A knowledge reliability modeling method and system based on temporal validity constraints

By unifying the knowledge representation of multi-source policy data and constructing a time-series effectiveness knowledge structure, the problem of old knowledge entering the target time point processing results after local knowledge changes is solved, and the temporal consistency and reliability of knowledge retrieval and processing are achieved.

CN123068615APending Publication Date: 2026-11-10ZHEJIANG STARSINO INFORMATION TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611586782.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-14
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In existing technologies, after local knowledge undergoes subsequent changes, old knowledge that has ceased to be effective may still be included in the processing results at the target time point, resulting in inconsistencies between the knowledge content and the target time point, making it difficult to limit its participation scope while preserving historical content.

Method used

By unifying the knowledge representation of multi-source policy data, multiple knowledge fragments are formed, their scope of application and state inheritance relationship are determined, a time-series effectiveness knowledge structure is constructed, and effective candidate fragments are screened by combining content association index and time-series effectiveness index with target time point, and erroneous data is identified and the index is corrected in reverse.

Benefits of technology

This approach preserves historical knowledge while limiting its scope of participation, ensuring consistency in the temporal state, traceability of sources, and reliability of processing in the knowledge retrieval and processing process, and preventing fragments of old knowledge that have already ceased to be used from re-entering the processing path.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123068615A_ABST
    Figure CN123068615A_ABST
Patent Text Reader

Abstract

This invention relates to the field of computer data processing technology, and discloses a knowledge trust modeling method and system based on temporal validity constraints. The method includes: acquiring multi-source policy data and performing unified knowledge representation to form multiple first knowledge fragments and determine their effective ranges; forming second knowledge fragments when subsequent changes occur in the multi-source policy data, determining the state inheritance relationship between the first and second knowledge fragments, and constructing a temporal validity knowledge structure by associating it with source-related data and effective ranges; establishing content association indexes and temporal validity indexes based on the temporal validity knowledge structure, generating valid candidate fragments by combining target query data and target time points, establishing a trustworthy processing path to identify erroneous data, and correcting the two types of indexes in reverse; constraining the knowledge participation state at the target time point through fragment-level effective ranges and state inheritance relationships, which can preserve historical knowledge and prevent old knowledge fragments that have ended their effects from re-entering the processing path at the target time point.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and more specifically, to a knowledge reliability modeling method and system based on time-series validity constraints. Background Technology

[0002] As policy documents, business rules, and other time-sensitive knowledge data continue to accumulate, knowledge management systems typically need to simultaneously store historical and current content, and perform knowledge retrieval, rule matching, or inference processing based on user queries. For data that has undergone subsequent changes such as modification, replacement, or discontinuation, the knowledge content corresponding to different time stages may undergo local changes. Therefore, the way the knowledge content is organized and its effective status at the target time point directly affects the subsequent processing results.

[0003] Patent application CN121562837A discloses a method and system for compliance reasoning in large language models based on domain knowledge graph constraints. By labeling entity relation triples with timestamp intervals and confidence levels, it organizes the state changes of the same entity at different points in time into a knowledge evolution tracking graph, and selects currently valid graph knowledge fragments for model reasoning based on temporal constraints. This scheme can perform knowledge calibration by incorporating temporal changes in knowledge during large language model reasoning. Patent application CN120806085A discloses an adaptive optimization method and system for human resources and social security knowledge graphs based on uncertain information. It calculates the credibility of entities and relations in the knowledge graph through policy indexing, fuzzy sets, multi-stage Markov processes, policy timeliness factors, and multi-hop attention layers, and performs version management and incremental updates on the optimized graph structure. This scheme can continuously update the temporal credibility state of knowledge graph nodes and edges.

[0004] However, in knowledge data processing with local subsequent changes, when the source file as a whole remains valid, but some conditions, processing actions, or processing results in the file have been modified, replaced, or stopped by subsequent content, the knowledge content corresponding to the target time point may simultaneously obtain both old content and subsequent content. The old content may even be actually adopted because it is closer to the query expression, causing local knowledge that has ended its effect to enter the current processing result, thereby causing qualification judgment, rule matching, or business processing to adopt knowledge content that is inconsistent with the target time point. Summary of the Invention

[0005] Subsequent changes to policy content do not necessarily affect the entire source document. Some changes may only involve a condition, a processing action, or a processing result, while other content in the source document still needs to participate in subsequent processing. If the knowledge state is still controlled uniformly by the source document, after partial changes, it is difficult to simply shut down the changed content, or to limit the scope of participation of historical content at the target time while retaining historical content. At the same time, content-related searches may still recall old content that has already ceased to be effective. Therefore, it is necessary to implement knowledge content, time-bound scope, and the relationship between changes into knowledge units that can independently participate in processing, and to subject the query entry point to the common constraints of the knowledge state corresponding to the target time.

[0006] To overcome the shortcomings of existing technologies where outdated knowledge, after undergoing subsequent changes, may still be included in the processing results at the target time point, this invention provides a knowledge credibility modeling method and system based on temporal validity constraints. This method forms knowledge fragments by unifying the knowledge representation of multi-source policy data, determines the scope of effect and inheritance relationships of these fragments, constructs a temporal validity knowledge structure, and uses content association indexes and temporal validity indexes combined with the target time point to filter valid candidate fragments. Furthermore, it identifies erroneously entered data based on the credible processing path and reverse-corrects the two types of indexes, enabling historical knowledge to be preserved and traced, while restricting outdated knowledge fragments from participating in the data processing at the corresponding target time point.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A knowledge reliability modeling method based on temporal validity constraints includes:

[0009] Acquire multi-source policy data, perform unified knowledge representation on the multi-source policy data to form multiple first knowledge fragments, determine the scope of the first knowledge fragments, and record the source-related data corresponding to the first knowledge fragments; when the multi-source policy data undergoes subsequent changes, form second knowledge fragments and record the source-related data corresponding to the second knowledge fragments, determine the state inheritance relationship between the first knowledge fragments and the second knowledge fragments, and store the first knowledge fragments, second knowledge fragments, source-related data, scope of scope, and state inheritance relationship in association to construct a temporal effectiveness knowledge structure;

[0010] By leveraging the knowledge structure of time-series effectiveness, we establish content association indexes and time-series effectiveness indexes. We combine the acquired target query data with the target time point to generate valid candidate fragments. Based on the valid candidate fragments, we establish a reliable processing path to identify erroneous data. We then use the erroneous data to reverse-correct the content association indexes and time-series effectiveness indexes.

[0011] A knowledge reliability modeling system based on temporal validity constraints is provided to implement the aforementioned knowledge reliability modeling method based on temporal validity constraints. The system includes:

[0012] The temporal effectiveness knowledge structure construction module is used to acquire multi-source policy data, perform unified knowledge representation on the multi-source policy data, form multiple first knowledge fragments, and determine the scope of the first knowledge fragments; when the multi-source policy data undergoes subsequent changes, second knowledge fragments are formed, the state inheritance relationship between the first knowledge fragments and the second knowledge fragments is determined, and the temporal effectiveness knowledge structure is constructed.

[0013] The trusted index closed-loop correction module is used to establish content association index and time-series effectiveness index using the time-series effectiveness knowledge structure, generate effective candidate fragments by combining the acquired target query data and target time point, establish a trusted processing path based on the effective candidate fragments to identify erroneous data, and use the erroneous data to reverse correct the content association index and time-series effectiveness index.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] This invention unifies the knowledge representation of multi-source policy data, forming a first knowledge fragment of policy content, and using an effective range to characterize the scope of participation of the first knowledge fragment in the time dimension; when subsequent changes occur in the multi-source policy data, a second knowledge fragment is formed, and a state inheritance relationship is established between the first and second knowledge fragments. The first knowledge fragment, the second knowledge fragment, source-related data, effective range, and state inheritance relationship are then stored together, so that changes in local knowledge content can form a continuous temporal effectiveness state at the knowledge fragment level. After establishing content association indexes and time-series effectiveness indexes based on the time-series effectiveness knowledge structure, the target query data can form effective candidate fragments by combining the target time point. This further constrains the content association results by the scope of knowledge fragments and the state inheritance relationship, preventing local old knowledge fragments that have already ended their function from re-entering the processing path of the corresponding target time point simply because of content association when the source file can still participate in processing. Furthermore, a reliable processing path is established based on the effective candidate fragments. The erroneous data that actually enters the reliable processing path is identified, and the erroneous data is used to reverse correct the content association index and the time-series effectiveness index. This ensures that subsequent query entry points continue to correspond to the scope of knowledge fragments and the state inheritance relationship. Thus, while preserving the knowledge content and source association relationship before and after, it achieves the distinction between the knowledge participation state before and after local subsequent changes and the limitation of the knowledge participation scope at the target time point, ensuring that the knowledge retrieval and processing process maintains the consistency of time-series status, traceability of sources, and reliability of processing. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a multi-source policy data access and execution device scenario in this invention;

[0018] Figure 2 This is a flowchart of a knowledge reliability modeling method based on temporal validity constraints in this invention;

[0019] Figure 3 This is a schematic diagram illustrating the formation of a first knowledge fragment based on the content boundaries of the policy content in this invention;

[0020] Figure 4 This is a schematic diagram illustrating the backtracking of knowledge fragment source-related data in this invention;

[0021] Figure 5 This is a flowchart illustrating the source merging process for the multi-source representation in this invention.

