Inference intention driven multi-form mixed knowledge base retrieval method and system

Abstract

The application discloses a multi-form mixed knowledge base retrieval method and system based on reasoning intention driving, which can be applied to the field of artificial intelligence technology. The method comprises the following steps: receiving a query request of a user and identifying a corresponding reasoning intention; matching a target representation form according to the reasoning intention; determining a corresponding target storage module from a mixed knowledge base, and retrieving target knowledge data corresponding to the query request from the target storage module, and then returning the target knowledge data; the mixed knowledge base stores knowledge data of at least two different representation forms corresponding to a same knowledge entity, the same knowledge entity is associated with a unified entity identifier in each representation form, and a cross-form mutual reference relationship is established between different representation forms through the entity identifier. In this way, through the form routing driven by the intention and the mixed knowledge base architecture of multi-form mutual reference, the problems of poor adaptability of a single-form knowledge base and split multi-form data are solved, and the retrieval accuracy and retrieval efficiency in different reasoning scenarios are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a multi-form hybrid knowledge base retrieval method and system based on reasoning intent. Background Technology

[0002] With the development of enterprise digital knowledge management, knowledge base retrieval technology is widely used in scenarios such as business query, fault diagnosis, and logical deduction. Common knowledge representation forms include natural language text, structured entity data, and relation triples, which are used to carry business context logic, precise attribute information, and entity association relationships, respectively.

[0003] In existing technologies, most knowledge bases are constructed and retrieved using a single knowledge representation format. This single format can only adapt to specific types of query needs, and generally suffers from poor adaptability when facing different reasoning scenarios such as business deduction, precise anchoring, and relation traversal. This can easily lead to information redundancy or missing key information. Furthermore, for some knowledge base solutions that employ multiple formats, the independent storage of each format and inconsistent entity identification result in fragmented knowledge data across different formats. The lack of an effective cross-format association mechanism can easily lead to entity misalignment and data inconsistency, making it difficult to balance retrieval accuracy and efficiency across different reasoning scenarios.

[0004] Therefore, how to improve retrieval accuracy and efficiency in different reasoning scenarios is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a multi-form hybrid knowledge base retrieval method and system based on reasoning intent-driven approaches. By employing intent-driven form routing and a hybrid knowledge base architecture with multi-form mutual referencing, it solves the problems of poor adaptability of single-form knowledge bases and fragmented multi-form data, thereby improving retrieval accuracy and efficiency in different reasoning scenarios.

[0006] In a first aspect, embodiments of this application provide a multi-form hybrid knowledge base retrieval method based on reasoning intent, including: Receive user query requests and identify the reasoning intent corresponding to the query requests; Match the corresponding target representation form according to the reasoning intent; The target storage module corresponding to the target representation form is determined from the hybrid knowledge base, and the target knowledge data corresponding to the query request is retrieved from the target storage module and returned. The hybrid knowledge base stores knowledge data corresponding to at least two different representation forms of the same knowledge entity. The same knowledge entity is associated with a unified entity identifier in each representation form, and cross-representation mutual reference relationship is established between different representation forms through the entity identifier.

[0007] Optionally, the hybrid knowledge base is constructed using the following methods: Receive original knowledge sources; At least two different representations of knowledge data corresponding to each knowledge entity are extracted from the original knowledge source; Associate a unified entity identifier with the same knowledge entity under different representation forms, and establish cross-representation mutual reference relationships between different representation forms; Perform cross-format consistency checks on knowledge data corresponding to different representations of the same knowledge entity; A hybrid knowledge base is constructed based on the knowledge data of each valid representation.

[0008] Optionally, the different representation forms include narrative text form, structured entity form, and relation triplet form; The narrative text is natural language text that retains the business context and causal logic chain, and the natural language text is embedded with anchor marks of unified entity identifiers corresponding to knowledge entities. The structured entity form is a structured field object carrying source code path and line number anchors. The structured field object contains a field that is a unified entity identifier for the corresponding knowledge entity. The relation triplet is a subject-predicate-object structure, where the subject and object correspond to knowledge entities and are associated with a unified entity identifier of their respective knowledge entities.

[0009] Optionally, the reasoning intent includes business deduction, precise anchoring, and relationship traversal. The step of matching the corresponding target representation based on the reasoning intent includes: When the reasoning intent corresponding to the query request is identified as a business deduction type, the narrative text form is matched as the corresponding target representation form; When the reasoning intent corresponding to the query request is identified as a precise anchoring class, the structured entity form is matched with the corresponding target representation form. When the reasoning intent corresponding to the query request is identified as a relation traversal class, the form of the matching relation triple is the corresponding target representation form.

[0010] Optionally, the method further includes: When the query request is a compound query, the target knowledge data is assembled into a structured report according to different representation forms and then output.

