Intelligent reasoning method and device, computer equipment and storage medium

Abstract

The invention discloses an intelligent reasoning method and device, computer equipment and a storage medium, belongs to the technical field of artificial intelligence, and is applied to trend reasoning in specific scenes such as science and technology finance or old-age care medical treatment. According to the method, the multi-level knowledge graph of the basic data layer, the business logic layer and the strategic target layer is constructed, and the cross-layer jump reasoning mechanism based on the reasoning value density is introduced, so that the intelligent reasoning process can dynamically adjust the reasoning path according to the user intention and the reasoning value; therefore, the matching degree and effectiveness between the reasoning result and the core task are improved. Meanwhile, through comprehensive calculation of multi-dimensional evaluation parameters in the reasoning process, quantitative evaluation of reasoning quality is achieved, and the system can give consideration to reasoning efficiency and resource utilization rate on the premise that reasoning accuracy is guaranteed. Key data in the whole reasoning process is written into a block chain log, it is ensured that reasoning paths and results are completely recorded and cannot be tampered, and traceability and auditing performance of the reasoning process are achieved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically relating to an intelligent reasoning method, device, computer equipment, and storage medium. Background Technology

[0002] With the rapid development of artificial intelligence technology, data-driven intelligent reasoning methods are widely used in business decision support, risk control, intelligent operation and maintenance, and strategic planning. Existing intelligent reasoning systems often rely on single-level data models or flat knowledge bases, lacking systematic modeling of the hierarchical relationships between raw data, business logic, and strategic objectives. This makes it difficult for the reasoning process to dynamically adapt to user needs of varying complexity. Furthermore, existing technologies typically execute reasoning along fixed paths or according to preset rules, lacking a mechanism for quantitatively evaluating the value of the reasoning results. When the information density of the reasoning path is insufficient or deviates from the core objective, effective dynamic adjustments cannot be made, easily leading to low reasoning efficiency and insufficient relevance of results.

[0003] As intelligent systems are increasingly applied in critical business scenarios, higher demands are being placed on the credibility, traceability, and security of the reasoning process. For example, when a system conducts risk assessments based on a user's historical claims data, health status, and other basic information, if reasoning is performed only at the basic data level in a single dimension, the assessment results may not match the actual risks due to insufficient information granularity.

[0004] Traditional reasoning systems mostly focus only on the reasoning result itself, neglecting the complete recording and auditing of the reasoning process, making it difficult to meet the needs of compliance supervision, accountability, and system self-optimization. Especially in multi-level knowledge interaction and cross-level reasoning scenarios, the lack of unified reasoning identifiers and reliable log storage mechanisms leads to unreproducible reasoning paths, difficulty in verifying reasoning rules, and risks of data tampering and unverifiable results.

[0005] Therefore, there is an urgent need for an intelligent reasoning method that can perform dynamic reasoning based on multi-level knowledge structures, has the ability to assess reasoning value and jump between levels, and can securely and immutably record the entire reasoning process, so as to improve the intelligence, reliability and application value of the system. Summary of the Invention

[0006] The purpose of this application is to propose an intelligent reasoning method, device, computer equipment, and storage medium, so as to provide an intelligent reasoning method that can perform dynamic reasoning based on a multi-level knowledge structure, has reasoning value assessment and cross-level jump capabilities, and can securely and tamper-proofly record the entire reasoning process.

[0007] To address the aforementioned technical problems, this application provides an intelligent reasoning method, employing the following technical solution: An intelligent reasoning method, comprising: The system receives raw input data and constructs a multi-level knowledge graph based on the raw input data according to preset attribute rules and target classification. The multi-level knowledge graph includes a basic data layer knowledge graph, a business logic layer knowledge graph, and a strategic target layer knowledge graph. Each layer of the knowledge graph has an independent input interface, storage structure, and association rules. The multi-level knowledge graph enables cross-layer information interaction through association keys. The core task is determined based on user intent recognition, and the starting reasoning level is located in a multi-level knowledge graph based on the core task. Perform hierarchical reasoning starting from the initial reasoning level and calculate the reasoning value density; If the reasoning value density is less than the preset value density threshold, a cross-level jump will be triggered in the multi-level knowledge graph until the reasoning value density is greater than or equal to the preset value density threshold. During the hierarchical reasoning process, a reasoning ID is generated for each reasoning step, and the input data ID, reasoning level, reasoning rule ID, calculated reasoning result, and cross-level jump conditions corresponding to the reasoning ID are recorded in the blockchain log.

[0008] To address the aforementioned technical problems, this application also provides an intelligent reasoning device, which employs the following technical solution: An intelligent reasoning device, comprising: The knowledge graph construction module is used to receive raw input data and construct multi-level knowledge graphs based on the raw input data according to preset attribute rules and target classification. The multi-level knowledge graphs include basic data layer knowledge graphs, business logic layer knowledge graphs, and strategic goal layer knowledge graphs. Each layer of knowledge graph has an independent input interface, storage structure, and association rules. The multi-level knowledge graphs realize cross-layer information interaction through association keys. The reasoning hierarchy module is used to identify and determine the core task based on the user's intent, and to locate the starting reasoning hierarchy in the multi-level knowledge graph based on the core task. The initial inference module is used to perform hierarchical inference starting from the initial inference level and to calculate the inference value density; The cross-layer reasoning module is used to trigger cross-layer jumps in the multi-level knowledge graph if the reasoning value density is less than the preset value density threshold, until the reasoning value density is greater than or equal to the preset value density threshold. The reasoning results module is used to generate a reasoning ID for each reasoning step during the hierarchical reasoning process, and to record the input data ID, reasoning level, reasoning rule ID, calculated reasoning result, and cross-level jump conditions corresponding to the reasoning ID to the blockchain log.

[0009] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: A computer device includes a memory and a processor, the memory storing computer-readable instructions, the processor executing the computer-readable instructions to implement the steps of the intelligent reasoning method as described in any of the preceding claims.

[0010] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below: A computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the steps of the intelligent reasoning method as described in any one of the preceding descriptions.