[0022] Figure 6 This is a time sequence diagram illustrating the effective period of the project filing knowledge fragment in this invention;

[0023] Figure 7 This is a functional block diagram of a knowledge credibility modeling system based on temporal validity constraints in this invention. Detailed Implementation

[0024] The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. Unless the context clearly indicates otherwise, the singular form also includes the meaning of the plural form. It should be understood that although terms such as first, second, etc., may be used in this invention to describe various information, such information should not be limited to these terms, which are only used to distinguish information of the same type from one another.

[0025] This invention provides a knowledge trust modeling method based on temporal validity constraints. The method can be executed by a server, edge computing device, or cloud platform. The execution device obtains policy content and policy time information from policy release platforms, policy document repositories, historical archives, or business systems through data interfaces, and performs unified knowledge representation, index building, query processing, and index updating through a processor. Scanned documents can first be converted into processable text through character recognition. For example... Figure 1The diagram shown is a schematic of a multi-source policy data access and execution device scenario provided in an embodiment of the present invention. Policy content and policy time information from the policy release platform, policy document library, historical archive library, and business system are input into the execution device via a data interface. For scanned documents, they are first processed into processable text through character recognition before being connected to the data interface. The processor in the execution device performs unified knowledge representation on the accessed data and updates the index.

[0026] This invention uses a corporate policy application scenario as the original scenario. The multi-source policy data can simultaneously store original policy documents and subsequent policy data. The subsequent policy data can modify only certain application conditions, eligibility conditions, processing actions, or processing results from the original policy documents. The device forms knowledge fragments according to the content boundaries of the policy content itself, while simultaneously storing the source location and time information corresponding to the knowledge fragments. When a user initiates a policy query, the target query data includes the query object and the target time point; the query object represents the policy matter to be retrieved, and the target time point limits the policy status to be adopted. The device first forms candidates based on content association, then filters them according to the participation status corresponding to the target time point, and returns any erroneous data discovered during the actual processing to the index update stage.

[0027] In the above application scenarios, such as Figure 2 As shown, the method includes the following steps:

[0028] Step S10: Obtain multi-source policy data, perform unified knowledge representation on the multi-source policy data to form multiple first knowledge fragments, and determine the scope of effect of the first knowledge fragments; when the multi-source policy data undergoes subsequent changes, form second knowledge fragments, determine the state inheritance relationship between the first knowledge fragments and the second knowledge fragments, and construct a temporal effectiveness knowledge structure.

[0029] Specifically, multi-source policy data is a collection of policy content and corresponding time information obtained from different issuing entities, different source documents, or different data presentation formats. Multi-source policy data can include the policy text as well as release information, effectiveness information, modification information, replacement information, cessation of implementation information, and continued use information. Unified knowledge representation converts policy content from different sources and in different formats into a consistent field structure. A unified field at least carries a semantic object representing the policy matter, condition items representing the conditions for its establishment, processing actions representing the handling or execution behavior, and processing results representing the processing conclusion, and is associated with the source location. Unified knowledge representation can be achieved through sentence parsing, clause segmentation, entity recognition, and semantic role recognition, as long as the subsequent content boundary judgment and content matching use the same field structure. For example, when performing unified knowledge representation on the exemplary policy content "If a company has submitted a project plan and provided an equipment list, it can apply for project filing," the semantic object is "project filing," the condition items are "submitted project plan" and "provided equipment list," the processing action is "process," and the processing result is "project filing can be processed." Subsequent policy data refers to policy data that is later in time than the associated policy content and can establish a sequential relationship with the existing knowledge segment through source document identification, the location of modified clauses, modification information, replacement information, information on discontinuation of implementation, or information on continued use. Subsequent changes in multi-source policy data refer to changes in the content, applicable conditions, processing actions, processing results, or participation status of the existing knowledge segment caused by the aforementioned subsequent policy data. A change in participation status also occurs when subsequent policy data forms a new version node or a new source record, and clearly replaces the previous version through replacement information, causing the existing knowledge segment to cease to have a participation status from the new starting point. Policy content that is only published later but does not have a clear sequential revision relationship with the existing knowledge segment is not considered a subsequent change to the existing knowledge segment.

[0030] Data fragments formed from multi-source policy data after unified knowledge representation, based on content boundaries with independent conditional semantics, independent processing actions, or independent processing results, are called first knowledge fragments, serving as fragment-level state carriers. Specifically, during the division, candidate boundaries can be formed first using periods, semicolons, or clause numbers. Then, the condition items, processing actions, and processing results on both sides of the candidate boundaries are read: if one side of the boundary already contains independently identifiable conditions and their corresponding actions or results, it is divided at that boundary; if the latter processing action or result depends on the former condition to be valid, they are merged into the same first knowledge fragment. For example, when a policy statement simultaneously records application conditions and "applicants meeting the aforementioned conditions can obtain the corresponding processing result," the application conditions and the corresponding processing result are retained in the same first knowledge fragment; when a subsequent statement records a processing matter that can be independently judged without depending on the aforementioned conditions, it can form another first knowledge fragment. The time range within which the first knowledge fragment can participate in data processing at the target time point, determined by the time information corresponding to the first knowledge fragment, is called the scope of action. The state of whether a knowledge fragment can participate in data processing at the target time point is called the participation state. A subsequent data segment formed from multi-source policy data that changes later and has a preceding-follower state relationship with the first knowledge segment is called a second knowledge segment. The first and second knowledge segments represent two roles in a preceding-follower state relationship; when the second knowledge segment undergoes further changes, it is processed according to the first knowledge segment in the subsequent preceding-follower state relationship, and a state inheritance relationship is established between it and the subsequent second knowledge segment. For example... Figure 3 The diagram shown is a schematic diagram of policy content forming a first knowledge segment according to content boundaries provided in an embodiment of the present invention. Taking the policy content as sequentially recording application conditions, corresponding processing results, and handling matters as an example, the content dependency relationship is determined at the candidate boundary; when the corresponding processing result depends on the preconditions, the application conditions and the corresponding processing result are retained in the same first knowledge segment; when the handling matters do not depend on the aforementioned conditions, the handling matters are formed into another first knowledge segment.

[0031] The relationship between two knowledge segments, defined by the type of content change between them or by the termination information pointing to the first knowledge segment, is called a state inheritance relationship. This relationship is used to record complete inheritance relationships, partial replacement relationships, condition addition relationships, condition deletion relationships, or termination relationships. The data structure formed by storing the first knowledge segment, the second knowledge segment, source-related data, scope, and state inheritance relationship is called a temporal effectiveness knowledge structure. This structure provides the basis for the participation state at the target time point while preserving historical knowledge content. The temporal effectiveness knowledge structure can be stored using relational tables, graph structures, or key-value structures, as long as it allows locating the corresponding scope, source-related data, and state inheritance relationship from a single knowledge segment.

[0032] When policy content undergoes partial changes, the content text and the participation status of knowledge fragments belong to different data attributes. The first knowledge fragment can continue to be stored in the content storage; however, if subsequent policy data does not explicitly stipulate parallel application, the first knowledge fragment will no longer have a participation status from the point when the subsequent change takes effect. The scope of application and the relationship of status inheritance separate the expression of "whether it is stored" from "whether it can participate at the target time point," and provide a basis for step S20 to find the subsequent knowledge status at the target time point. If the aforementioned fragment-level status information is lacking, step S20 can only obtain the content association result and cannot determine the end boundary of the old knowledge fragment and the search path for the subsequent knowledge fragment.