[0011] Optionally, the cross-morphological consistency verification follows consistency constraint rules; the consistency constraint rules include: entity identifier uniqueness rule, field synchronization rule, relationship integrity rule, and evidence mutual citation rule; The unique entity identifier rule requires that the same knowledge entity use the same entity identifier in all representation forms; The field synchronization rule requires that key fields in the structured entity form have corresponding descriptions in the narrative text form; The relation integrity rule requires that the entity relations in the relation triple form have corresponding causal explanations in the narrative text form; The evidence citation rule requires that different representations of the same knowledge entity share the same source evidence information.

[0012] Optionally, the method further includes: When the cross-morphological consistency check results show data inconsistency, an alarm list containing missing or contradictory entity information is output.

[0013] Secondly, embodiments of this application provide a multi-form hybrid knowledge base retrieval system based on reasoning intent, including: The reasoning intent recognition module is used to receive user query requests and identify the reasoning intent corresponding to the query requests; The matching module is used to match the corresponding target representation form according to the reasoning intent; The morphological routing retrieval module is used to determine the target storage module corresponding to the target representation morphology from the hybrid knowledge base, retrieve the target knowledge data corresponding to the query request from the target storage module, and return the target knowledge data; the hybrid knowledge base stores knowledge data corresponding to at least two different representation morphologies of the same knowledge entity, the same knowledge entity is associated with a unified entity identifier under each representation morphology, and cross-morphological mutual reference relationships are established between different representation morphologies through the entity identifier.

[0014] Optionally, the system further includes: The receiving module is used for the original knowledge source; The multi-form extraction module is used to extract knowledge data corresponding to each knowledge entity in at least two different representation forms from the original knowledge source. The entity identifier association module is used to associate a unified entity identifier with the same knowledge entity under different representation forms, and to establish cross-representation mutual reference relationships between different representation forms. The consistency verification module is used to perform cross-format consistency verification on knowledge data corresponding to different representations of the same knowledge entity. The database module is used to build a hybrid knowledge base based on knowledge data in various valid representations.

[0015] Optionally, the system further includes: a joint assembly return module; The joint assembly return module is used to assemble the target knowledge data into a structured report according to different representation forms and output it when the query request is a composite query.

[0016] As can be seen from the above technical solutions, compared with the prior art, this application has the following advantages: This application provides a multi-form hybrid knowledge base retrieval method based on reasoning intent. First, it receives a user's query request and identifies the reasoning intent corresponding to the query request. Then, it matches the corresponding target representation form according to the reasoning intent. Finally, it determines the target storage module corresponding to the target representation form from the hybrid knowledge base, retrieves the target knowledge data corresponding to the query request from the target storage module, and returns the target knowledge data. The hybrid knowledge base stores knowledge data corresponding to at least two different representation forms for the same knowledge entity. The same knowledge entity is associated with a unified entity identifier in each representation form, and cross-form mutual reference relationships are established between different representation forms through the entity identifier. Thus, through intent-driven form routing and a multi-form mutual reference hybrid knowledge base architecture, it solves the problems of poor adaptability of single-form knowledge bases and fragmented multi-form data, improving retrieval accuracy and efficiency in different reasoning scenarios. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a multi-form hybrid knowledge base retrieval method based on reasoning intent-driven principles, provided in this application embodiment; Figure 2 A knowledge retrieval flowchart provided for an embodiment of this application; Figure 3 A flowchart illustrating a method for constructing a hybrid knowledge base, as provided in an embodiment of this application; Figure 4 A knowledge construction flowchart is provided for an embodiment of this application; Figure 5 A timing diagram for cross-morphological consistency verification provided in an embodiment of this application; Figure 6 A schematic diagram of a three-morphological coexistence structure provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a multi-form hybrid knowledge base retrieval system based on reasoning intent, provided in an embodiment of this application. Detailed Implementation

[0018] As mentioned earlier, existing technologies struggle to balance retrieval accuracy and efficiency across different reasoning scenarios. Specifically, most existing knowledge bases employ a single knowledge representation format for construction and retrieval. This single format can only adapt to specific types of query needs, generally exhibiting poor adaptability when facing different reasoning scenarios such as business deduction, precise anchoring, and relation traversal, easily leading to information redundancy or missing key information. Furthermore, for some knowledge base solutions that utilize multiple formats, the independent storage of each format, inconsistent entity identifiers, and fragmented knowledge data across different formats, coupled with a lack of effective cross-format association mechanisms, easily result in entity misalignment and data inconsistency, making it difficult to balance retrieval accuracy and efficiency across different reasoning scenarios.