[0011] Compared with the prior art, the embodiments of this application have the following main advantages: This application discloses an intelligent reasoning method, apparatus, computer equipment, and storage medium, belonging to the field of artificial intelligence technology, and applied to trend reasoning in specific scenarios such as fintech or elderly care and healthcare. This application constructs a multi-level knowledge graph consisting of a basic data layer, a business logic layer, and a strategic goal layer, and introduces a cross-layer jump reasoning mechanism based on reasoning value density. This enables the intelligent reasoning process to dynamically adjust the reasoning path according to user intent and reasoning value, avoiding information redundancy or reasoning deviation problems caused by traditional fixed reasoning links, thereby improving the matching degree and effectiveness between reasoning results and core tasks. Simultaneously, by comprehensively calculating multi-dimensional evaluation parameters such as reliability, contribution, time consumption, and resource consumption in the reasoning process, a quantitative assessment of reasoning quality is achieved, enabling the system to balance reasoning efficiency and resource utilization while ensuring reasoning accuracy. A unique reasoning ID is generated for each reasoning step, and key data of the entire reasoning process is written to a blockchain log to ensure the complete recording and immutability of the reasoning path and results, achieving traceability and auditability of the reasoning process. Attached Figure Description

[0012] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 An exemplary system architecture diagram is shown, in which this application can be applied; Figure 2 A flowchart of one embodiment of the intelligent reasoning method according to this application is shown; Figure 3 It shows Figure 2 A flowchart of one embodiment of step S201; Figure 4 A schematic diagram of the structure of one embodiment of the intelligent reasoning device according to this application is shown; Figure 5 It shows Figure 4 A schematic diagram of an embodiment of the Chinese map construction module 401; Figure 6 A schematic diagram of the structure of one embodiment of a computer device according to this application is shown. Detailed Implementation

[0014] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0015] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0016] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0017] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0018] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0019] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0020] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0021] It should be noted that the intelligent reasoning method provided in the embodiments of this application is generally executed by a server / terminal device, and correspondingly, the intelligent reasoning device is generally set in the server / terminal device.

[0022] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; the system can have any number of terminal devices, networks, and servers depending on implementation needs.

[0023] Continue to refer to Figure 2 A flowchart of an embodiment of the intelligent reasoning method according to this application is shown. The intelligent reasoning method includes the following steps: S201, Receive raw input data, and construct a multi-level knowledge graph based on the raw input data according to preset attribute rules and target classification. The multi-level knowledge graph includes a basic data layer knowledge graph, a business logic layer knowledge graph, and a strategic target layer knowledge graph. Each layer of knowledge graph has an independent input interface, storage structure, and association rules. The multi-level knowledge graph realizes cross-layer information interaction through association keys. Specifically, the system first receives raw input data from different data sources through a unified data access interface. This raw input data can include structured, semi-structured, and unstructured data. To ensure the accuracy and consistency of the knowledge graph construction, the system performs preprocessing operations on the received raw input data. These preprocessing operations include at least data cleaning, format conversion, noise removal, missing value handling, and semantic parsing. Subsequently, the system classifies the preprocessed data according to pre-defined classification rules, which divide the data into attribute-based data, rule-based data, and target-based data. Attribute-based data describes objective entities and their characteristics, rule-based data describes the logical relationships or constraints between entities, and target-based data describes strategic objectives and constraint indicators.

[0024] After classification, the system constructs corresponding levels of knowledge graphs based on different types of data: attribute-based data is used to construct the basic data layer knowledge graph, rule-based data is used to construct the business logic layer knowledge graph, and target-based data is used to construct the strategic target layer knowledge graph. Each level of knowledge graph uses independent data input interfaces, storage structures, and association rules during construction, and establishes cross-level relationships through predefined association keys, thereby achieving information connectivity and interaction between multi-level knowledge graphs.

[0025] In one specific embodiment of this application, the input to the basic data layer includes: receiving structured technical parameter data (such as chip computing power, software response time, hardware power consumption, etc.) and unstructured performance test reports; extracting key indicators through OCR recognition and entity extraction technology; and then eliminating the influence of dimensions through normalization processing. The preprocessing formula is as follows:

[0026] In the formula, These are the normalized data values; These are the original technical parameter values; , These are the maximum and minimum sample values ​​for this parameter, respectively. , The target range for normalization (default [0,1]) is set to ensure that technical parameters of different magnitudes can be compared horizontally.

[0027] For input to the business logic layer: It receives enterprise business process specifications, rules relating technology and business (e.g., "server response time ≤ 500ms" corresponds to "e-commerce platform peak user retention rate ≥ 98%)", and a historical decision case library. The rule engine then transforms these natural language rules into machine-readable logical expressions, such as the mapping expression for "technical parameter compliance rate → business indicator achievement rate".

[0028] In the formula, Let be the predicted value of the i-th business indicator; The influence weight of the j-th technical parameter on the ith business indicator (obtained through training with historical cases); This is the normalized value of the j-th technical parameter; This is an error correction item (value range [-0.05, 0.05], dynamically adjusted based on real-time business feedback).

[0029] For the strategic goal layer input: receive the company's medium and long-term development goals (such as "increase market share by 20% within 3 years"), industry policy constraints, resource budget limits and other qualitative and quantitative indicators, extract core goal parameters through semantic segmentation technology, construct a strategic goal tree model, and decompose the top-level goal into quantifiable sub-goals (such as "annual technology investment ratio ≤ 15% of revenue" and "new business technology adaptation rate ≥ 80%").

[0030] In this embodiment, the association key of the multi-level knowledge graph adopts a composite index structure design, which consists of three parts: "hierarchical identifier - entity unique ID - association type code". The hierarchical identifier is used to distinguish the basic data layer (code 01), the business logic layer (code 02) and the strategic goal layer (code 03). The entity unique ID adopts a 128-bit UUID format to ensure global uniqueness. The association type code identifies the strength level of data association through 3 bits (001 indicates strong association, 010 indicates medium association and 100 indicates weak association). For example, the entity ID of a server in the basic data layer is "uuid-20230101-server-001". Its association key with the "e-commerce platform response speed rule" (entity ID "uuid-20230101-rule-005") in the business logic layer can be represented as "01-uuid-20230101-server-001-001-02-uuid-20230101-rule-005". This association key allows direct retrieval of all rule data related to the server's performance from both layers of the knowledge graph. Simultaneously, the system periodically verifies cross-layer associations, performing a validity scan of association keys at 3 AM daily. When the update frequency of the entity node data corresponding to a certain association key exceeds a preset threshold (e.g., ≥5 updates within 24 hours), a reassessment of the association strength is automatically triggered. The association type code is adjusted by calculating the cosine similarity of the associated data (threshold set to 0.85) to ensure the accuracy and timeliness of cross-layer information exchange.

[0031] Furthermore, regarding the storage structure of the multi-level knowledge graph: the basic data layer uses the graph database Neo4j to store entity-attribute associations (such as "Chip A-Computing Power-100TOPS"); the business logic layer uses the relational database MySQL to store rule expressions and association weights; and the strategic goal layer uses the document database MongoDB to store the goal tree and constraints. The three layers of data are linked for querying through the association key "Entity ID-Rule ID-Goal ID".

[0032] S202, determine the core task based on user intent recognition, and locate the starting reasoning level in the multi-level knowledge graph based on the core task; Specifically, the system receives a query request or task description input by the user and performs semantic parsing processing on the query request or task description. Semantic parsing processing identifies keywords, key phrases, and semantic relationships in the user input to extract intent feature information that represents the user's true needs. Based on this intent feature information, the system classifies and judges the user's needs through a user intent recognition module, thereby determining the core task corresponding to that user need. The core task clarifies the target direction of this reasoning, such as focusing on technical parameter analysis, business logic derivation, or strategic goal matching. After determining the core task, the system matches the core task with pre-set reasoning hierarchy mapping rules, which describe the correspondence between different types of core tasks and various levels in a multi-level knowledge graph. Through the matching results, the system locates the level most relevant to the core task in the multi-level knowledge graph and determines this level as the starting reasoning level.