[0033] Further, step S10 includes:

[0034] Step S11: Obtain multi-source policy data, merge the sources of the multi-source policy data and unify the knowledge representation, form multiple first knowledge fragments according to the content boundaries, and record the source association data corresponding to the first knowledge fragments.

[0035] In this embodiment of the invention, the process of forming the first knowledge fragment and recording source-related data specifically includes: parsing the publishing entity, source document identifier, and content location identifier in multi-source policy data; grouping data whose publishing entity, source document identifier, and content location identifier all point to the same policy content into the same source group; performing a unified knowledge representation on the policy content in the source group; and then dividing the content boundaries according to the conditional semantics that can be independently judged as true or false, the processing actions that can be independently executed, or the processing results that can be independently established. When the two sides of the content boundary correspond to knowledge content that can independently participate in subsequent matching and state judgment, the content is divided at the content boundary; when a piece of content needs to rely on the same preconditions to form a complete judgment, the preconditions and the corresponding processing actions or processing results are retained in the same first knowledge fragment to avoid splitting sentences that cannot be independently established into mutually independent knowledge fragments.

[0036] For example, the device can first form candidate segments from the clause text using periods, semicolons, or clause numbers. Then, for each candidate segment, it generates four types of content fields: semantic object, condition item, processing action, and processing result. If a candidate segment only contains dependency expressions such as "meets the preceding condition" but not the corresponding precondition, it is merged with its predecessor. If a candidate segment can independently determine its semantic object, condition item, and corresponding processing action or result, it is retained as an independent first knowledge segment. This process does not require pre-setting thresholds for segment length, sentence count, or character count, avoiding the disruption of semantic dependencies between conditions and processing results due to fixed-length segmentation.

[0037] The combined relationship between the publishing entity, source document identifier, superior source node, and content location identifier is called source association data. This data is used to record the source location of the first or second knowledge fragment in multi-source policy data and to support subsequent reliable processing path backtracking. The source document identifier can be a policy document number, a unique document number, or a stable document number within the data source; the superior source node can be a chapter, section, article, or attachment node; the content location identifier can be an article number, paragraph number, or a stable fragment number after unified knowledge expression. Source association data does not require a fixed encoding format, as long as it allows backtracking from a knowledge fragment to the corresponding source document and content location, and can identify other knowledge fragments with the same source document identifier. For example... Figure 4 The diagram shown is a schematic representation of the source association data composition for knowledge fragments provided in this embodiment of the invention. The source association data is composed of the publishing entity, the source document identifier, the superior source node, and the content location identifier. The source document identifier can be a policy document number, the superior source node can correspond to a chapter, section, or article, and the content location identifier can correspond to an article number. The first knowledge fragment can be traced back to the corresponding source document through the source association data.

[0038] When the same policy content exists simultaneously in webpage text, electronic document text, and scanned document parsing text, the system first determines whether different data representations point to the same source location based on the publishing entity, source file identifier, and content location identifier. If the source locations are consistent, the semantic objects, condition items, processing actions, and processing results after unified knowledge expression are compared to see if they are consistent. When the source locations are consistent and the four types of content fields are consistent, different data representations are grouped into the same source-related data, without duplicate first knowledge fragments. When the source locations are consistent but the four types of content fields differ, the differing content is retained and the system proceeds to step S12 for time information verification to distinguish between parsing differences and subsequent changes. When time information verification confirms that subsequent policy data clearly points to an existing knowledge fragment and changes the content or participation status of the existing knowledge fragment, it is processed as a subsequent change. When a clear revision relationship cannot be confirmed, no status inheritance relationship is established, and the source association records corresponding to the differing content are retained separately. When the source locations cannot be matched, first knowledge fragments are formed separately, and source association data is recorded separately. The above branches ensure that source attribution does not rely on a preset similarity threshold, while avoiding merging different clauses solely based on text similarity. Figure 5The diagram shown is a flowchart of the multi-source representation source merging process provided by an embodiment of the present invention. When the same policy content exists simultaneously, it is first determined whether different data representations point to the same source location; if they do not point to the same source location, a first knowledge fragment is formed and the source-related data is recorded separately; if they point to the same source location, the four types of content fields—semantic object, condition item, processing action, and processing result—are compared after the unified knowledge expression. If the four types of content fields are consistent, they are merged into the same source-related data; if there are differences, the differences are retained and time information is verified.

[0039] The granularity of the first knowledge segment determines the minimum range within which subsequent scopes can be closed. If the content boundaries are not determined according to independent conditional semantics, independent processing actions, or independent processing results, and the entire source file is treated as a single first knowledge segment, then any local modification will force the entire content to share the same scope. If clauses that cannot be independently established are broken down too finely, the conditions and corresponding processing actions may be assigned to different knowledge segments, and subsequent content matching will not be able to form a stable one-to-one correspondence.

[0040] Step S12: Determine the scope of the first knowledge fragment based on the first knowledge fragment and the source association data corresponding to the first knowledge fragment. When the multi-source policy data undergoes subsequent changes, respond to the modification or replacement information of the subsequent multi-source policy data, form a second knowledge fragment based on the modification or replacement information, update the scope of the first knowledge fragment, and obtain the scope of the second knowledge fragment.

[0041] In this embodiment of the invention, the "subsequent changes in multi-source policy data" adopts the aforementioned definition, and the judgment basis is whether the subsequent policy data clearly points to an existing knowledge fragment and changes the content or participation status of the existing knowledge fragment; data that is only published later but does not have a revision relationship between the previous and subsequent versions will not be used to form a second knowledge fragment.

[0042] In this embodiment of the invention, the process of determining the scope of action and forming the second knowledge fragment specifically includes: locating policy release information and effectiveness information based on the source-related data corresponding to the first knowledge fragment; obtaining the starting point of action of the first knowledge fragment from the effectiveness information; then checking modification information, replacement information, and termination information along the same source-related data and subsequent policy data that can point to the first knowledge fragment; and obtaining the ending point of action from the action time information that first changed the participation status of the first knowledge fragment. The starting point of action and the ending point of action together determine the scope of action of the first knowledge fragment. If no subsequent policy data appears that can end the participation status of the first knowledge fragment, the ending point of action remains undefined, and the first knowledge fragment continues to have a participation status from the starting point of action. If the policy data only records the release time but not the effectiveness time, and the policy text or release page does not explicitly state "effective from the date of release" or other content that can determine the starting point of action, then the starting point of action is marked as pending verification, and the release time is not directly used to replace the effectiveness time; after obtaining explicit effectiveness information directly corresponding to the policy document, it is then written into the scope of action. During the initial period marked as pending verification, the corresponding knowledge fragment is not identified as having a participation status until clear effectiveness information is obtained before it is included in the target time-end screening. This avoids confusing the release time with the effectiveness time, which could cause the knowledge fragment to enter the participation status prematurely.

[0043] To prevent adjacent knowledge states from participating repeatedly at the same target time point, this embodiment of the invention adopts a boundary convention of inclusion at the beginning and exclusion at the end for adjacent action intervals: when a subsequent change takes effect from time T, the action interval of the first knowledge segment ends before time T, and the action interval of the second knowledge segment begins from time T; when the target time point is equal to time T, the second knowledge segment has an action state, and the first knowledge segment does not have an action state. When subsequent policy data explicitly stipulates that old content and new content are applicable in parallel within the same time range, the action intervals are determined separately based on the parallel time range recorded in the subsequent policy data, and the aforementioned adjacent boundary convention is not mandatory. The boundary convention is only used in cases where the policy data does not separately stipulate parallel application. Here, time T represents the time point at which the modification, replacement, or other subsequent changes recorded in the subsequent policy data begin to take effect. The precision of time T follows the time precision of the source policy data itself; when the source only records the date, the date is used as the comparison unit; when the source records a specific time, the corresponding time is used as the comparison unit, without artificially expanding date-level information to a finer time precision.