[0019] To address the aforementioned issues, this application provides a multi-form hybrid knowledge base retrieval method driven by reasoning intent, comprising: first, receiving a user's query request and identifying the reasoning intent corresponding to the query request; then, matching the corresponding target representation form according to the reasoning intent; finally, determining the target storage module corresponding to the target representation form from the hybrid knowledge base, retrieving the target knowledge data corresponding to the query request from the target storage module, and returning the target knowledge data. The hybrid knowledge base stores knowledge data corresponding to at least two different representation forms for the same knowledge entity. The same knowledge entity is associated with a unified entity identifier in each representation form, and cross-form mutual reference relationships are established between different representation forms through the entity identifier.

[0020] Thus, by using an intent-driven morphological routing and a hybrid knowledge base architecture with multi-morphological cross-referencing, the problems of poor adaptability of single-morphological knowledge bases and fragmentation of multi-morphological data are solved, thereby improving retrieval accuracy and efficiency in different reasoning scenarios.

[0021] It should be noted that the multi-form hybrid knowledge base retrieval method and system based on reasoning intent driven by the embodiments of this application can be applied to the field of artificial intelligence technology. The above are merely examples and do not limit the application field of the multi-form hybrid knowledge base retrieval method and system based on reasoning intent driven by this application.

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0023] This application performs retrieval based on a pre-built hybrid knowledge base, which stores knowledge data in at least two different representations of the same knowledge entity. The same knowledge entity is associated with a unified entity identifier in each representation, and cross-representation mutual reference relationships are established between different representations through the entity identifier.

[0024] Furthermore, the different representation forms include narrative text form, structured entity form, and relation triplet form; The narrative text is natural language text that retains the business context and causal logic chain, and the natural language text is embedded with anchor marks of unified entity identifiers corresponding to knowledge entities. The structured entity form is a structured field object carrying source code path and line number anchors. The structured field object contains a field that is a unified entity identifier for the corresponding knowledge entity. The relation triplet is a subject-predicate-object structure, where the subject and object correspond to knowledge entities and are associated with a unified entity identifier of their respective knowledge entities.

[0025] In practical applications, the retrieval operation of this application relies on a pre-built hybrid knowledge base. This hybrid knowledge base stores knowledge data in at least two different representation forms for each knowledge entity, providing a data foundation for multi-scenario retrieval. For the same knowledge entity, a unique and unified entity identifier is configured across all (knowledge) representation forms. Based on this unified entity identifier, cross-form mutual referencing relationships are established between knowledge data of different forms, enabling mutual indexing and linked queries of data of different forms. The symbol for the entity identifier can be “\(E_i\)”, where i = 1, 2, ..., N; N is a positive integer. Furthermore, in this embodiment, the hybrid knowledge base includes three types of (knowledge) representation forms: The first type is narrative text form, which uses natural language to record the complete business context and causal logic. Anchor markers are embedded within the text, and each anchor is bound to the unified entity identifier of the corresponding knowledge entity, making it easy to quickly locate the business description corresponding to the knowledge entity; The second type is structured entity form, which stores data in structured field objects, with source code paths, line numbers, and other traceability anchors. The objects have specially set fields to store the unified entity identifier of the corresponding knowledge entity, which can quickly read the entity's precise parameters and original traceability basis; The third type is relational triple form, which uses the subject-predicate-object standard structure to record entity relationships. The subject and object in the triple correspond to independent knowledge entities, and both are bound to the unified entity identifier of their respective entities, which can completely restore the business relationships such as dependencies and calls between entities. The notation for narrative text can be “\(F_1(E_i)\)”, a lightweight markup text (Markdown) format; the notation for structured entity form can be “\(F_2(E_i)\)”, where structured field objects refer to JavaScript Object Notation (JSON) objects; the notation for relational triples can be “\(F_3(E_i)\)”; and the notation for source code anchor information can be the “backend_code subfield”. In this way, by unifying entity identifiers to connect the three types of data, the defects of fragmented multi-form knowledge data and chaotic entity correspondences are eliminated. It supports quickly retrieving the corresponding data for other forms of the same entity based on any one form, ensuring that multi-form knowledge content is of the same origin and traceable.

[0026] Figure 1 A flowchart illustrating a multi-form hybrid knowledge base retrieval method based on reasoning intent-driven approaches provided in this application embodiment. Figure 2 This is a knowledge retrieval flowchart provided as an embodiment of this application. (Combined with...) Figure 1 and Figure 2 As shown in the embodiments of this application, a multi-form hybrid knowledge base retrieval method based on reasoning intent-driven methods may include: S101: Receive the user's query request and identify the reasoning intent corresponding to the query request.

[0027] In practical applications, the system first receives a query request from the user. This query request is the search text entered by the user based on business needs, represented by the symbol "q". Then, the query request undergoes semantic parsing and requirement classification to identify the underlying reasoning intent. This reasoning intent is represented by the symbol "\(I(q)\)", which characterizes the user's core search request type, distinguishing whether the user needs complete business causal context, precise entity attribute parameters, or a link between entity relationships. By pre-identifying the reasoning intent, a basis for determining the knowledge representation form for subsequent targeted matching can be provided, avoiding indiscriminate full-database searches and enabling on-demand distribution of search paths, thus reducing the computational cost of retrieving invalid data from the source.