[0033] S203, start hierarchical reasoning from the initial reasoning level and calculate the reasoning value density; Specifically, the system uses the initial reasoning level as the entry point for reasoning. It calls reasoning rules matching the core task from the knowledge graph of the corresponding level to perform reasoning calculations on the associated node data. During the reasoning calculation process, the system analyzes the input data item by item according to the structural features and association rules of the knowledge graph at that level, generating intermediate reasoning results. These intermediate reasoning results can be numerical results, logical judgment results, or relational derivation results. After generating the intermediate reasoning results, the system verifies them to determine whether the reasoning process meets the requirements of data integrity, consistency, and rule adaptability. Based on successful verification, the system further collects evaluation information related to this reasoning, including but not limited to reliability parameters of the reasoning results, the degree of correlation between the reasoning results and the core task, the time consumed in the reasoning process, and system resource usage. Based on the collected evaluation information, the system performs combined calculations on each evaluation parameter according to preset calculation rules to calculate the reasoning value density corresponding to the current level of reasoning, which is used to characterize the comprehensive evaluation level of the current reasoning result.

[0034] S204. If the reasoning value density is less than the preset value density threshold, then a cross-level jump is triggered in the multi-level knowledge graph until the reasoning value density is greater than or equal to the preset value density threshold. Specifically, the system first compares the reasoning value density calculated at the current level with a preset value density threshold to determine if the conditions for continuing reasoning at the current level are met. When the comparison result shows that the reasoning value density is less than the preset value density threshold, the system determines that the information density or relevance of the reasoning result at the current level is insufficient, and supplementary information needs to be obtained through cross-level reasoning. At this time, based on the preset level jump rules and the cross-level association keys associated with the current reasoning level, the system searches for other levels that are related to the current level in the multi-level knowledge graph, forming a set of candidate jump levels. The system filters the candidate jump levels and determines the level with a high degree of matching with the current core task as the target jump level. Subsequently, the system passes the reasoning result of the current level as input to the target jump level, and re-executes the reasoning process and reasoning value density calculation process at that level. The above comparison and jump process can be repeated until the reasoning value density reaches or exceeds the preset value density threshold, thereby ending the cross-level jump process.

[0035] In one specific embodiment of this application, after the system receives a user request (such as "technology selection for server upgrade of an e-commerce platform"), the intelligent agent determines the core reasoning task through the intent recognition module and automatically locates the starting level of the reasoning. If the request includes a query for specific technical parameters, it starts from the basic data layer; if it includes business impact analysis, it starts from the business logic layer; if it includes a strategic matching degree judgment, it starts from the strategic goal layer.

[0036] Redirection is triggered based on the "inference value density" calculation result. When the value density of the current level's inference result is lower than a threshold, automatic redirection to the associated level occurs. The formula for calculating inference value density is as follows:

[0037] In the formula, The inference value density (value range [0,10]); The reliability of the current inference result is determined by data integrity and rule matching degree. ,in These are the weighting coefficients. For data integrity (0-1). (Rule matching degree (0-1)) The contribution of the current result to the core task (calculated from the goal relevance, 0-1); The time taken for reasoning at the current level (in seconds); System resources consumed for inference at the current level (CPU utilization, 0-1). (threshold) When the default value is 3 (which can be adjusted according to the complexity of the task), a cross-level jump is triggered.

[0038] For example, in server upgrade selection, the agent first infers from the basic data layer that "server B's computing power normalization value is 0.9 and power consumption normalization value is 0.3", and then calculates the value at this time. Continue reasoning at this level; when reasoning reaches "the purchase cost of server B", data integrity... (Prices for some models are missing), calculation This triggers a jump to the business logic layer, where the reasoning continues based on the rule of "procurement cost → operating cost ratio".

[0039] After the jump, the logic verification module ensures the continuity of the reasoning chain by matching and verifying the "previous result - current level rule". For example, when jumping from the basic data layer to the business logic layer, it verifies whether the "server computing power parameter" meets the rule threshold of the business logic layer's "e-commerce peak processing capacity". If it does not meet the threshold, it returns to the basic data layer to re-filter the parameters. If it meets the threshold, it calculates the impact value of business indicators based on the rules.

[0040] S205 generates a reasoning ID for each reasoning step during the hierarchical reasoning process and records the input data ID, reasoning level, reasoning rule ID, calculated reasoning result, and cross-level jump conditions corresponding to the reasoning ID in the blockchain log.

[0041] Specifically, throughout the entire process of hierarchical reasoning and cross-level jumps, the system generates a unique reasoning ID for each reasoning operation. This reasoning ID identifies a single reasoning action and serves as an index for the reasoning process record. After generating the reasoning ID, the system summarizes and organizes the key information associated with it. This key information includes at least the input data ID, the current level of reasoning, the ID of the invoked reasoning rule, the intermediate and final results generated during the reasoning calculation, and the judgment conditions corresponding to triggering cross-level jumps. The system encapsulates the above information into reasoning log data according to a preset data structure and writes it to the blockchain storage system through a blockchain interface. The blockchain log writing process uses a chain structure for storage, creating a sequential relationship between each reasoning log, thereby ensuring the continuity and integrity of the reasoning process record. This method achieves structured recording of the entire reasoning process.

[0042] For example, when the intelligent reasoning method of this application conducts risk assessment based on a user's historical claims data, health status, and other information, it first receives raw input data such as the user's historical claims records, various health indicators (such as blood pressure, blood sugar, and blood lipids) from a medical examination report, and descriptions of past medical history. Then, it constructs a multi-level knowledge graph according to preset attribute rules and target classifications. The basic data layer knowledge graph will integrate these raw data and assign an independent input interface and corresponding storage structure to each data item (such as a claim amount or a specific blood pressure value). For example, historical claim data will be stored in a specific database table in chronological order, while health indicators will be stored according to the type of indicator. The business logic layer knowledge graph will establish association rules based on the insurance industry's underwriting rules and claims policies, such as setting normal blood pressure ranges for different age groups and the relationship between a certain disease and the probability of claims. The strategic goal layer knowledge graph focuses on the insurance company's risk control objectives and profit objectives, such as setting related rules that require an annual risk assessment accuracy rate of 95% and a risk assessment efficiency improvement of 30%.