[0044] In response to subsequent changes in multi-source policy data, including modifications or replacements, the device first locates the corresponding first knowledge fragment based on the source document identifier, the location of the modified clause, and the semantic object after unified knowledge representation. Then, it forms a second knowledge fragment from the modified or replaced policy content according to the same content boundaries as in step S11. The device records the source association data corresponding to the second knowledge fragment based on the issuing entity, source document identifier, superior source node, and content location identifier in the subsequent policy data. The device updates the end-of-use time of the first knowledge fragment according to the effective time recorded in the modification or replacement information, and uses the same effective time as the start-of-use time of the second knowledge fragment. If no further modifications, replacements, or termination information appear for the second knowledge fragment, the end-of-use time remains undefined; if further changes occur, a subsequent second knowledge fragment is formed in the same manner.

[0045] If subsequent policy data replaces the entire content of the first knowledge fragment, only the scope of the replaced first knowledge fragment is closed, and a corresponding second knowledge fragment is formed. If subsequent policy data only changes some of the first knowledge fragments among multiple first knowledge fragments with the same source document identifier, only the end point of the changed first knowledge fragments is updated, while other first knowledge fragments with the same source document identifier but unchanged retain their original scope. When subsequent policy data explicitly stops executing the first knowledge fragment, the end point of the first knowledge fragment is determined based on the execution time information recorded in the cessation information, and a second knowledge fragment without actual knowledge content is not formed. When subsequent policy data explicitly continues to use the first knowledge fragment and no different knowledge content is formed, the scope of the first knowledge fragment is maintained, and a second knowledge fragment with the same content is not formed repeatedly. "Continues to use" is not equivalent to "complete inheritance relationship" in step S13. If the information continues to be used only to confirm that the original knowledge content remains valid and no new content version has been formed, a second knowledge fragment will not be formed. A complete inheritance relationship applies when subsequent policy data has formed a new version node or a new source record, and the replacement information of subsequent policy data clearly states that the new version replaces the previous version and can determine the new starting point of action. In this case, a second knowledge fragment is formed according to the same content boundary as in step S11, the source association data corresponding to the second knowledge fragment is recorded, and the ending point of action of the first knowledge fragment is updated with the starting point of action of the second knowledge fragment. When the four types of content fields after unified knowledge expression—semantic object, condition item, processing action, and processing result—are all consistent with the previous knowledge fragment, it is determined to be a complete inheritance relationship.

[0046] For example, the first knowledge fragment corresponding to project filing takes effect from January 1, 2023, and subsequent policy data effective from March 1, 2025, deletes the project plan submission conditions from the first knowledge fragment. The device generates a second knowledge fragment based on the modified information, deleting the project plan submission conditions. The effective period of the first knowledge fragment is defined as from January 1, 2023, to March 1, 2025, and the effective period of the second knowledge fragment is defined as from March 1, 2025, continuing until subsequent changes occur. At the target time point of March 1, 2025, only the second knowledge fragment is involved; at the target time point of June 1, 2024, only the first knowledge fragment participates according to the aforementioned time sequence. In this example, if subsequent policy data only deletes the project plan submission conditions, while the semantic object, retained condition items, processing actions, and processing results are all maintained, then step S13 determines the preceding and following knowledge fragments as a condition deletion relationship; if deleting the project plan submission conditions also changes the processing action or processing result, then it is classified as a partial replacement relationship, no longer classified as a condition deletion relationship. Figure 6 The figure shown is a time sequence diagram illustrating the effective range of the project filing knowledge fragment provided in an embodiment of the present invention. Figure 6 The time intervals of the first and second knowledge segments are represented by time. The conditional deletion relationship formed after the submission of the subsequent policy data deletion project plan, which took effect on March 1, 2025, is shown, as well as the participation status of the corresponding knowledge segments when the target time points are June 1, 2024 and March 1, 2025.

[0047] In step S12, both the time boundary and the knowledge content fall onto the fragment-level object. If the scope of the changed first knowledge fragment is not simultaneously closed when the second knowledge fragment is formed, the first and second knowledge fragments will remain in an active state simultaneously within the modified time range; even if the target time point is obtained in step S21, the old state cannot be excluded from the time dimension. Conversely, if a local modification causes all first knowledge fragments with the same source file identifier to close simultaneously, the unchanged clauses will be incorrectly excluded. Step S12 limits subsequent changes to the first knowledge fragments that have actually changed and connects adjacent knowledge states into a continuous time series, providing the temporally aligned preceding and following knowledge fragments for step S13 to determine the change type.

[0048] Step S13: Perform content matching on the first knowledge fragment and the second knowledge fragment to determine the state inheritance relationship between the first knowledge fragment and the second knowledge fragment. Associate and store the first knowledge fragment, the second knowledge fragment, the source associated data, the scope of application, and the state inheritance relationship to construct a temporal effectiveness knowledge structure. The content matching includes semantic object matching, condition item matching, processing action matching, and processing result matching.

[0049] In this embodiment of the invention, the process of determining the state inheritance relationship specifically includes: reading the semantic objects, condition items, processing actions, and processing results of the first and second knowledge segments, which have already formed a temporal relationship in step S12, after unifying their knowledge expressions. Semantic object matching is used to determine whether the two knowledge segments process the same policy matter, which can be determined by the semantic object identifier or the result of the matter name normalization; condition item matching is used to compare the retention, addition, deletion, and content changes in the condition item set; processing action matching is used to compare whether the processing actions of the two knowledge segments are consistent; and processing result matching is used to compare whether the processing results of the two knowledge segments are consistent.

[0050] To ensure the uniqueness of the state inheritance relationship classification results, this embodiment of the invention determines the state inheritance relationship in the following order: When subsequent policy data contains explicit information indicating termination of execution, the termination relationship is determined first. When there is no information indicating termination of execution and the semantic object matching result is not the same policy matter, a state inheritance relationship is not established between the two knowledge fragments, and the two knowledge fragments are stored according to their respective source-related data and scope. When the semantic object matching result is the same policy matter, if the condition item, processing action, and processing result are consistent after unified knowledge expression, a complete inheritance relationship is determined; if only a new condition item is added and the original condition item, processing action, and processing result are maintained, a condition addition relationship is determined; if only a condition item is deleted and the condition item, processing action, and processing result are retained, a condition deletion relationship is determined; except for complete inheritance, condition addition, and condition deletion relationships, as long as the content of the condition item, processing action, or processing result changes, a partial replacement relationship is determined. When addition and deletion occur simultaneously, or when a change in condition item is accompanied by a change in processing action or processing result, it is classified as a partial replacement relationship. For termination relationships, the state inheritance relationship is recorded on the first knowledge fragment that has been terminated, and it is not required to form a second knowledge fragment without actual knowledge content.

[0051] Condition matching can express condition items as structured items of "condition object - comparison relationship - condition content", and then compare the set relationships of structured items; processing actions and processing results can be compared using normalized action identifiers and result identifiers respectively. When using structured fields for set comparison, complete inheritance, addition only, deletion only, and other mutually exclusive set relationships do not require setting semantic similarity thresholds. When there are synonymous rewrites in text descriptions, the same structured fields can be obtained first using synonym normalization, entity linking, or semantic role recognition in the unified knowledge expression stage, and then proceed to set relationship judgment. When field normalization cannot be completed and the same policy matter cannot be confirmed, state inheritance relationships are not forcibly established to avoid linking knowledge fragments with similar sources but different actual matters into the same temporal chain. Synonym normalization can be achieved by mapping the abbreviation, full name, or standardized name of the same matter in different documents to the same semantic object identifier through a pre-maintained policy terminology mapping table; entity linking can be constrained using parsed fields such as matter name, source document identifier, and content location identifier. If the same text may map to multiple policy items and the existing fields cannot uniquely determine it, it will be retained as unconfirmed and will not be included in the status inheritance relationship classification.

[0052] After determining the state inheritance relationship, the device writes the first knowledge fragment, the second knowledge fragment, the source-related data, the scope of action, and the state inheritance relationship into the same temporal validity knowledge structure. Except where subsequent policy data explicitly stipulates that the first and second knowledge fragments apply in parallel, for complete inheritance relationships, partial replacement relationships, conditional addition relationships, and conditional deletion relationships, the first knowledge fragment no longer has a participating state from the starting point of the second knowledge fragment's action, and the second knowledge fragment participates in the target time point processing according to its scope of action. When applied in parallel, the first and second knowledge fragments participate in the target time point processing according to their respective scopes of action determined in step S12. For termination relationships, the first knowledge fragment no longer has a participating state from the end point of action, while retaining its historical state before the end point of action. Other first knowledge fragments with the same source document identifier but without subsequent changes do not change their scope of action according to the aforementioned state inheritance relationship.