[0028] S102: Match the corresponding target representation form according to the reasoning intent.

[0029] Furthermore, the reasoning intent includes business deduction, precise anchoring, and relationship traversal. The step of matching the corresponding target representation based on the reasoning intent includes: When the reasoning intent corresponding to the query request is identified as a business deduction type, the narrative text form is matched as the corresponding target representation form; When the reasoning intent corresponding to the query request is identified as a precise anchoring class, the structured entity form is matched with the corresponding target representation form. When the reasoning intent corresponding to the query request is identified as a relation traversal class, the form of the matching relation triple is the corresponding target representation form.

[0030] In practical applications, the identified reasoning intent can be combined with a pre-defined mapping rule to match the appropriate target knowledge representation format, thereby achieving precise routing of retrieval paths. Reasoning intents include business deduction, precise anchoring, and relationship traversal. For example, when a query request contains patterns such as "why," "because," "reason," or "design intention," the reasoning intent should correspond to the business deduction category; when a query request contains patterns such as "interface path," "field name," "line number," or "specific value," the reasoning intent corresponds to the precise anchoring category; and when a query request contains patterns such as "which," "dependency," "call chain," or "scope of influence," the reasoning intent corresponds to the relationship traversal category. This application's embodiment categorizes user retrieval requests into these three types of reasoning intents, with different types of reasoning intents corresponding to knowledge storage formats with different adaptation advantages. Specifically, if the inference intent is identified as business deduction, it means the user needs continuous descriptive information such as complete business processes and causal logic. Therefore, narrative text is matched as the target representation form, relying on the complete context to support the business logic analysis. If the inference intent is identified as precise anchoring, it means the user needs deterministic field information such as precise entity parameters and code source location. Therefore, structured entity is matched as the target representation form to quickly retrieve standardized and accurate data. If the inference intent is identified as relation traversal, it means the user needs to query the connection links such as calls and dependencies between multiple entities. Therefore, relation triples are matched as the target representation form to facilitate traversing entity relationships. It can be understood that the inference intent is selected from business deduction, precise anchoring, and relation traversal, and the target representation form is selected from narrative text, structured entity, and relation triples. One query request can correspond to multiple inference intents and multiple (knowledge) representation forms. Thus, by using a matching mechanism that corresponds one-to-one between reasoning intent and knowledge form, knowledge data of the appropriate type can be retrieved only in a targeted manner, without having to traverse all multi-form knowledge base content, avoiding interference from irrelevant information, and simultaneously improving retrieval speed and result matching accuracy.

[0031] S103: Determine the target storage module corresponding to the target representation form from the hybrid knowledge base, retrieve the target knowledge data corresponding to the query request from the target storage module, and return the target knowledge data.

[0032] In practical applications, after matching the target representation, the system locates the target storage module specific to that representation based on the pre-defined correspondence between the representation and the storage module. It is understood that in this embodiment, data of different knowledge representations are stored independently in partitions, with each type of storage module storing only a single type of knowledge data, achieving physical data isolation. Subsequently, the system retrieves target knowledge data matching the current query request only within the target storage module, eliminating the need to scan the full data of other storage modules in the hybrid knowledge base, significantly reducing the retrieval scope and computational cost. After the retrieval is complete, the filtered target knowledge data is directly output and returned. Thus, this partitioned targeted retrieval method avoids redundant computation caused by cross-module full-domain retrieval, shortens retrieval response time, and outputs only knowledge content suitable for the user's reasoning needs, reducing interference from irrelevant information and improving the effectiveness of retrieval results.

[0033] Furthermore, since query requests are not all the same, this application embodiment can describe one possible query request.

[0034] In one instance, the method further includes: When the query request is a compound query, the target knowledge data is assembled into a structured report according to different representation forms and then output.

[0035] In practical applications, when a user's query request is identified as a composite query, it means that the query simultaneously contains at least two types of reasoning needs, corresponding to at least two of the three reasoning intentions: business deduction, precise anchoring, and relationship traversal. Alternatively, if it doesn't belong to any of these three categories, knowledge data in a single representation format cannot fully meet the user's retrieval needs. The system retrieves multiple sets of target knowledge data belonging to different representation formats from various corresponding storage modules. Then, it systematically integrates and typesets the multi-source data according to the classification rules of narrative text, structured entities, and relation triples, generating a clear and structured report. Finally, the integrated report is uniformly output. This processing method can summarize multi-dimensional information such as business logic, precise parameters, and entity relationships in one go, eliminating the need for users to initiate multiple independent searches. This solves the problem of scattered information and cumbersome viewing in multi-demand scenarios, improving the efficiency of knowledge reading and use in composite query scenarios.