[0043] These three layers of knowledge graph achieve cross-layer information interaction through unique user identifiers (such as ID card numbers) as association keys. In the basic data layer, a user's ID card number is associated with the underwriting rule judgment result corresponding to that user in the business logic layer. The judgment result of the business logic layer is then associated with the risk level assessment of the corresponding user in the strategic goal layer. Then, based on user intent recognition, the core task is determined to be "user risk level assessment," and based on this, the starting reasoning level in the multi-level knowledge graph is located as the business logic layer, because the core logical rules of risk assessment are mainly located in the business logic layer. Hierarchical reasoning begins from the business logic layer, such as preliminary reasoning based on business rules like whether the user's blood pressure is within the normal range or whether they have a history of serious illness. The reasoning value density at this point is calculated, i.e., the degree to which the current reasoning result contributes to the final risk assessment. If the initial reasoning based solely on some rules from the business logic layer yields a reasoning value density less than the preset value density threshold (e.g., the assessment results based solely on blood pressure and past medical history are not comprehensive enough and contribute less than 60% to the risk level judgment), then a cross-layer jump will be triggered in the multi-level knowledge graph. This may involve jumping to the basic data layer to obtain more detailed data such as historical claim frequency and the reasons for each claim, or jumping to the strategic goal layer to refer to the company's current special assessment strategy for high-risk users, until the reasoning value density is greater than or equal to the preset value density threshold (e.g., after comprehensively considering detailed claim data from the basic data layer, underwriting rules from the business logic layer, and risk control objectives from the strategic goal layer, the reasoning value density reaches 75%).

[0044] Throughout the hierarchical reasoning process, a unique reasoning ID is generated for each reasoning step. For example, the first reasoning step at the business logic layer generates reasoning ID 001. The input data ID corresponding to this reasoning ID (such as the claim data ID and health indicator data ID associated with the user's ID card number), the reasoning level (business logic layer), the reasoning rule ID (such as the rule number R001 for judging whether blood pressure is normal), the calculated reasoning result (such as the preliminary risk level being medium), and the cross-layer jump condition (reasoning value density 60% is less than the threshold 70%) are recorded in the blockchain log to ensure that the entire reasoning process is traceable and tamper-proof.

[0045] To address the issues of broken reasoning chains and poor interpretability in existing technologies, a dual-module approach of "reasoning trace log + dynamic feedback adjustment" is designed to achieve traceability of the reasoning process and continuous optimization of the logic.

[0046] A unique "inference ID" is generated for each cross-layer inference step. The log records all information, including "input data ID - inference level - rule ID used - calculation process - output result - jump trigger condition," and uses blockchain technology to store the log data to ensure immutability. Users can trace the entire process back through the inference ID. For example, if the technology selection result does not meet expectations, the problem of "unreasonable weight coefficient setting" in a certain jump step can be located.

[0047] Furthermore, after the inference output, user feedback is received (e.g., "the results are too conservative," "the technical feasibility analysis is insufficient"), and the correlation parameters are adjusted based on feedback factors. The influence weight of the business logic layer is used. Taking adjustment as an example, the corrected formula is as follows:

[0048] Parameter meaning: The adjusted weights; The weights before adjustment; Set as the learning rate (default 0.01 to avoid sudden weight changes); These are the actual values ​​of the business metrics; This is the inference prediction value; These are technical parameter values. Through this correction mechanism, the system can continuously optimize its inference logic based on actual application results.

[0049] Further, please see Figure 3 The steps of receiving raw input data and constructing a multi-level knowledge graph based on the raw input data according to preset attribute rules and target classification include: S301, perform data cleaning, format unification and semantic parsing on the original input data to obtain standardized data; S302, based on preset classification rules, divides standardized data into attribute data, rule data, and target data, and maps them to the basic data layer knowledge graph, business logic layer knowledge graph, and strategic target layer knowledge graph, respectively; S303, for the knowledge graph of the basic data layer, constructs entity nodes and attribute nodes based on attribute data, and establishes the association between entity nodes and attributes to form the graph structure of the basic data layer; S304, for the knowledge graph of the business logic layer, construct rule nodes based on rule data, and establish the association between the rule nodes and the corresponding entity nodes in the knowledge graph of the basic data layer to form the business logic layer graph structure; S305, for the strategic target layer knowledge graph, construct target nodes based on target data, and establish the relationship between target nodes and rule nodes in the business logic layer knowledge graph, while establishing the relationship between target nodes and corresponding entity nodes in the basic data layer knowledge graph; S306 provides a unified configuration of association keys for multi-level knowledge graphs, enabling cross-layer connections between the basic data layer knowledge graph, the business logic layer knowledge graph, and the strategic goal layer knowledge graph based on these association keys.

[0050] In this embodiment, the system first acquires raw input data from different sources through a unified data access module. This raw input data may include structured parameter data, semi-structured business rule text, and unstructured target description information. For this raw input data, the system sequentially performs data cleaning, format standardization, and semantic parsing operations to eliminate redundant information, outliers, and format differences. Semantic parsing extracts key elements with clear business meaning, thereby forming standardized data that can be processed by the system.

[0051] After standardization, the system classifies the standardized data according to pre-defined attribute and target classification rules. Data describing the characteristics of objective entities is identified as attribute data, data describing logical constraints or deductive relationships between entities is identified as rule data, and data describing business development directions, constraints, or strategic intentions is identified as target data. Subsequently, the system maps the attribute data, rule data, and target data to the basic data layer knowledge graph, business logic layer knowledge graph, and strategic target layer knowledge graph for storage and modeling, respectively.

[0052] In the basic data layer, the system constructs entity nodes and attribute nodes based on attribute data, and describes the subordinate or characteristic relationships between entities and attributes through preset association relationships, thereby forming a basic data layer graph structure that can reflect the underlying factual information. In the business logic layer, the system constructs rule nodes based on rule data, and associates the rule nodes with the corresponding entity nodes in the basic data layer to describe the influence of entity attributes on business logic results. In the strategic goal layer, the system constructs goal nodes based on goal data, and establishes association relationships between goal nodes and rule nodes in the business logic layer, as well as between goal nodes and entity nodes in the basic data layer, to reflect the constraint and guidance role of strategic goals on business rules and underlying entities.

[0053] For example, in the scenario of "server upgrade technology selection for an e-commerce platform", the basic data layer includes entity nodes such as "Server Model A (computing power 8000HPS / power consumption 350W)" and "Server Model B (computing power 10000HPS / power consumption 420W)" as well as attribute nodes such as "computing power", "power consumption", and "procurement cost". Entity nodes and attribute nodes are associated through the "having attribute" relationship. The business logic layer includes rule nodes such as "operating cost calculation rules (annual power consumption = power consumption × runtime × electricity price)" and "processing capacity matching rules (computing power ≥ peak concurrency × single request computing power consumption)". Rule nodes are associated with the "server model" entity node in the basic data layer through the "constrained entity" relationship. The strategic goal layer includes goal nodes such as "green and low-carbon goals (PUE value ≤ 1.3)" and "business continuity goals (annual failure rate ≤ 0.5%)". Goal nodes are associated with the "operating cost calculation rules" node in the business logic layer through the "guidance rules" relationship, and are also directly associated with the "server model" entity node in the basic data layer through the "constrained entity" relationship. By uniformly configuring "server model ID" as a cross-layer association key, cross-layer connections are realized between the "server model B" entity node in the basic data layer, the "processing capacity matching rule" node in the business logic layer, and the "green and low-carbon target" node in the strategic goal layer. When reasoning about "server B procurement cost" in the basic data layer, the association key can be used to quickly jump to the business logic layer to calculate "annual operating cost", and then jump to the strategic goal layer to evaluate "whether green and low-carbon target is met".