[0053] State inheritance transforms the concept of "adjacent time periods" into a computable relationship of "how knowledge states change." If there is only an effective range without a state inheritance relationship, and the target time point is equal to or later than the end time point of the old knowledge fragment, step S21 can only exclude the old knowledge fragment but cannot determine which subsequent knowledge fragment to continue checking. If the state inheritance relationship is established solely based on content differences without the effective range of step S12, the preceding and following knowledge fragments lack participation time boundaries. Steps S12 and S13 together form a temporally effective knowledge structure that can be continuously traversed along time and state changes. Based on this, step S20 can both exclude knowledge states that have ended before the target time point and find the corresponding subsequent state at the target time point. A traversable path from the old knowledge fragment to the effective knowledge fragment at the target time point is established through fragment-level effective ranges and state inheritance relationships: the effective range first excludes old states that have ended participation in time, and the state inheritance relationship then provides the direction for locating the subsequent state, ensuring that even if a partially revised old clause is recalled by the content association index due to high text similarity, it cannot be directly used as effective input at the target time point. This constraint separates content relevance from time-based effectiveness, which can reduce the likelihood of old knowledge fragments re-entering the processing path at new time points due to relying solely on content association.

[0054] Step S20: Utilize the time-series effectiveness knowledge structure to establish a content association index and a time-series effectiveness index. Combine the obtained target query data and target time point to generate effective candidate fragments. Based on the effective candidate fragments, establish a reliable processing path to identify erroneous data. Use the erroneous data to reverse correct the content association index and the time-series effectiveness index.

[0055] Specifically, the retrieval structure built from semantic objects, conditional terms, processing actions, processing results, and knowledge fragment identifiers in the temporal effectiveness knowledge structure is called a content association index. It is used to locate potentially related first and second knowledge fragments based on content information in the target query data. The content association index can be implemented using inverted indexes, vector indexes, entity relationship indexes, or a combination of these. The retrieval structure built from knowledge fragment identifiers, scope of action, and state inheritance relationships in the temporal effectiveness knowledge structure is called a temporal effectiveness index. It is used to quickly determine the participating states of a knowledge fragment and locate subsequent knowledge fragments based on the target time point. The temporal effectiveness index can use an interval index organized according to the start and end times of action, and it stores predecessor and successor identifiers for state inheritance relationships.

[0056] Data containing the query object and the target time point is called target query data, used to provide content association conditions and time-series validity conditions. The query object is the policy matter to be retrieved, which can be determined by the matter name, eligibility issues, or policy matter identifiers passed from the business system. The target time point is the time when the policy status needs to be applied, which can be the time when the application occurred, the time when the business action occurred, or a historical retrospective time point. If there is no identifiable target time point in the target query data, the device will prompt for supplementation of the target time point, and will not substitute the device's current time for the unspecified business occurrence time.

[0057] The first or second knowledge fragment generated by the content association index that has not yet completed the target time point participation state verification is called a candidate knowledge fragment; candidate knowledge fragments form a candidate knowledge fragment set. Candidate knowledge fragments that have been verified by the temporal effectiveness index according to the target time point, scope of action, and state inheritance relationship are determined to have a participation state and are called valid candidate fragments, used as trusted input for subsequent knowledge reasoning or rule matching. The record structure formed by the valid candidate fragments used in an actual processing, the corresponding source association data, and the state inheritance relationship is called a trusted processing path, used to store which knowledge fragments were used in this processing and the source and previous / next state relationships of the knowledge fragments. Data that does not meet the target time point participation state but actually enters the trusted processing path, obtained from re-verification by the trusted processing path, is called erroneous data, used to trigger reverse correction by the content association index and temporal effectiveness index.

[0058] The content association index is responsible for forming a candidate range related to the query object, while the time-series effectiveness index is responsible for limiting the participation status according to the target time point. Both indexes are derived from the time-series effectiveness knowledge structure, but index reconstruction, cache refresh, or batch updates may be temporarily out of sync with the time-series effectiveness knowledge structure. Therefore, step S22 directly re-verifies the actual trusted processing path using the time-series effectiveness knowledge structure, and step S23 returns the verification result to the index, so that the index entry can continue to be constrained by the fragment-level participation status in subsequent queries.

[0059] Further, step S20 includes:

[0060] Step S21: Utilize the temporal effectiveness knowledge structure to establish a content association index and a temporal effectiveness index, obtain target query data containing the query object and the target time point, perform association matching between the target query data and the first and second knowledge fragments in the temporal effectiveness knowledge structure based on the content association index to generate a set of candidate knowledge fragments containing candidate knowledge fragments, and use the temporal effectiveness index to verify the scope of application and state inheritance relationship of the candidate knowledge fragments in combination with the target time point, and select effective candidate fragments from them.

[0061] In this embodiment of the invention, the process of generating valid candidate fragments specifically includes: extracting the query object and target time point from the target query data; performing the same field parsing as step S11 (unified knowledge representation) on the query object to obtain the semantic object corresponding to the query object, as well as the condition items, processing actions, or processing results that can be determined from the query object; when the query object is a policy item identifier passed in by the business system, mapping the policy item identifier to the corresponding semantic object. The device uses the semantic object, condition items, processing actions, or processing results corresponding to the query object to access the content association index. The content association index returns a first knowledge fragment and a second knowledge fragment that have the same fields, entity links, or semantic associations with the query object, forming a candidate knowledge fragment set. When using an inverted index, candidates can be generated from inverted items corresponding to the same semantic object identifier or the same term; when using a vector index, candidates can be generated according to the proximity sorting of the vector retrieval structure itself; when using a combined index, knowledge fragments returned by different indexes can be deduplicated according to the knowledge fragment identifier. The candidate knowledge fragment set is only used as input for temporal validity verification and does not directly enter knowledge reasoning or rule matching. Among them, the vector index is only used to expand the candidate recall and does not directly determine whether a knowledge fragment has a participation status based on the vector similarity. If the specific implementation requires limiting the number of candidate recalls per instance, the recall number can be gradually increased on the test data consisting of known query objects and their corresponding historical and subsequent versions, until further increasing the recall number no longer increases the coverage of known versions of the same query object. Then, the corresponding recall number is used as the engineering configuration value for the index. This configuration value only affects the retrieval overhead and does not replace the verification of the scope and state inheritance relationship.

[0062] For each candidate knowledge fragment, the device uses the temporal effectiveness index to read the scope and state inheritance relationship. If the target time point is within the scope of the candidate knowledge fragment, and there is no complete inheritance relationship, partial replacement relationship, condition addition relationship, condition deletion relationship, or action termination relationship that would cause the candidate knowledge fragment to end its participation in the state by the target time point, then the candidate knowledge fragment is determined as a valid candidate fragment. If the starting time point of the candidate knowledge fragment is marked as pending verification, the candidate knowledge fragment is not determined as a valid candidate fragment; if the target time point is earlier than the starting time point of the candidate knowledge fragment, the candidate knowledge fragment is not determined as a valid candidate fragment. If the target time point is equal to or later than the ending time point of the candidate knowledge fragment, if there is no subsequent second knowledge fragment, the candidate knowledge fragment is not determined as a valid candidate fragment; if there is a subsequent second knowledge fragment, the subsequent second knowledge fragment is located along the state inheritance relationship, and the verification within the same scope continues.

[0063] When searching along state inheritance relationships, a subsequent second knowledge fragment can still enter the target time point verification even if it does not pass the initial association match of the content association index. When multiple state inheritance relationships exist consecutively, the search proceeds sequentially according to the predecessor and successor identifiers until a second knowledge fragment with a participating state and whose target time point is within the scope of effect is found, or until the state inheritance relationship has no successor. If the same first knowledge fragment forms multiple simultaneously existing second knowledge fragments in subsequent policy data, target time point verification is performed on each of the multiple second knowledge fragments. Second knowledge fragments with a participating state and whose target time point is within the scope of effect can all enter the valid candidate fragments. Subsequent knowledge reasoning or rule matching then determines the specific applicable branch based on the corresponding condition items. Multiple simultaneously existing second knowledge fragments only occur when subsequent policy data clearly forms parallel applicable branches; each second knowledge fragment stores its own condition items and scope of effect. The target time point is used to determine whether there is a participating state in time, and the specific query object ultimately enters which parallel branch, which is then determined by the corresponding condition items, thereby avoiding the erroneous merging of parallel applicable content into a single knowledge state.