[0036] Furthermore, taking the knowledge engineering of a supply chain finance platform as an example, its knowledge scale can include: narrative text, approximately 66 product business documents (including a global knowledge map and cross-product line analysis); JSON entities, approximately 5,000, covering more than 30 entity types (interfaces / processes / configurations / verification codes, etc.); and triple relationships, approximately 15,000 (calls / inclusions / dependencies / changes). The corresponding unified entity identification scheme is E001~E, approximately 5,000, cross-referenced across three forms. The corresponding intent reasoning test data are as follows: business deduction, corresponding to narrative text form, average token consumption of approximately 800, accuracy rate of 92%; precise anchoring, corresponding to structured entity form, average token consumption of approximately 250, accuracy rate of 98%; relationship traversal, corresponding to relationship triple form, average token consumption of approximately 400, accuracy rate of 95%; full return (comparison), corresponding to single text, average token consumption of approximately 3,500, accuracy rate of 78%. As can be seen, intent routing reduces token consumption by approximately 60% to 93% and improves accuracy by approximately 14% to 20% compared to full return.

[0037] Figure 3 A flowchart illustrating a method for constructing a hybrid knowledge base, as provided in an embodiment of this application. Figure 4 This is a flowchart illustrating a knowledge construction process as provided in an embodiment of this application. (Combined with...) Figure 3 and Figure 4 As shown in the embodiments of this application, a method for constructing a hybrid knowledge base may include: S201: Receive the original knowledge source.

[0038] In practical applications, the system pre-receives various raw knowledge sources. These raw knowledge sources are the original materials carrying business-related information, and may include various types of files such as product requirement documents, source code, and business process diagrams. The system reads all the original information, including text, parameters, and relationships, from the raw knowledge sources as the basic material for subsequent extraction and decomposition to generate multi-form knowledge data. This completes the data input step before knowledge is stored in the database, providing raw data support for building a hybrid knowledge base containing multiple representation forms.

[0039] S202: Extract knowledge data in at least two different representation forms corresponding to each knowledge entity from the original knowledge source.

[0040] In practical applications, the system performs semantic parsing, information decomposition, and classification extraction on the received raw knowledge sources. For each knowledge entity, it extracts and generates knowledge data in at least two different representation formats. Specifically, it extracts complete business descriptions from the raw materials to form narrative text, extracts fixed parameters, code locations, and other definite information to form structured entity data, and mines relationship triples and other logical connections between entities. By parsing the raw knowledge source once and simultaneously generating multiple forms of supporting knowledge, it ensures that all types of information for the same knowledge entity originate from the same set of raw materials. This guarantees the consistency and uniformity of multi-form knowledge content from the source, providing standardized multi-form data for building a hybrid knowledge base with cross-form mutual referencing capabilities.

[0041] S203: Associate a unified entity identifier with the same knowledge entity under different representation forms, and establish cross-representation mutual reference relationships between different representation forms.

[0042] In practical applications, for knowledge entities belonging to different representation forms but corresponding to the same business object, the system assigns a unique and unified entity identifier for binding and association, ensuring that the same knowledge entity uses the exact same identifier number in the corresponding narrative text form, structured entity form, and relation triplet form. Based on this unified entity identifier, cross-form mutual referencing relationships are established between data of different forms, enabling data of any form to be quickly indexed and retrieved from other forms of the same entity through the entity identifier. This breaks down data isolation barriers between different knowledge forms, realizes multi-form information linkage query, and provides a unified matching basis for subsequent cross-form consistency verification.

[0043] S204: Perform cross-format consistency verification on knowledge data of different representations corresponding to the same knowledge entity.

[0044] Figure 5 This document provides a timing diagram for cross-morphological consistency verification in an embodiment of this application. Combined with... Figure 5As shown, after binding multi-form knowledge data with unified entity identifiers and establishing cross-form mutual reference relationships, the consistency verification module in the system will perform cross-form consistency verification operations on all representation forms of knowledge data under the same knowledge entity name, and obtain the corresponding cross-form consistency status, denoted as "\(C(E_i)\)". The extracted content includes extracting the entity identifier (entity identifier anchor point) corresponding to the target knowledge entity from the storage module corresponding to the knowledge data in narrative text form, extracting the entity identifier (entity_id field) corresponding to the target knowledge entity from the storage module corresponding to the knowledge data in structured entity form, and extracting the entity identifier (subject and object identifiers) corresponding to the target knowledge entity from the storage module corresponding to the knowledge data in relation triplet form. Verification methods include finding the intersection / difference. This verification relies on preset consistency constraint rules and uses the unified entity identifier as the matching benchmark. It simultaneously checks the content matching, information completeness, and traceability consistency of three types of corresponding data: text form, structured entity form, and relation triple form. It identifies abnormal issues such as missing information, content contradictions, and inconsistent traceability information among knowledge data in different representation forms, ensuring that the data content of multiple representation forms under the same knowledge entity is from the same source and logically consistent, and avoiding the distortion of subsequent search results due to the asynchrony of multiple forms of data.

