding business source and logical level. This enhances the interpretability of the rule constraint set and provides a clear rule framework and location basis for subsequent decision text generation, verification, and tracing.

[0051] In one embodiment, the hierarchical path identifier corresponding to each rule constraint in the rule constraint subset can be obtained by the following method: First, based on the governance objectives and business scope of the procurement governance task, at least one business module related to the procurement governance task is determined. Then, in the rule tree corresponding to each business module, a partial set of nodes related to the procurement governance task is extracted, wherein the partial set of nodes includes at least one upper-level governance target node and a lower-level business rule node associated with the upper-level governance target node. Finally, based on the partial set of nodes, the hierarchical path identifier of each rule constraint in the rule constraint subset in the rule tree corresponding to the target business module is determined.

[0052] In this embodiment, when obtaining the hierarchical path identifier corresponding to each rule in the rule constraint subset, the governance objectives declared by the procurement governance task and the specific business scope it covers can be analyzed first, such as controlling procurement costs or ensuring supplier compliance. Based on this, at least one related business module can be identified, such as a cost analysis module or a compliance review module. Next, for each identified business module, its corresponding rule tree structure can be accessed. By matching the governance objectives and business scope of the task, a partial set of nodes related to the procurement governance task can be filtered and extracted from the tree structure. The partial set of nodes contains at least one upper-level governance objective node that describes the governance objective, and several lower-level business rule nodes related to the specific operation of the task under the objective. Finally, based on the subtree structure formed by this extracted partial set of nodes, the position and level of each rule constraint in the rule tree corresponding to its business module can be determined, thereby generating a complete hierarchical path identifier reflecting the path from the upper-level governance objective node to the rule itself.

[0053] This embodiment locates the relevant business module based on the procurement governance task and extracts the relevant node set containing governance objectives and specific rules from the rule tree corresponding to the business module. This provides correct contextual information and search scope for generating hierarchical path identifiers. Furthermore, by generating detailed hierarchical path identifiers for each rule based on a partial node set, the origin and logical hierarchy of each rule constraint become readily apparent, thereby greatly enhancing the manageability, interpretability, and traceability of the rules.

[0054] In one embodiment, step 104 can be implemented as follows: First, based on the hierarchical path identifier of each rule constraint invoked during the initial decision text generation and correction process, and the rule tree where the rule constraint is located, a rule application path corresponding to the decision text is generated. Then, based on the source data identifier and data generation time of each assertion and data basis in the initial decision text, the data source information corresponding to the decision text is obtained. Next, based on the multi-dimensional verification results, multi-dimensional verification time, and verification execution unit identifier of the initial decision text, multi-dimensional verification information corresponding to the decision text is obtained. Finally, the verified and corrected decision text, rule application path, data source information, and multi-dimensional verification information are associated and encapsulated to form a traceable decision record.

[0055] In this embodiment, when generating traceable decision records, the initial decision text generation log can be traced back first. The hierarchical path identifiers carried by each rule constraint invoked during the generation or regeneration of the initial decision text, as well as the rule tree identifier to which that rule constraint belongs, are combined and sorted according to the log records to generate a complete rule application path describing the rule invocation order and source. Next, semantic parsing of the initial decision text can be performed to identify each key assertion and supporting data basis in the text. The unique source data identifiers corresponding to these assertions and data in the system, along with the timestamps of generation or acquisition, are extracted, thus summarizing the data source information corresponding to the decision text. Furthermore, verification information output from the multi-dimensional information verification process can be obtained, including the specific verification results, execution time, and service unit identifier for each of the authenticity verification, governance objective verification, logical consistency verification, and compliance verification, and this information is organized into multi-dimensional verification information corresponding to the decision text. Finally, the finalized decision text, the generated rule application path, data source information, and multi-dimensional verification information can be bound and packaged using data encapsulation technology to form a data package with a unique identifier, thereby completing the construction of a traceable decision record.

[0056] This embodiment collects and records the rule application path, data source information, and multi-dimensional verification information used in the process of generating decision text, and associates and encapsulates this information with the decision text, which makes the logic of decision content generation, data basis, and verification process more transparent and traceable.

[0057] In one embodiment, step 105 can be implemented by integrating traceable decision records through a preset presentation format to generate a procurement governance decision report for procurement governance tasks. The procurement governance decision report includes formatted decision text and supplementary files containing rule application paths, data source information, and multi-dimensional verification information.