[0064] Step S21 separates the content association result and the participation status at the target time point into two consecutive judgment steps. Content association can find knowledge fragments that are related to the query object but are in different historical states. The temporal validity verification then determines the state corresponding to the target time point based on the scope of effect and the state inheritance relationship. If the subsequent second knowledge fragment is not searched along the state inheritance relationship, the subsequent content with a large degree of text rewriting may not enter the initial content association result, and the old knowledge fragment will be excluded without any subsequent content to fill it in; if the scope of effect verification is not performed, the historical first knowledge fragment may still be a candidate and directly enter the processing chain. The valid candidate fragments output by step S21 thus simultaneously satisfy the requirements of content association and the participation status at the target time point, and provide a temporally constrained input boundary for step S22 to establish a reliable processing path. Furthermore, the content association index solves the recall problem of "potentially relevant", and the temporal validity index solves the state problem of "whether it can participate at the target time point". The two types of judgments use different data bases. Even if an older knowledge fragment is closer to the query text, it will still be excluded if the target time point has passed its end point. Meanwhile, state inheritance can include subsequent knowledge fragments with significant textual changes in the validation scope. This reduces the historical version bias caused by relying solely on semantic similarity.

[0065] Step S22: Based on the selected valid candidate fragments and their corresponding source association data and state inheritance relationship, a trusted processing path required for actual processing is formed; the first or second knowledge fragment entering the trusted processing path is re-verified using the temporal effectiveness knowledge structure, and the first or second knowledge fragment that does not meet the target time point participation state but actually enters the trusted processing path is recorded as erroneous data.

[0066] In this embodiment of the invention, the process of forming a trusted processing path and identifying erroneous data specifically includes: inputting the valid candidate fragments obtained in step S21 into the knowledge reasoning, rule matching, or policy response generation process, and synchronously recording the knowledge fragment identifier, source association data, target time point, and corresponding state inheritance relationship when the knowledge fragment is actually used. Specifically, a knowledge fragment is determined to be actually used when it is read as a judgment basis by knowledge reasoning or rule matching, or when it is written into the input context of policy response generation. Only knowledge fragments actually used to form judgment conditions, processing actions, or processing results are written into the trusted processing path; valid candidate fragments that are not actually used are not written into the trusted processing path. By recording the actual users, subsequent verification is performed on knowledge fragments that have already affected the processing results, preventing candidates that are only recalled but not adopted from being mistakenly recorded as erroneous data.

[0067] After the trusted processing path is formed, the device no longer uses the temporal validity index as the sole verification source, but directly accesses the temporal validity knowledge structure formed in step S10. For each first or second knowledge fragment in the trusted processing path, the device reads the corresponding scope and state inheritance relationship, and checks the target time point according to the same boundary convention as in step S21. If the target time point is not located within the corresponding scope, the corresponding knowledge fragment is recorded as erroneous data. Even if the target time point falls within the index participation range of the temporal validity index record, if the temporal validity knowledge structure record has already established a complete inheritance relationship, partial replacement relationship, condition addition relationship, condition deletion relationship, or scope termination relationship up to the target time point and ended the participation state of the corresponding knowledge fragment, the corresponding knowledge fragment is also recorded as erroneous data.

[0068] For knowledge fragments with multiple consecutive state inheritance relationships, the device checks the subsequent second knowledge fragment along the temporal validity knowledge structure. If the knowledge fragment actually used in the trusted processing path has ended its participation state by the target time point, and a second knowledge fragment with a participation state at the target time point can be found along the state inheritance relationship, then the actually used old knowledge fragment is recorded as mis-entered data, and the second knowledge fragment that can participate at the target time point is provided to step S23. If a second knowledge fragment with a participation state cannot be found, the knowledge fragment that has ended its participation state is still recorded as mis-entered data, but step S23 only corrects the temporal validity index participation range corresponding to the mis-entered data and does not establish supplementary associations without basis.

[0069] If all knowledge fragments in the trusted processing path meet the participation status at the target time point, then this processing will not generate erroneous data, and the content association index and time-series effectiveness index will remain in their current state. Therefore, the re-verification process has two clear branches: "generating erroneous data" and "not generating erroneous data," and will not necessarily modify the indexes simply because verification is performed.

[0070] The temporal effectiveness knowledge structure stores the complete state relationships determined in steps S12 and S13, while the temporal effectiveness index is a derived structure built for query efficiency. After new policy data is added, the temporal effectiveness knowledge structure can be updated first, followed by asynchronous updates to the interval index, cache, or vector association. Before the derived structure update is complete, the index record may still treat knowledge fragments that have ended their participation state as eligible objects. Step S22 uses the actual reliable processing path as the observation point and the temporal effectiveness knowledge structure as the verification benchmark, accurately assigning objects that "have been recalled and actually used, but the target time point state is not satisfied" to specific knowledge fragments. If the actual reliable processing path is not used as the verification scope, and all candidates are uniformly corrected, historical candidates that have not actually been used will also be treated as abnormal entry points, making it impossible to distinguish between normal historical retention and actual erroneous entry. The erroneous data output by step S22 and the second knowledge fragment with a participation state therefore provide a locatable correction object and alternative entry point for step S23.

[0071] For example, following the project filing example in step S12, the first knowledge fragment had a participation status before March 1, 2025, and the second knowledge fragment has a participation status since March 1, 2025. If the temporal effectiveness knowledge structure has recorded this change, but the temporal effectiveness index cache has not been refreshed, and the target time point is March 1, 2025 or later, the first knowledge fragment may still be indexed and actually enter processing. After directly reading the temporal effectiveness knowledge structure in step S22, it can be determined that the scope of the first knowledge fragment has ended, the first knowledge fragment is recorded as mis-entered data, and the second knowledge fragment is located along the state inheritance relationship. In this example, the judgment basis for mis-entered data is not whether the policy response result is correct, but whether the knowledge fragment actually used meets the participation status at the target time point, thus forming a repeatable program judgment.

[0072] Step S23: Determine the first or second knowledge fragment corresponding to the erroneous data, the target time point, and the state inheritance relationship. If there is a second knowledge fragment with a participating state, further determine the second knowledge fragment with a participating state. Based on the relationship between the target time point and the corresponding effective interval of the erroneous data, update the index participation range of the knowledge fragment corresponding to the erroneous data in the time-series effectiveness index, and establish a supplementary association between the second knowledge fragment with a participating state and the query object to complete the closed-loop correction of the content association index and the time-series effectiveness index.

[0073] In this embodiment of the invention, the time range within which a knowledge fragment is allowed as a candidate entry point for a target time point in the temporal effectiveness index is called the index participation range. This range is used to convert the scope of action in the temporal effectiveness knowledge structure into a time range that the index layer can directly determine. The additional content association between a second knowledge fragment with a participating state, derived from the state inheritance relationship and recorded, and the query object is called supplementary association. This association is used to allow subsequent second knowledge fragments to directly enter candidate generation or prioritize temporal effectiveness verification in subsequent processing of the same query object, without relying on subsequent text and query text forming the same degree of semantic association again. Supplementary association only adds an entry point from the query object to the subsequent second knowledge fragment; it does not change the scope of action or the state inheritance relationship of the second knowledge fragment. The index participation range is a derived representation of the scope of action in the retrieval layer, not an independently set validity period parameter. When the index is first established, the index participation range is consistent with the scope of action recorded in the temporal effectiveness knowledge structure. After a correction of an error, the index participation range can only be shrunk or subsequent entry points added based on the confirmed start and end points of action and state inheritance relationships in the temporal effectiveness knowledge structure; the effective time range of the knowledge fragment is not expanded separately through manual experience.

[0074] The device first locates the first or second knowledge fragment based on the knowledge fragment identifier in the mistakenly entered data, and then reads the inheritance relationship between the target time point and the corresponding state. If a state inheritance relationship exists between the knowledge fragment corresponding to the mistakenly entered data, the search continues along the state inheritance relationship until a second knowledge fragment with a participating state is determined within the scope of the target time point, or until no subsequent second knowledge fragment exists. Subsequently, the index participation range of the knowledge fragment corresponding to the mistakenly entered data is updated according to the scope of the knowledge period in the temporal effectiveness knowledge structure, ensuring that the index participation range does not cross the time boundary where the knowledge fragment has ended its participation state. When there are multiple content entries pointing to the same knowledge fragment in the index, the index participation range corresponding to multiple entries is updated synchronously for the same knowledge fragment identifier, avoiding only correcting one query term entry while retaining others.