[0045] Furthermore, since the rules followed for cross-morphological consistency verification are not entirely the same, this application embodiment can describe one possible rule.

[0046] In one case, the cross-morphological consistency verification follows consistency constraint rules; the consistency constraint rules include: entity identifier uniqueness rule, field synchronization rule, relationship integrity rule, and evidence mutual citation rule; The unique entity identifier rule requires that the same knowledge entity use the same entity identifier in all representation forms; The field synchronization rule requires that key fields in the structured entity form have corresponding descriptions in the narrative text form; The relation integrity rule requires that the entity relations in the relation triple form have corresponding causal explanations in the narrative text form; The evidence citation rule requires that different representations of the same knowledge entity share the same source evidence information.

[0047] In practical applications, the system rigorously verifies multi-form knowledge data item by item according to multiple preset consistency constraint rules. These four constraint rules ensure data consistency from four dimensions: unique entity identifiers, field synchronization, relationship integrity, and mutual citation of evidence. Specifically, the unique entity identifier rule requires that the same knowledge entity share the same entity identifier across all representations in narrative text, structured entities, and relation triples, preventing entity number confusion and misalignment. The field synchronization rule verifies key fields in structured entities, ensuring corresponding matching text descriptions exist in the narrative text and preventing the disconnect between structured data and business text content. The relationship integrity rule verifies entity dependencies, calls, and other associations recorded in triples, requiring corresponding business causal statements to support them in the narrative text, preventing the generation of unfounded entity associations. The mutual citation of evidence rule constrains all forms of data for the same knowledge entity to share a single evidence chain, namely, source code paths and line numbers, ensuring that multi-form knowledge data originates from the same original material. Key fields are the core business parameters, attribute information, and source code tracing and positioning fields carried by the structured entity, such as source code paths and line numbers, and are the core data for achieving accurate retrieval. In this way, through multi-layer rule joint verification, problems such as missing information, content contradictions, and inconsistent sources between different representation forms are eliminated in all aspects, ensuring that the multi-form data in the hybrid knowledge base is logically unified and traceable.

[0048] S205: Construct a hybrid knowledge base based on the knowledge data of each representation form that has passed verification.

[0049] In practical applications, after all cross-morphological consistency checks are completed and the results are satisfactory, the multi-morphological knowledge data representing the same knowledge entity satisfies all constraint rules and does not exhibit anomalies such as incorrect identification, disjointed content, missing relationships, or inconsistent source attribution. The system will then classify and store all validated narrative text, structured entity, and relation triplet knowledge data according to the morphological partitioning storage logic, building a hybrid knowledge base with unified entity identification and cross-morphological referencing capabilities. This rule-based validation before data entry ensures logical consistency, information matching, and source consistency of the multi-morphological knowledge data within the database from the source, preventing unqualified and corrupted data from affecting subsequent retrieval accuracy and providing a high-quality, standardized data source for subsequent targeted retrieval based on reasoning intent.

[0050] Furthermore, since different verification results correspond to different processing methods, this application embodiment can describe one possible processing method.

[0051] In one instance, the method further includes: When the cross-morphological consistency check results show data inconsistency, an alarm list containing missing or contradictory entity information is output.

[0052] In practical applications, the system checks knowledge data in different representations of the same knowledge entity against each consistency constraint rule. If any data inconsistency is detected, such as inconsistent entity identifiers, missing text descriptions for key fields, lack of causal explanations for entity relationships, or mismatched tracing evidence, the check item is deemed abnormal. The system automatically captures the knowledge entity numbers, anomaly types, and corresponding knowledge form information for missing or contradictory content, integrates them to generate a complete alarm list containing information on missing or contradictory entities, and outputs it externally. This allows maintenance personnel to quickly locate entities with data defects, perform targeted corrections to original knowledge materials, and re-extract knowledge data, eliminating information conflicts in multiple forms of knowledge data from the source and ensuring the consistency of the knowledge data entering the database.

[0053] Figure 6 This is a schematic diagram of a three-morphological coexistence structure provided in an embodiment of this application. (Combined with...) Figure 6 As shown, the same knowledge entity can simultaneously correspond to knowledge data in three different representation forms: narrative text form, structured entity form, and relation triple form. The same knowledge entity is associated with a unified entity identifier across all representation forms, and cross-representational referencing relationships are established between different representation forms through this entity identifier.