[0058] In this embodiment, after a traceable decision record is created, the decision text portion of the record can be formatted according to preset document styles and layout rules to generate a formatted decision text suitable for review or archiving. Subsequently, the rule application paths, data source information, and multi-dimensional verification information from the decision record can be extracted and organized, and an appendix can be generated according to a predefined template or structure. Finally, the formatted decision text can be used as the main body of the report, bound to the generated appendix, and together output as a complete procurement governance decision report.

[0059] This embodiment integrates decision records through a preset presentation format, ensuring that the final report has a consistent and professional appearance. Furthermore, by providing the formatted decision text along with supplementary documents containing complete traceability information, transparency and traceability of all key elements in the decision-making process are ensured, thereby simultaneously meeting the dual requirements of result delivery and content auditing.

[0060] In one embodiment, after step 105, the above method may further include the following steps: First, the rule application path corresponding to the decision text is parsed to identify the node position and level of the rule constraints applied by the decision text in the rule tree. Then, a decision logic explanation text is generated based on the node position and level. The decision logic explanation text is used to explain the rules on which the decision points in the decision text are based and the business logic positioning based on the rules. Finally, the decision logic explanation text is output as an independent chapter of the procurement governance decision report or as part of the supplementary document of the procurement governance decision report.

[0061] In this embodiment, after generating the procurement governance decision report, the rule application paths in the encapsulated decision records can be parsed to identify the specific node position and hierarchy of each rule constraint applied in the decision text within the rule tree corresponding to its business module. For example, it can be determined that a certain rule is a direct child node under a certain upper-level governance target node. Then, based on the parsed node positions and hierarchical relationships, a natural language generation model can be invoked to generate a decision logic explanation text. This explanation text can be used to explain the specific rules upon which each key decision point or conclusion in the decision text is based, and to elucidate the logical position of the rule in the rule tree, such as explaining what governance objective it serves. Finally, the generated decision logic explanation text can be added as a new independent chapter to the procurement governance decision report, or it can be merged into an existing supplementary file and output to the user along with the main report.

[0062] This embodiment, by parsing the rule application path and converting it into natural language, can clearly show the specific rules on which each conclusion is based and their position in the overall business logic framework. This can significantly enhance the transparency, understandability, and auditability of the decision-making process, making it easier for users to quickly grasp the underlying basis of the decision.

[0063] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. In addition, the labels corresponding to each step in the above embodiments are only for identification purposes and are not intended to limit the execution order of the steps. The execution order of the steps in each embodiment can be set according to the actual situation.

[0064] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0065] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0066] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A rule-constrained decision generation method for a centralized procurement governance center based on AI, wherein the method is executed in the centralized procurement governance center, characterized in that, The method includes: Receive procurement governance tasks and obtain a set of rules and constraints related to the procurement governance tasks, wherein the set of rules and constraints includes governance objective rules, compliance boundary rules, and business execution rules; The text generation format is determined based on the set of rules and constraints, and the decision content generation model is invoked to generate initial decision text under the constraints of the set of rules and constraints and the text generation format. Based on the set of rules and constraints, the initial decision text is verified in multiple dimensions, and the initial decision text that fails the verification is rolled back and regenerated to obtain the corrected decision text. The verified and corrected decision text is associated and stored with the corresponding rule application path, data source information, and multi-dimensional verification information to form a traceable decision record. A procurement governance decision report for the procurement governance task is generated based on the traceable decision records. The step of associating and storing the verified and corrected decision text with the corresponding rule application path, data source information, and multi-dimensional verification information to form a traceable decision record includes: Based on the hierarchical path identifier of each rule constraint called during the initial decision text generation and correction process and the rule tree in which the rule constraint is located, a rule application path corresponding to the verified and corrected decision text is generated. The rule tree includes upper-level governance target nodes, middle-level rule boundary nodes, and lower-level business rule nodes. The rule application path is the complete hierarchical path of each rule constraint in the rule tree corresponding to its business module. Based on the source data identifier and data generation time of each assertion and data basis in the initial decision text, the data source information corresponding to the verified and corrected decision text is obtained; Based on the multi-dimensional verification results, multi-dimensional verification time, and the identifier of the verification execution unit of the initial decision text, the multi-dimensional verification information corresponding to the verified and corrected decision text is obtained. The verified and corrected decision text, the rule application path, the data source information, and the multi-dimensional verification information are associated and encapsulated to form the traceable decision record.

2. The method according to claim 1, characterized in that, The step of determining the text generation format based on the set of rules and constraints, and then calling the decision content generation model to generate initial decision text under the constraints of the set of rules and constraints and the text generation format, includes: The rule constraint set is parsed, and governance objective rules, compliance boundary rules, and business execution rules are extracted from the rule constraint set; Based on the governance objective rules, the compliance boundary rules, and the business execution rules, a structured generation instruction is generated to control text generation. The decision content generation model is invoked to generate initial decision text whose text content and format both conform to the rule requirements, under the constraints of the structured generation instructions.