[0075] For different inheritance relationships, the device adopts a mutually exclusive update method. For data errors corresponding to complete inheritance relationships, starting from the time the second knowledge fragment begins to take effect, the index participation range of the first knowledge fragment no longer includes the time of the second knowledge fragment's start date and subsequent times, and the index participation range of the second knowledge fragment is determined according to its effective period. For data errors corresponding to partial replacement relationships, starting from the time the second knowledge fragment begins to take effect, the index participation range of the partially replaced first knowledge fragment no longer includes the time of the second knowledge fragment's start date and subsequent times, and the index participation range of the second knowledge fragment is determined according to its effective period, without changing other first knowledge fragments with the same source file identifier but without subsequent changes. For data errors corresponding to conditional addition and conditional deletion relationships, starting from the time the second knowledge fragment begins to take effect, the index participation range of the first knowledge fragment no longer includes the time of the second knowledge fragment's start date and subsequent times, and the index participation range of the second knowledge fragment is determined according to its effective period. For erroneous data corresponding to the termination relationship, starting from the end point of the first knowledge segment's effect, the index participation range of the first knowledge segment will no longer include the end point of effect and the time thereafter, while retaining the index participation range before the end point of effect.

[0076] When a second knowledge fragment with a participation status exists, the device establishes a supplementary association between the second knowledge fragment and the query object in step S21. If the query object already has a supplementary association with the same second knowledge fragment, the same record is not created repeatedly. When the query object corresponds to multiple second knowledge fragments with a participation status at the target time point, supplementary associations can be established separately, and the subsequent actual use object is determined by the scope of application and condition items. When no second knowledge fragment with a participation status exists, no supplementary association is established; only the participation scope of the time-series effectiveness index is corrected. When the erroneous data corresponds to multiple content association entry points, both the supplementary association and the index participation scope update are performed using the knowledge fragment identifier as a unified object to avoid conflicting participation statuses at different entry points. Before establishing a supplementary association, the device reconfirms that the semantic object corresponding to the query object is consistent with the semantic object of the second knowledge fragment with a participation status, and confirms that the second knowledge fragment can be reached from the knowledge fragment corresponding to the erroneous data along the state inheritance relationship; if any condition is not met, no supplementary association is established. Therefore, the supplementary association only corrects subsequent entry points of the same policy matter that has actually occurred, and does not extend the result of one erroneous entry to other policy matters that do not have a state inheritance relationship.

[0077] After the update is complete, subsequent target queries will continue to use the updated content association index and time-series effectiveness index. When the same query object arrives again, the old knowledge fragment will not continue to be eligible for index participation within the time range where its participation status has ended, simply because it retains historical text. If a subsequent second knowledge fragment exists, supplementary associations will allow the subsequent second knowledge fragment to enter candidate generation and be verified again at the target time point. When the target time point returns to the historical effective range of the old knowledge fragment, the index participation range before the end of its effective period still allows the old knowledge fragment to participate in historical queries. Therefore, the corrective action applies to the time range corresponding to the erroneously entered data, rather than deleting the old knowledge fragment from the storage structure.

[0078] If only erroneous data is deleted from a single trusted processing path without updating the index participation scope, the same content entry may still return the same erroneous data in subsequent queries. If only the index participation scope of old knowledge fragments is closed without establishing supplementary associations, subsequent second knowledge fragments with significant text changes may still need to rely on content similarity to be recalled. Step S23 simultaneously uses erroneous data to locate old entry points and uses state inheritance relationships to locate subsequent entry points, and continues to use the updated index in the next step S21, so that the temporal effectiveness knowledge structure, effective candidate fragments, trusted processing paths, erroneous data, and index updates form a continuous feedback chain. This closed-loop correction does not adjust model weights based on a single query result, but writes the time boundary and reachable subsequent entry points corresponding to the actual erroneous knowledge fragments back to the index layer. Therefore, when the same query object reappears, the old knowledge fragments can still be used for retrospective queries within their own historical scope, but lose their index participation eligibility within the subsequent scope; at the same time, the subsequent knowledge fragments obtain stable supplementary entry points. This processing preserves both historical traceability and target time-point state accuracy, avoiding the exchange of current retrieval accuracy for physical deletion of old knowledge content.

[0079] The data conditions addressed in this invention include long-term storage of policy content, local changes affecting only certain content units, different knowledge states corresponding to the same policy matter at different target time points, and a potential time lag between index updates and knowledge structure updates. These data conditions require that the storage status of knowledge content, the participation status at the target time point, and the query entry status be expressed separately, and that their correspondence be maintained through knowledge fragment identifiers.

[0080] Step S11 establishes the independence of source location and content in the first knowledge fragment, giving local changes a state carrier that can be closed independently. Step S12 converts the time information corresponding to activation, modification, replacement, and termination into fragment-level scope intervals. Step S13 then converts the change types of the preceding and following knowledge content into traversable state inheritance relationships. Step S21 excludes old states that are not involved at the target time point based on the scope interval, while simultaneously adding the subsequent states corresponding to the target time point along the state inheritance relationship. Step S22 compares the actual usage path with the temporal effectiveness knowledge structure again. Step S23 returns the differences to the content association index and temporal effectiveness index. The output of the preceding steps continuously constitutes the judgment basis for the following steps, and the final index update enters the next step S21.

[0081] Under the combined effect of the aforementioned fragmented expression, temporal validity constraints, trusted processing path verification, and index closed-loop correction, historical knowledge content does not need to be overwritten or physically deleted, and the historical scope of action can still be hit by the target time point. Modified queries can exclude old knowledge fragments that have ended their participation state and find the second knowledge fragment along the state inheritance relationship when subsequent text expressions change. Incorrectly entered data only corrects the knowledge fragments that occurred with the error and their corresponding time ranges, preventing other unchanged first knowledge fragments from the same source file from simultaneously exiting the index. Supplementary associations convert the actual incorrectly entered paths into subsequent entry points that can be used by subsequent queries, enabling the index state to be updated according to the temporal validity knowledge structure as multi-source policy data continuously changes. Because the correction object is limited to specific knowledge fragments, specific scopes of action, and reachable subsequent knowledge fragments, closed-loop updates will not spread a single local revision into the invalidation of the entire policy document, nor will they allow the old state to re-enter the processing path at a new target time point by retaining historical content.

[0082] Ultimately, the target time point, scope of action, state inheritance relationship, and knowledge fragment identifier maintain consistent constraints across the three stages of valid candidate fragment selection, trusted processing path verification, and index participation scope update. For historical target time points, the device can return knowledge fragments within the corresponding historical scope of action; for subsequent target time points, the device can exclude knowledge fragments that have ended their participation state and add second knowledge fragments with a participation state; for cases where the index derivation structure experiences state lag, the device can locate and correct the corresponding index entry through incorrectly entered data.

[0083] In another policy application scenario, multi-source policy data can be policies related to government procurement, qualification certification, or administrative licensing. The first knowledge fragment corresponds to independent qualification conditions, material requirements, or processing conclusions. When subsequent policy data only adjusts part of the content, the fragment-level scope and state inheritance relationship are still formed according to steps S11 to S13, and the corresponding target query data is processed according to steps S21 to S23.

[0084] In the third policy application scenario, multi-source policy data can come from different issuing entities of the same policy matter. A unified knowledge representation can store the issuing entity level and applicable region as content association fields, but the applicable region does not replace the temporal validity verification of the target time point, scope of effect, and state inheritance relationship. When there is no clear sequential revision relationship between different issuing entities, no state inheritance relationship is established; each entity participates in processing according to its respective scope of effect.

[0085] Of course, the above process is just one example of credible modeling of policy knowledge. All implementation methods that can record the scope of action and state inheritance relationship according to the granularity of knowledge fragments, filter knowledge fragments according to the target time point, and use actual mis-entered data to correct the index participation state are within the scope of protection of this invention.

[0086] Based on the same application concept as the above method, embodiments of the present invention also provide a knowledge trust modeling system based on temporal validity constraints. This system can be implemented in software, hardware, or a combination of both. Taking software implementation as an example, the system can be formed by a processor in an electronic device reading computer program instructions from non-volatile memory, or it can be formed by multiple processing nodes in a distributed server.