[0054] Finally, the system uniformly configures association keys for the three-layer knowledge graph, and realizes data mapping and information interaction between knowledge graphs of different levels through the association keys, thereby constructing a multi-level knowledge graph system with clear hierarchy, complete structure and support for cross-layer association.

[0055] Through the above steps, the constructed multi-level knowledge graph can achieve hierarchical modeling and orderly association of different types of data. At the same time, the association key mechanism between levels ensures the accuracy and logic of data retrieval. This knowledge graph construction method not only meets the storage characteristics requirements of different types of data, but also realizes the connection from the underlying basic data to the high-level strategic goals through cross-layer association design.

[0056] Furthermore, the steps of identifying the core task based on user intent and locating the starting reasoning level in a multi-level knowledge graph based on the core task specifically include: Receive task description input from the user, perform semantic parsing on the task description, and extract key semantic features that represent the user's needs; User intent is identified based on key semantic features to determine the core tasks corresponding to user needs; Match the core task with the preset reasoning level mapping rules to determine the target reasoning level corresponding to the core task; The determined target reasoning level is used as the starting reasoning level, and the starting reasoning level is located in the multi-level knowledge graph.

[0057] In this embodiment, the system first receives task description information input by the user, which can be in natural language or structured instruction form. The system processes the task description through a semantic parsing module, analyzing keywords, semantic roles, and contextual relationships to extract key semantic features that reflect the user's true needs. Based on these key semantic features, the system uses preset intent recognition rules or models to determine the user's intent, categorizing the user's needs into specific task types, and accordingly determining the core task to be completed in this reasoning. After determining the core task, the system calls pre-configured reasoning level mapping rules to match the core task with different levels in a multi-level knowledge graph to identify the most suitable reasoning level to carry the core task. Through this matching process, the system can choose between the basic data layer, business logic layer, and strategic goal layer, and use the matching result as the target reasoning level. Finally, the system locates and initializes the target reasoning level in the multi-level knowledge graph, setting it as the starting reasoning level.

[0058] Through the above steps, the starting point of reasoning based on user intent is accurately located, improving the relevance and execution efficiency of the reasoning process.

[0059] Furthermore, the steps of performing hierarchical reasoning starting from the initial reasoning level and calculating the reasoning value density specifically include: Starting at the initial inference level, inference rules matching the core task are invoked to perform inference operations on the associated key semantic features and generate intermediate inference results. The validity of intermediate inference results is verified to determine whether the intermediate inference results meet the preset integrity and consistency conditions. Based on the verified intermediate inference results, calculate the inference evaluation parameters corresponding to the current inference level. The inference evaluation parameters include at least the reliability of the inference results, the contribution of the inference results to the core task, the inference time, and the resource consumption. Based on the reasoning evaluation parameters, the reasoning value density of the current level of reasoning is calculated, whereby the reasoning value density is used to characterize the overall effectiveness of the reasoning results at the current level.

[0060] In this embodiment, the system uses the determined initial reasoning level as the entry point for reasoning execution. It retrieves reasoning rules matching the core task from the knowledge graph corresponding to that level and performs reasoning operations on key semantic features and node data related to user needs based on these rules, thereby generating intermediate reasoning results. These intermediate reasoning results can include numerical reasoning conclusions, logical judgment conclusions, or relational derivation results. After generating the intermediate reasoning results, the system performs validity verification on them. By checking the completeness of the input data, the suitability of the reasoning rules, and the consistency between the reasoning results, it determines whether the intermediate reasoning results meet preset completeness and consistency conditions. When the intermediate reasoning results pass verification, the system further collects multi-dimensional evaluation information related to this reasoning process and calculates reasoning evaluation parameters corresponding to the current reasoning level based on the collected information. These parameters include reliability parameters reflecting the credibility of the reasoning results, contribution parameters reflecting the degree of support the reasoning results have for the core task, and parameters reflecting reasoning time and system resource consumption. Based on this, the system performs comprehensive calculations on each reasoning evaluation parameter according to preset calculation rules, thereby obtaining the reasoning value density of the current level of reasoning, which is used to quantitatively describe the overall effectiveness of the reasoning results at that level.

[0061] Through the above steps, a quantitative evaluation of the results of hierarchical reasoning is achieved, ensuring that the reasoning process can be dynamically adjusted according to the actual effect, and avoiding resource waste caused by invalid or excessive reasoning.

[0062] Furthermore, the step of calculating the reasoning value density of the current level of reasoning based on the reasoning evaluation parameters specifically includes: Based on the intermediate inference results, a first evaluation parameter is obtained to characterize the reliability of the inference results. The first evaluation parameter is determined at least based on the completeness of the input data and the degree of matching of the inference rules. Based on the correlation between intermediate inference results and core tasks, a second evaluation parameter is obtained to characterize the contribution of inference results to core tasks; During the execution of hierarchical reasoning, the reasoning time information is recorded to obtain a third evaluation parameter used to characterize the time cost of reasoning; During the execution of hierarchical reasoning, the system resource usage is monitored to obtain a fourth evaluation parameter to characterize the consumption of reasoning resources; Based on the first evaluation parameter, the second evaluation parameter, the third evaluation parameter, and the fourth evaluation parameter, the reasoning value density of the current level of reasoning is calculated according to the preset combination calculation rules.

[0063] In this embodiment, after generating and verifying intermediate inference results, the system further refines the calculation of inference evaluation parameters based on the quality and cost of the inference results. First, based on the input data corresponding to the intermediate inference results, the system analyzes the completeness of the input data and the degree of matching between the invoked inference rules and data features, thereby obtaining a first evaluation parameter characterizing the reliability of the inference results. Second, by analyzing the correlation between the intermediate inference results and the determined core tasks, the system determines the extent to which the inference results can support or advance the completion of the core tasks, thereby obtaining a second evaluation parameter characterizing the contribution of the inference results to the core tasks. Simultaneously, during the execution of hierarchical inference, the system records the start and end times of inference, calculates the time consumed by inference, and obtains a third evaluation parameter characterizing the time cost of inference. Furthermore, during inference execution, the system continuously monitors system resource usage such as processor utilization and memory usage, calculating a fourth evaluation parameter characterizing the level of inference resource consumption. Finally, the system performs comprehensive calculations on the first, second, third, and fourth evaluation parameters according to the pre-set combination calculation rules, generating a reasoning value density index that can comprehensively reflect the reasoning quality and cost at the current level, and is used to characterize the comprehensive effectiveness of the reasoning results at the current level.