[0054] In summary, this application first receives a user's query request and identifies the reasoning intent corresponding to the query request. Then, it matches the corresponding target representation form based on the reasoning intent. Finally, it determines the target storage module corresponding to the target representation form from the hybrid knowledge base, retrieves the target knowledge data corresponding to the query request from the target storage module, and returns the target knowledge data. The hybrid knowledge base stores knowledge data corresponding to at least two different representation forms for the same knowledge entity. The same knowledge entity is associated with a unified entity identifier in each representation form, and cross-representational referencing relationships are established between different representation forms through the entity identifier. Thus, through intent-driven form routing and a hybrid knowledge base architecture with multi-form referencing, the application solves the problems of poor adaptability of single-form knowledge bases and fragmented multi-form data, improving retrieval accuracy and efficiency in different reasoning scenarios.

[0055] Figure 7 This is a schematic diagram illustrating the structure of a multi-form hybrid knowledge base retrieval system driven by reasoning intent, provided as an embodiment of this application. Combined with... Figure 7 As shown, the multi-form hybrid knowledge base retrieval system 700 includes: The reasoning intent recognition module 701 is used to receive a user's query request and recognize the reasoning intent corresponding to the query request; Matching module 702 is used to match the corresponding target representation form according to the reasoning intent; The morphological routing retrieval module 703 is used to determine the target storage module corresponding to the target representation morphology from the hybrid knowledge base, retrieve the target knowledge data corresponding to the query request from the target storage module, and return the target knowledge data; the hybrid knowledge base stores knowledge data corresponding to at least two different representation morphologies of the same knowledge entity, the same knowledge entity is associated with a unified entity identifier under each representation morphology, and cross-morphological mutual reference relationships are established between different representation morphologies through the entity identifier.

[0056] Furthermore, the different representation forms include narrative text form, structured entity form, and relation triplet form; The narrative text is natural language text that retains the business context and causal logic chain, and the natural language text is embedded with anchor marks of unified entity identifiers corresponding to knowledge entities. The structured entity form is a structured field object carrying source code path and line number anchors. The structured field object contains a field that is a unified entity identifier for the corresponding knowledge entity. The relation triplet is a subject-predicate-object structure, where the subject and object correspond to knowledge entities and are associated with a unified entity identifier of their respective knowledge entities.

[0057] Furthermore, the reasoning intent includes business deduction, precise anchoring, and relationship traversal. Matching module 702 is specifically used for: When the reasoning intent corresponding to the query request is identified as a business deduction type, the narrative text form is matched as the corresponding target representation form; When the reasoning intent corresponding to the query request is identified as a precise anchoring class, the structured entity form is matched with the corresponding target representation form. When the reasoning intent corresponding to the query request is identified as a relation traversal class, the form of the matching relation triple is the corresponding target representation form.

[0058] As one implementation method, regarding how to construct a hybrid knowledge base, the aforementioned multi-form hybrid knowledge base retrieval system 700 further includes: The receiving module is used for the original knowledge source; The multi-form extraction module is used to extract knowledge data corresponding to each knowledge entity in at least two different representation forms from the original knowledge source. The entity identifier association module is used to associate a unified entity identifier with the same knowledge entity under different representation forms, and to establish cross-representation mutual reference relationships between different representation forms. The consistency verification module is used to perform cross-format consistency verification on knowledge data corresponding to different representations of the same knowledge entity. The database module is used to build a hybrid knowledge base based on knowledge data in various valid representations.

[0059] Furthermore, the cross-morphological consistency verification follows consistency constraint rules; these consistency constraint rules include: entity identifier uniqueness rule, field synchronization rule, relationship integrity rule, and evidence mutual citation rule; The unique entity identifier rule requires that the same knowledge entity use the same entity identifier in all representation forms; The field synchronization rule requires that key fields in the structured entity form have corresponding descriptions in the narrative text form; The relation integrity rule requires that the entity relations in the relation triple form have corresponding causal explanations in the narrative text form; The evidence citation rule requires that different representations of the same knowledge entity share the same source evidence information.

[0060] As one implementation method, regarding how to output structured reports, the aforementioned multi-format hybrid knowledge base retrieval system 700 further includes: a joint assembly return module; this joint assembly return module is used for: When the query request is a compound query, the target knowledge data is assembled into a structured report according to different representation forms and then output.

[0061] As one implementation method, the above-mentioned data entry module is further used to handle data with inconsistent verification results, specifically for: When the cross-morphological consistency check results show data inconsistency, an alarm list containing missing or contradictory entity information is output.

[0062] In summary, this application first receives a user's query request and identifies the reasoning intent corresponding to the query request. Then, it matches the corresponding target representation form based on the reasoning intent. Finally, it determines the target storage module corresponding to the target representation form from the hybrid knowledge base, retrieves the target knowledge data corresponding to the query request from the target storage module, and returns the target knowledge data. The hybrid knowledge base stores knowledge data corresponding to at least two different representation forms for the same knowledge entity. The same knowledge entity is associated with a unified entity identifier in each representation form, and cross-representational referencing relationships are established between different representation forms through the entity identifier. Thus, through intent-driven form routing and a hybrid knowledge base architecture with multi-form referencing, the application solves the problems of poor adaptability of single-form knowledge bases and fragmented multi-form data, improving retrieval accuracy and efficiency in different reasoning scenarios.