[0087] like Figure 7 As shown, the system includes a time-series effectiveness knowledge structure construction module and a trusted index closed-loop correction module.

[0088] The temporal effectiveness knowledge structure construction module is used to acquire multi-source policy data, perform unified knowledge representation on the multi-source policy data, form multiple first knowledge fragments, and determine the scope of the first knowledge fragments; when the multi-source policy data undergoes subsequent changes, second knowledge fragments are formed, the state inheritance relationship between the first knowledge fragments and the second knowledge fragments is determined, and the temporal effectiveness knowledge structure is constructed.

[0089] The trusted index closed-loop correction module is used to establish content association index and time-series effectiveness index using the time-series effectiveness knowledge structure, generate effective candidate fragments by combining the acquired target query data and target time point, establish a trusted processing path based on the effective candidate fragments to identify erroneous data, and use the erroneous data to reverse correct the content association index and time-series effectiveness index.

[0090] This invention also provides an electronic device, which includes a processor, a memory, and a data interface. The memory stores the computer program and the multi-source policy data, temporal validity knowledge structure, content association index, and temporal validity index required to execute the method. The data interface receives the multi-source policy data and target query data, and outputs the policy knowledge processing results. The processor reads and executes the computer program in the memory, enabling the electronic device to execute the aforementioned knowledge trust modeling method based on temporal validity constraints. The electronic device can be a server, a cloud computing node, an edge computing device, or a distributed device composed of multiple computing nodes.

[0091] The various modules of the system of the present invention can be integrated into one unit or deployed separately; the above modules can be merged into one module or further divided into multiple sub-modules.

[0092] The above-disclosed embodiments are merely specific examples of the present invention, but the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A knowledge reliability modeling method based on temporal validity constraints, characterized in that, The method includes: Acquire multi-source policy data, perform unified knowledge representation on the multi-source policy data to form multiple first knowledge fragments, determine the scope of the first knowledge fragments, and record the source-related data corresponding to the first knowledge fragments; when the multi-source policy data undergoes subsequent changes, form second knowledge fragments and record the source-related data corresponding to the second knowledge fragments, determine the state inheritance relationship between the first knowledge fragments and the second knowledge fragments, and store the first knowledge fragments, second knowledge fragments, source-related data, scope of scope, and state inheritance relationship in association to construct a temporal effectiveness knowledge structure; By leveraging the knowledge structure of time-series effectiveness, we establish content association indexes and time-series effectiveness indexes. We combine the acquired target query data with the target time point to generate valid candidate fragments. Based on the valid candidate fragments, we establish a reliable processing path to identify erroneous data. We then use the erroneous data to reverse-correct the content association indexes and time-series effectiveness indexes.

2. The knowledge reliability modeling method based on temporal validity constraints according to claim 1, characterized in that, The multi-source policy data is a collection of policy content and corresponding time information obtained from different issuing entities, different source documents, or different data presentation formats. Multi-source policy data includes the policy text as well as information on its release, effectiveness, modification, replacement, cessation of implementation, and continuation of use.

3. The knowledge reliability modeling method based on temporal validity constraints according to claim 2, characterized in that, The unified knowledge representation transforms policy content into a consistent field structure. Each unified field at least carries a semantic object representing a policy matter, a condition item representing the conditions for its establishment, a processing action representing the handling or execution behavior, and a processing result representing the processing conclusion, and is associated with the source location.

4. The knowledge reliability modeling method based on temporal validity constraints according to claim 3, characterized in that, The process of forming multiple first knowledge fragments includes: merging sources and unifying knowledge representation of multi-source policy data, dividing content boundaries according to conditional semantics that can be independently judged as true or false, processing actions that can be independently executed, or processing results that can be independently established, and forming multiple first knowledge fragments at the content boundaries.

5. The knowledge reliability modeling method based on temporal validity constraints according to claim 1, characterized in that, The source association data is a combination of the publishing entity, source file identifier, superior source node, and content location identifier, used to record the source location of the first or second knowledge fragment in multi-source policy data.

6. The knowledge reliability modeling method based on temporal validity constraints according to claim 1, characterized in that, The effective range is the time range within which the first knowledge segment can participate in the data processing at the target time point, determined by the time information corresponding to the first knowledge segment; the state in which the knowledge segment can participate in data processing at the target time point is called the participation state.

7. The knowledge reliability modeling method based on temporal validity constraints according to claim 2, characterized in that, The determination of the effective range of the first knowledge fragment includes: locating the release information and the effective information based on the source association data corresponding to the first knowledge fragment, obtaining the starting point of action from the effective information; then checking the modification information, replacement information, and termination information of subsequent policy data pointing to the first knowledge fragment, obtaining the ending point of action from the action time information that first changed the participation status, and determining the effective range by the starting point of action and the ending point of action.

8. The knowledge reliability modeling method based on temporal validity constraints according to claim 6, characterized in that, The scope of action adopts a boundary convention that includes the beginning and excludes the end: when subsequent changes take effect from time T, the scope of action of the first knowledge segment ends before time T, and the scope of action of the second knowledge segment begins from time T; when the target time point is equal to time T, the second knowledge segment has a participating state, and the first knowledge segment does not have a participating state.

9. A knowledge reliability modeling method based on temporal validity constraints according to claim 3, characterized in that, The state inheritance relationship is used to record complete inheritance relationships, partial replacement relationships, condition addition relationships, condition deletion relationships, or effect termination relationships; The determination of the state inheritance relationship between the first knowledge fragment and the second knowledge fragment includes: determining the termination relationship when there is stop execution information; determining the complete inheritance relationship when the semantic objects match the same policy matter and the condition items, processing actions and processing results are consistent; determining the condition addition relationship when only condition items are added; determining the condition deletion relationship when only condition items are deleted; and determining the partial replacement relationship when the content of the condition items, processing actions or processing results changes.

10. A knowledge reliability modeling method based on temporal validity constraints according to claim 3, characterized in that, The target query data includes data containing the query object and the target time point; the query object is the policy matter to be retrieved, and the target time point is the time when the policy status needs to be adopted. The content association index is a retrieval structure built from semantic objects, condition items, processing actions, processing results, and knowledge fragment identifiers in the temporal effectiveness knowledge structure; the temporal effectiveness index is a retrieval structure built from knowledge fragment identifiers, scope of action, and state inheritance relationships in the temporal effectiveness knowledge structure.

11. The knowledge reliability modeling method based on temporal validity constraints according to claim 1, characterized in that, The generation of valid candidate fragments includes: matching the target query data with the first knowledge fragment and the second knowledge fragment based on the content association index, generating a set of candidate knowledge fragments containing candidate knowledge fragments, verifying the scope of application and state inheritance relationship of the candidate knowledge fragments using the time-series effectiveness index in conjunction with the target time point, and selecting valid candidate fragments from them.

12. The knowledge reliability modeling method based on temporal validity constraints according to claim 1, characterized in that, The identification of erroneous data includes: forming a trusted processing path based on valid candidate fragments and corresponding source association data and state inheritance relationship; using the temporal effectiveness knowledge structure to perform a second verification on the first or second knowledge fragment in the trusted processing path; and recording the first or second knowledge fragment that does not meet the target time point participation state but actually enters the trusted processing path as erroneous data.

13. A knowledge reliability modeling system based on temporal validity constraints, used to implement the knowledge reliability modeling method based on temporal validity constraints according to any one of claims 1-12, characterized in that, The system includes: The temporal effectiveness knowledge structure construction module is used to acquire multi-source policy data, perform unified knowledge representation on the multi-source policy data, form multiple first knowledge fragments, and determine the scope of the first knowledge fragments; when the multi-source policy data undergoes subsequent changes, second knowledge fragments are formed, the state inheritance relationship between the first knowledge fragments and the second knowledge fragments is determined, and the temporal effectiveness knowledge structure is constructed. The trusted index closed-loop correction module is used to establish content association index and time-series effectiveness index using the time-series effectiveness knowledge structure, generate effective candidate fragments by combining the acquired target query data and target time point, establish a trusted processing path based on the effective candidate fragments to identify erroneous data, and use the erroneous data to reverse correct the content association index and time-series effectiveness index.

Citation Information

Patent Citations

  • Person and social knowledge graph adaptive optimization method and system based on uncertain information

    CN120806085A

  • Large language model compliance reasoning method and system based on domain knowledge graph constraint

    CN121562837A