3. The method according to claim 2, characterized in that, The step of generating structured generation instructions for controlling text generation based on the governance objective rules, the compliance boundary rules, and the business execution rules includes: The intent of the governance objective rules is identified to obtain the core conclusions and evaluation dimensions that the procurement governance task needs to achieve. Conditions are extracted from the aforementioned compliance boundary rules to obtain the compliance statements and prohibited expressions that the decision text must follow; The business execution rules are logically deconstructed to obtain the decision logic, paragraph structure, and paragraph arrangement order of the decision text; The core conclusions, evaluation dimensions, compliance statements, prohibited expressions, decision-making logic, paragraph structure, and paragraph order are parameterized and encapsulated to obtain the structured generation instructions.

4. The method according to claim 1, characterized in that, The step of performing multi-dimensional information verification on the initial decision text based on the set of rules and constraints includes: The entity information, data information, and legal clauses referenced in the initial decision text are compared and verified with a preset knowledge database or a certified external data source to obtain the authenticity verification result. Check whether the core conclusions in the initial decision text are consistent with the governance objective rules in the rule constraint set to obtain the governance objective verification result; Check whether the analysis and reasoning content in the initial decision text is consistent with the business execution rules in the rule constraint set to obtain the logical consistency verification result; Determine whether the decision suggestions and conclusions in the initial decision text are within the compliance boundary defined by the compliance boundary rules in the rule constraint set, and obtain the compliance verification result.

5. The method according to claim 2 or 3, characterized in that, The process of rolling back and regenerating the initial decision text that failed the verification to obtain the corrected decision text includes: When the initial decision text fails the multi-dimensional information verification, a text correction mechanism is automatically triggered, wherein the text correction mechanism includes at least one of the following: The text that failed the verification and the reason for the failure are fed back to the decision content generation model, so that the decision content generation model can regenerate part or all of the text based on the set of rule constraints. Adjust the constraint parameters in the generation instruction according to the verification results, and control the decision content generation model to regenerate part or all of the text based on the rule constraint set based on the adjusted generation instruction; Generate a verification anomaly report and submit the verification anomaly report to the manual review interface to receive the manually corrected text returned by the manual review interface; Based on the regenerated partial or complete text, as well as the manually corrected text, the revised decision text is obtained.

6. The method according to claim 1, characterized in that, The acquisition of the set of rule constraints related to the procurement governance task includes: Based on the business areas involved in the procurement governance task, a subset of rule constraints corresponding to the procurement governance task is retrieved from the rule management module of the centralized procurement governance center; Based on the rule trees pre-built for multiple business modules, the hierarchical path identifier corresponding to each rule constraint in the rule constraint subset is obtained, wherein the rule tree includes upper-level governance target nodes, middle-level rule boundary nodes, and lower-level business rule nodes; A corresponding hierarchical path identifier is attached to each rule constraint in the subset of rule constraints to obtain a set of rule constraints related to the procurement governance task.

7. The method according to claim 6, characterized in that, The step of obtaining the hierarchical path identifier corresponding to each rule constraint in the rule constraint subset, based on a rule tree pre-constructed for multiple business modules, includes: Based on the governance objectives and business scope of the procurement governance task, identify at least one business module related to the procurement governance task; In the rule tree corresponding to each of the business modules, a partial set of nodes related to the procurement governance task is extracted, wherein the partial set of nodes includes at least one upper-level governance target node and a lower-level business rule node associated with the upper-level governance target node; Based on the aforementioned set of nodes, determine the hierarchical path identifier of each rule constraint in the rule tree corresponding to the business module within the rule constraint subset.

8. The method according to claim 1, characterized in that, The process of generating a procurement governance decision report for the procurement governance task based on the traceable decision records includes: The traceable decision records are integrated through a preset presentation format to generate the procurement governance decision report, which includes a formatted decision text and ancillary files containing the rule application path, the data source information, and the multi-dimensional verification information.

9. The method according to claim 8, characterized in that, The method further includes: The rule application path corresponding to the formatted decision text is parsed to identify the node position and level of the rule constraints applied to the formatted decision text in the rule tree. Based on the node location and hierarchy, a decision logic explanation text is generated. The decision logic explanation text is used to explain the basis rules of the decision points in the formatted decision text and the business logic positioning of the basis rules. The explanation of the decision-making logic may be output as a separate chapter of the procurement governance decision-making report or as part of an appendix to the procurement governance decision-making report.

Images