[0064] For example, in the scenario of "server upgrade technology selection for an e-commerce platform", if the "processing capacity matching rule" reasoning is executed in the business logic layer, the intermediate reasoning result generated is "server model B with a computing power of 10,000 HPS meets the peak concurrency requirements". At this point, the first evaluation parameter (reliability) is determined by checking the completeness of the input "server model B computing power data" (such as whether it contains complete parameters such as peak computing power and continuous computing power) and "rule matching degree" (such as whether the formula "computing power ≥ peak concurrency × single request computing power consumption" is accurately referenced). If the data is complete and the rule matching degree is high, the first evaluation parameter is set to 0.9. The second evaluation parameter (contribution) is determined by analyzing the degree of support of the result for the core task of "server upgrade technology selection". Since computing power matching is a key prerequisite for selection, the contribution parameter is set to 0.85. The third evaluation parameter (inference time) records the time from data retrieval to result output for the rule inference as 0.3 seconds, which is converted to 0.95 according to the preset time consumption scoring standard (≤0.5 seconds is excellent). The fourth evaluation parameter (resource consumption) monitors the processor utilization rate of 15% and memory utilization of 8% during the inference process, which is converted to 0.9 according to the resource consumption threshold (processor utilization ≤20% and memory utilization ≤10% is reasonable). The preset combination calculation rule is "Inference Value Density = (First Evaluation Parameter × 0.4 + Second Evaluation Parameter × 0.3) × (Third Evaluation Parameter × 0.15 + Fourth Evaluation Parameter × 0.15)". Then, the inference value density of the current level is (0.9 × 0.4 + 0.85 × 0.3) × (0.95 × 0.15 + 0.9 × 0.15) = (0.36 + 0.255) × (0.1425 + 0.135) = 0.615 × 0.2775 ≈ 0.1707. This value quantifies the comprehensive effectiveness of this inference step in the business logic layer.

[0065] Through the above steps, a multi-dimensional quantitative evaluation of the value of reasoning results is achieved, making the reasoning process comparable and schedulable.

[0066] Furthermore, if the reasoning value density is less than a preset value density threshold, a cross-level jump is triggered in the multi-level knowledge graph until the reasoning value density is greater than or equal to the preset value density threshold. This process specifically includes: Compare the reasoning value density obtained from the current level with the preset value density threshold to determine whether the cross-level jump trigger condition is met. When the reasoning value density is less than the preset value density threshold, the target jump level associated with the current reasoning level is determined in the multi-level knowledge graph based on the preset hierarchical jump rules. The reasoning result of the current level is passed as cross-level input data to the target jump level, and re-reasoning is performed in the target jump level; Repeatedly perform hierarchical reasoning and reasoning value density calculation in the target jump level, and continuously compare the reasoning value density with the preset value density threshold until the condition for stopping cross-level jump is met.

[0067] In this embodiment, after calculating the reasoning value density of the current level, the system first compares and analyzes this reasoning value density with a preset value density threshold to determine whether the reasoning result of the current level meets the conditions for continuing to serve as the basis for reasoning. When the comparison result shows that the reasoning value density is lower than the preset value density threshold, the system considers that the information provided by the current level reasoning is insufficient to support further derivation of the core task, and it is necessary to introduce knowledge from other levels for supplementary analysis. At this time, based on the preset level jump rules and the association relationship of the current reasoning level in the multi-level knowledge graph, the system searches for candidate jump levels associated with the current reasoning level among the basic data layer, business logic layer, and strategic goal layer, and determines the target jump level with a high degree of matching with the core task. Subsequently, the system transmits the reasoning result obtained from the current level reasoning as cross-layer input data to the target jump level, and re-calls the corresponding reasoning rules to perform reasoning operations in the target jump level. After completing the reasoning, the system calculates the reasoning value density again for the new reasoning result and compares it with the preset value density threshold. The above-mentioned cross-level jump and re-reasoning process can be executed cyclically until the reasoning value density of a certain level reaches or exceeds the preset value density threshold, thereby ending the cross-level jump process.

[0068] Through the above steps, the reasoning process is dynamically switched between different knowledge levels, enabling the reasoning path to be adaptively adjusted according to the value assessment results.

[0069] Furthermore, based on preset hierarchical jump rules, the steps for determining the target jump level associated with the current reasoning level in a multi-level knowledge graph specifically include: Retrieve the hierarchy type information of the current inference level and the cross-level association key associated with the current inference level; Based on hierarchical type information and cross-level association keys, candidate jump levels that are related to the current reasoning level are retrieved in a multi-level knowledge graph; For candidate jump levels, the candidate jump levels are filtered according to the preset jump priority rules and the degree of matching of the core tasks; From the selected candidate jump levels, determine the level that best matches the current reasoning level and use it as the target jump level.

[0070] In this embodiment, when the system needs to perform a cross-level jump, it first obtains the level type information of the current inference level, such as whether the current level belongs to the basic data layer, business logic layer, or strategic goal layer. Simultaneously, it reads the cross-level association key information associated with the node at that level. The cross-level association key is used to identify the mapping relationship between different levels of the knowledge graph and is the foundation for cross-level retrieval and jump. After obtaining the level type information and cross-level association key, the system performs an association retrieval operation in the multi-level knowledge graph based on this information, searching for other levels that are structurally or semantically related to the current inference level, thus forming a set of candidate jump levels. Subsequently, the system comprehensively evaluates the candidate jump levels based on preset jump priority rules and the matching degree between each candidate level and the current core task, and then filters and sorts the candidate jump levels. The jump priority rules can be set according to task type, inference history, or system configuration to control the jump order between different levels. Finally, the system selects the level that best matches the current reasoning level and the core task from the selected candidate jump levels and identifies it as the target jump level.

[0071] Through the above steps, the orderly selection of cross-level jumps is achieved, ensuring the coherence and relevance of the cross-level reasoning process.

[0072] In this embodiment, the intelligent reasoning method runs on an electronic device (e.g., Figure 1 The server shown can receive instructions or acquire data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future wireless connection methods.

[0073] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned original input data, the aforementioned original input data can also be stored in a blockchain node.