[0063] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-form hybrid knowledge base retrieval method based on reasoning intent, characterized in that, The method includes: Receive user query requests and identify the reasoning intent corresponding to the query requests; Match the corresponding target representation form according to the reasoning intent; The target storage module corresponding to the target representation form is determined from the hybrid knowledge base, and the target knowledge data corresponding to the query request is retrieved from the target storage module and returned. The hybrid knowledge base stores knowledge data corresponding to at least two different representation forms of the same knowledge entity. The same knowledge entity is associated with a unified entity identifier in each representation form, and cross-representation mutual reference relationship is established between different representation forms through the entity identifier.

2. The method according to claim 1, characterized in that, The hybrid knowledge base is constructed using the following methods: Receive original knowledge sources; At least two different representations of knowledge data corresponding to each knowledge entity are extracted from the original knowledge source; Associate a unified entity identifier with the same knowledge entity under different representation forms, and establish cross-representation mutual reference relationships between different representation forms; Perform cross-format consistency checks on knowledge data corresponding to different representations of the same knowledge entity; A hybrid knowledge base is constructed based on the knowledge data of each valid representation.

3. The method according to claim 1 or 2, characterized in that, The different representation forms include narrative text form, structured entity form, and relation triplet form; The narrative text is natural language text that retains the business context and causal logic chain, and the natural language text is embedded with anchor marks of unified entity identifiers corresponding to knowledge entities. The structured entity form is a structured field object carrying source code path and line number anchors. The structured field object contains a field that is a unified entity identifier for the corresponding knowledge entity. The relation triplet is a subject-predicate-object structure, where the subject and object correspond to knowledge entities and are associated with a unified entity identifier of their respective knowledge entities.

4. The method according to claim 3, characterized in that, The reasoning intent includes business deduction, precise anchoring, and relationship traversal. The step of matching the corresponding target representation based on the reasoning intent includes: When the reasoning intent corresponding to the query request is identified as a business deduction type, the narrative text form is matched as the corresponding target representation form; When the reasoning intent corresponding to the query request is identified as a precise anchoring class, the structured entity form is matched with the corresponding target representation form. When the reasoning intent corresponding to the query request is identified as a relation traversal class, the form of the matching relation triple is the corresponding target representation form.

5. The method according to claim 1, characterized in that, The method further includes: When the query request is a compound query, the target knowledge data is assembled into a structured report according to different representation forms and then output.

6. The method according to claim 2, characterized in that, The cross-morphological consistency verification follows consistency constraint rules; the consistency constraint rules include: entity identifier uniqueness rule, field synchronization rule, relationship integrity rule, and evidence mutual citation rule; The unique entity identifier rule requires that the same knowledge entity use the same entity identifier in all representation forms; The field synchronization rule requires that key fields in the structured entity form have corresponding descriptions in the narrative text form; The relation integrity rule requires that the entity relations in the relation triple form have corresponding causal explanations in the narrative text form; The evidence citation rule requires that different representations of the same knowledge entity share the same source evidence information.

7. The method according to claim 2, characterized in that, The method further includes: When the cross-morphological consistency check results show data inconsistency, an alarm list containing missing or contradictory entity information is output.

8. A multi-form hybrid knowledge base retrieval system based on reasoning intent, characterized in that, include: The reasoning intent recognition module is used to receive user query requests and identify the reasoning intent corresponding to the query requests; The matching module is used to match the corresponding target representation form according to the reasoning intent; The morphological routing retrieval module is used to determine the target storage module corresponding to the target representation morphology from the hybrid knowledge base, retrieve the target knowledge data corresponding to the query request from the target storage module, and return the target knowledge data; the hybrid knowledge base stores knowledge data corresponding to at least two different representation morphologies of the same knowledge entity, the same knowledge entity is associated with a unified entity identifier under each representation morphology, and cross-morphological mutual reference relationships are established between different representation morphologies through the entity identifier.

9. The system according to claim 8, characterized in that, The system also includes: The receiving module is used for the original knowledge source; The multi-form extraction module is used to extract knowledge data corresponding to each knowledge entity in at least two different representation forms from the original knowledge source. The entity identifier association module is used to associate a unified entity identifier with the same knowledge entity under different representation forms, and to establish cross-representation mutual reference relationships between different representation forms. The consistency verification module is used to perform cross-format consistency verification on knowledge data corresponding to different representations of the same knowledge entity. The database module is used to build a hybrid knowledge base based on knowledge data in various valid representations.

10. The system according to claim 8, characterized in that, The system also includes: a joint assembly return module; The joint assembly return module is used to assemble the target knowledge data into a structured report according to different representation forms and output it when the query request is a composite query.

Images