[0074] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0075] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0076] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0078] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0079] Further reference Figure 4 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of an intelligent reasoning device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0080] like Figure 4 As shown, the intelligent reasoning device 400 described in this embodiment includes: The graph construction module 401 is used to receive raw input data and construct a multi-level knowledge graph based on the raw input data according to preset attribute rules and target classification. The multi-level knowledge graph includes a basic data layer knowledge graph, a business logic layer knowledge graph, and a strategic goal layer knowledge graph. Each layer of knowledge graph has an independent input interface, storage structure, and association rules. The multi-level knowledge graph realizes cross-layer information interaction through association keys. The reasoning level module 402 is used to identify the core task based on the user's intent and locate the starting reasoning level in the multi-level knowledge graph based on the core task. The initial reasoning module 403 is used to perform hierarchical reasoning starting from the initial reasoning level and to calculate the reasoning value density; The cross-layer reasoning module 404 is used to trigger cross-layer jumps in the multi-level knowledge graph if the reasoning value density is less than the preset value density threshold, until the reasoning value density is greater than or equal to the preset value density threshold. The reasoning result module 405 is used to generate a reasoning ID for each reasoning in the hierarchical reasoning process, and record the input data ID, reasoning level, reasoning rule ID, calculated reasoning result, and cross-level jump conditions corresponding to the reasoning ID to the blockchain log.

[0081] Further, please see Figure 5 The map construction module 401 specifically includes: The standardization processing unit 501 is used to perform data cleaning, format unification and semantic parsing on the raw input data to obtain standardized data. The data classification unit 502 is used to divide standardized data into attribute data, rule data and target data based on preset classification rules, and map them to the basic data layer knowledge graph, business logic layer knowledge graph and strategic target layer knowledge graph respectively. The first graph construction unit 503 is used to construct entity nodes and attribute nodes based on attribute data for the basic data layer knowledge graph, and to establish the relationship between entity nodes and attributes to form the basic data layer graph structure. The second graph construction unit 504 is used to construct rule nodes based on rule data for the business logic layer knowledge graph, and establish the association between the rule nodes and the corresponding entity nodes in the basic data layer knowledge graph to form a business logic layer graph structure. The third graph construction unit 505 is used to construct target nodes based on target data for the strategic target layer knowledge graph, and to establish the relationship between the target nodes and the rule nodes in the business logic layer knowledge graph, while also establishing the relationship between the target nodes and the corresponding entity nodes in the basic data layer knowledge graph. The graph association unit 506 is used to uniformly configure the association keys of multi-level knowledge graphs, and realize cross-layer connections between the basic data layer knowledge graph, the business logic layer knowledge graph, and the strategic goal layer knowledge graph based on the association keys.

[0082] Furthermore, the reasoning hierarchy module 402 specifically includes: The semantic parsing unit is used to receive the task description input by the user, perform semantic parsing on the task description, and extract key semantic features that represent the user's needs. The intent recognition unit is used to recognize user intents based on key semantic features and determine the core tasks corresponding to user needs. The rule matching unit is used to match the core task with the preset reasoning level mapping rules to determine the target reasoning level corresponding to the core task; The hierarchy determination unit is used to take the determined target reasoning hierarchy as the starting reasoning hierarchy and locate the starting reasoning hierarchy in the multi-level knowledge graph.

[0083] Furthermore, the initial inference module 403 specifically includes: The reasoning operation unit is used to start at the initial reasoning level, call the reasoning rules that match the core task, perform reasoning operations on the associated key semantic features, and generate intermediate reasoning results. The inference verification unit is used to verify the validity of intermediate inference results and determine whether the intermediate inference results meet the preset integrity and consistency conditions. The evaluation parameter unit is used to calculate the inference evaluation parameters corresponding to the current inference level based on the intermediate inference results that have passed the verification. The inference evaluation parameters include at least the reliability of the inference results, the contribution of the inference results to the core task, the inference time, and the resource consumption. The value density unit is used to calculate the reasoning value density of the current level of reasoning based on the reasoning evaluation parameters. The reasoning value density is used to characterize the overall effectiveness of the reasoning results at the current level.

[0084] Furthermore, the value density unit specifically includes: The first evaluation parameter subunit is used to obtain a first evaluation parameter to characterize the reliability of the inference result based on the intermediate inference result. The first evaluation parameter is determined at least based on the completeness of the input data and the degree of matching of the inference rule. The second evaluation parameter subunit is used to obtain a second evaluation parameter that characterizes the contribution of the reasoning results to the core task based on the correlation between intermediate reasoning results and core tasks. The third evaluation parameter subunit is used to record reasoning time information and obtain the third evaluation parameter to characterize the reasoning time cost during the execution of hierarchical reasoning. The fourth evaluation parameter subunit is used to monitor system resource usage and obtain a fourth evaluation parameter to characterize the consumption of inference resources during the execution of hierarchical inference. The value density subunit is used to calculate the reasoning value density of the current level of reasoning based on the first evaluation parameter, the second evaluation parameter, the third evaluation parameter, and the fourth evaluation parameter, according to a preset combination calculation rule.

[0085] Furthermore, the cross-layer reasoning module 404 specifically includes: The density comparison unit is used to compare the reasoning value density obtained from the current level of reasoning with the preset value density threshold to determine whether the cross-level jump triggering condition is met. The hierarchical jump unit is used to determine the target jump level associated with the current reasoning level in a multi-level knowledge graph based on preset hierarchical jump rules when the reasoning value density is less than a preset value density threshold. The cross-level reasoning unit is used to pass the reasoning result of the current level as cross-level input data to the target jump level, and to perform re-reasoning in the target jump level; The inference iteration unit is used to repeatedly perform hierarchical inference and inference value density calculation in the target jump level, and continuously compare the inference value density with the preset value density threshold until the condition for stopping cross-level jump is met.

[0086] Furthermore, the cross-layer reasoning unit specifically includes: The cross-level association subunit is used to obtain the level type information of the current inference level and the cross-level association key associated with the current inference level; Candidate level subunits are used to retrieve candidate jump levels that are related to the current reasoning level in a multi-level knowledge graph based on level type information and cross-level association keys; The hierarchical filtering subunit is used to filter candidate jump levels based on preset jump priority rules and the degree of matching of core tasks. The hierarchical matching subunit is used to determine the level that best matches the current inference level from the filtered candidate jump levels, and use it as the target jump level.

[0087] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0088] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0089] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0090] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for intelligent reasoning methods. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0091] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, such as executing computer-readable instructions for the intelligent inference method.

[0092] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0093] This application also provides an embodiment, namely, a computer device including a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the intelligent reasoning method described above.

[0094] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the intelligent reasoning method described above.

[0095] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the 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 this application.

[0096] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0097] It should be noted that the software tools or components not belonging to this company that appear in the various embodiments of this application are merely illustrative examples and do not represent actual use.

[0098] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. An intelligent reasoning method, characterized in that, include: The system receives raw input data and constructs a multi-level knowledge graph based on the raw input data according to preset attribute rules and target classification. The multi-level knowledge graph includes a basic data layer knowledge graph, a business logic layer knowledge graph, and a strategic target layer knowledge graph. Each layer of the knowledge graph has an independent input interface, storage structure, and association rules. The multi-level knowledge graph enables cross-layer information interaction through association keys. The core task is determined based on the user intent, and the starting reasoning level is located in the multi-level knowledge graph based on the core task. Hierarchical reasoning is performed starting from the initial reasoning level, and the reasoning value density is calculated; If the reasoning value density is less than the preset value density threshold, then a cross-level jump is triggered in the multi-level knowledge graph until the reasoning value density is greater than or equal to the preset value density threshold. During the hierarchical reasoning process, a reasoning ID is generated for each reasoning step, and the input data ID, reasoning level, reasoning rule ID, calculation reasoning result, and cross-level jump conditions corresponding to the reasoning ID are recorded in the blockchain log.

2. The intelligent reasoning method as described in claim 1, characterized in that, The step of receiving raw input data and constructing a multi-level knowledge graph based on the raw input data according to preset attribute rules and target classification specifically includes: The original input data is cleaned, formatted, and semantically parsed to obtain standardized data; Based on preset classification rules, the standardized data is divided into attribute data, rule data, and target data, and mapped to the basic data layer knowledge graph, the business logic layer knowledge graph, and the strategic target layer knowledge graph, respectively. For the knowledge graph of the basic data layer, entity nodes and attribute nodes are constructed based on the attribute data, and the association between the entity nodes and attributes is established to form the basic data layer graph structure. For the business logic layer knowledge graph, rule nodes are constructed based on the rule data, and the association between the rule nodes and the corresponding entity nodes in the basic data layer knowledge graph is established to form a business logic layer graph structure. For the strategic target layer knowledge graph, target nodes are constructed based on the target data, and the association between the target nodes and the rule nodes in the business logic layer knowledge graph is established. At the same time, the association between the target nodes and the corresponding entity nodes in the basic data layer knowledge graph is also established. The association keys of the multi-level knowledge graph are configured uniformly, and cross-layer connections between the basic data layer knowledge graph, the business logic layer knowledge graph, and the strategic goal layer knowledge graph are realized based on the association keys.

3. The intelligent reasoning method as described in claim 1, characterized in that, The step of determining the core task based on user intent recognition and locating the starting reasoning level in the multi-level knowledge graph based on the core task specifically includes: Receive a task description input by the user, perform semantic parsing on the task description, and extract key semantic features that represent the user's needs; Based on the key semantic features, user intent is identified to determine the core task corresponding to the user's needs; The core task is matched with a preset reasoning level mapping rule to determine the target reasoning level corresponding to the core task; The determined target reasoning level is used as the starting reasoning level, and the starting reasoning level is located in the multi-level knowledge graph.

4. The intelligent reasoning method as described in claim 3, characterized in that, The step of performing hierarchical reasoning starting from the initial reasoning level and calculating the reasoning value density specifically includes: Starting at the initial reasoning level, the reasoning rules matching the core task are invoked to perform reasoning operations on the associated key semantic features and generate intermediate reasoning results. The intermediate inference results are validated to determine whether they meet the preset integrity and consistency conditions. Based on the verified intermediate inference results, the inference evaluation parameters corresponding to the current inference level are calculated, wherein the inference evaluation parameters include at least the reliability of the inference results, the contribution of the inference results to the core task, the inference time, and the resource consumption. Based on the reasoning evaluation parameters, the reasoning value density of the current level reasoning is calculated, wherein the reasoning value density is used to characterize the overall effectiveness of the current level reasoning results.

5. The intelligent reasoning method as described in claim 4, characterized in that, The step of calculating the reasoning value density of the current level of reasoning based on the reasoning evaluation parameters specifically includes: Based on the intermediate inference results, a first evaluation parameter is obtained to characterize the reliability of the inference results, wherein the first evaluation parameter is determined at least based on the completeness of the input data and the degree of matching of the inference rules; Based on the correlation between the intermediate inference results and the core task, a second evaluation parameter is obtained to characterize the contribution of the inference results to the core task; During the execution of the hierarchical reasoning, the reasoning time information is recorded, and a third evaluation parameter is obtained to characterize the reasoning time cost. During the execution of the hierarchical reasoning, the system resource usage is monitored, and a fourth evaluation parameter is obtained to characterize the reasoning resource consumption. Based on the first evaluation parameter, the second evaluation parameter, the third evaluation parameter, and the fourth evaluation parameter, the reasoning value density of the current level reasoning is calculated according to the preset combination calculation rules.

6. The intelligent reasoning method as described in claim 5, characterized in that, The step of triggering a cross-level jump in the multi-level knowledge graph if the reasoning value density is less than a preset value density threshold, until the reasoning value density is greater than or equal to the preset value density threshold, specifically includes: Compare the reasoning value density obtained from the current level reasoning with the preset value density threshold to determine whether the cross-level jump triggering condition is met. If the reasoning value density is less than the preset value density threshold, a target jump level associated with the current reasoning level is determined in the multi-level knowledge graph based on preset hierarchical jump rules. The reasoning result of the current level is passed as cross-level input data to the target jump level, and re-reasoning is performed in the target jump level; In the target jump level, hierarchical reasoning and reasoning value density calculation are repeatedly performed, and the reasoning value density is continuously compared with the preset value density threshold until the condition for stopping cross-level jumps is met.

7. The intelligent reasoning method as described in claim 6, characterized in that, The step of determining the target jump level associated with the current reasoning level in the multi-level knowledge graph based on preset hierarchical jump rules specifically includes: Obtain the hierarchy type information of the current inference level and the cross-level association key associated with the current inference level; Based on the hierarchical type information and the cross-level association key, candidate jump levels that are associated with the current reasoning level are retrieved in the multi-level knowledge graph; For the candidate jump levels, the candidate jump levels are filtered according to the preset jump priority rules and the matching degree of the core task; The level that best matches the current inference level is determined from the selected candidate jump levels and used as the target jump level.

8. An intelligent reasoning device, characterized in that, include: The knowledge graph construction module is used to receive raw input data and construct a multi-level knowledge graph based on the raw input data according to preset attribute rules and target classification. The multi-level knowledge graph includes a basic data layer knowledge graph, a business logic layer knowledge graph, and a strategic goal layer knowledge graph. Each layer of knowledge graph has an independent input interface, storage structure, and association rules. The multi-level knowledge graph realizes cross-layer information interaction through association keys. The reasoning level module is used to identify and determine the core task based on the user's intent, and to locate the starting reasoning level in the multi-level knowledge graph based on the core task. An initial inference module is used to perform hierarchical inference starting from the initial inference level and to calculate the inference value density; The cross-layer reasoning module is used to trigger a cross-layer jump in the multi-level knowledge graph if the reasoning value density is less than a preset value density threshold, until the reasoning value density is greater than or equal to the preset value density threshold. The reasoning result module is used to generate a reasoning ID for each reasoning in the process of the hierarchical reasoning, and record the input data ID, reasoning level, reasoning rule ID, calculated reasoning result, and cross-level jump conditions corresponding to the reasoning ID to the blockchain log.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the intelligent reasoning method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the intelligent reasoning method as described in any one of claims 1 to 7.

Images