Verifiable traceability evidence storage and retrieval presentation method of enterprise memory system

Abstract

The invention discloses a verifiable traceability evidence storage and retrieval presentation method for an enterprise memory system. The method comprises the following steps: stabilizing and segmenting source contents, calculating segmented evidence fingerprints of each segment, summarizing all segmented evidence fingerprints to generate source version fingerprints, and forming source object evidence storage; extracting a memory entry and binding an evidence pointer set, wherein an evidence pointer comprises a source object identifier and a segment sequence number or range information; the candidate memory items are recalled during retrieval, the current evidence content is positioned and taken out according to the evidence pointer, the current segmented evidence fingerprints are recalculated by using the same stable rule version as the stored segmented evidence fingerprints, the current segmented evidence fingerprints are compared with the segmented evidence fingerprints stored in the database, and proof nodes containing comparison results, audit fields and necessary abstracts are generated. According to the verifiable traceability evidence storage and retrieval presentation method of the enterprise memory system, the retrieval items can be verified under the condition that full-text verification is prevented from being disclosed, and the source content can be reversely verified.

Description

Technical Field

[0001] This invention relates to the technical field of enterprise management systems, and specifically to a verifiable, traceable, and retrieval-based method for enterprise memory systems. Background Technology

[0002] With the deepening of enterprise digital transformation, enterprise knowledge management systems have become a crucial infrastructure for organizational operations. Current mainstream enterprise management systems, such as memory systems, knowledge base systems, and RAG (Retrieval-Augmented Generation) systems, typically collect text data from diverse and heterogeneous systems such as emails, meeting minutes, work orders, and document repositories. After preprocessing including cleaning, segmentation, and vectorization, the processed content is written into a vector database or full-text indexing system. When a user initiates a search or question-and-answer request, the system retrieves relevant segments based on semantic similarity or keyword matching, generates a summary or answer, and includes a source link for the user's reference.

[0003] In existing technical solutions, while the source links attached to the generated summaries or answers can be clicked to view the content, it cannot prove that the generated summaries or answers actually originate from the content within those links. That is, existing systems typically only attach the source URL (Uniform Resource Locator) or document ID (Document Identifier) ​​to the generated summary or answer result, but they cannot prove which specific section of the document the generated summary or answer actually comes from, nor can they verify whether the document has been modified after being entered into the database, nor can they trace the model version used in the extraction or summarization process or whether there has been any human rewriting. This "link-based" tracing method lacks verifiability, making it difficult to guarantee the credibility of the search results.

[0004] Meanwhile, existing technical solutions cannot complete verification without disclosing the original text. Under the company's management mechanism, many original documents are restricted by permissions and cannot be displayed arbitrarily, but the business requires proof of the credibility of the generated summaries or answers. Thus, the system either displays the original text, which poses a risk of information leakage and affects business information security, or it does not display the original text, which makes it impossible to verify the authenticity of the search results, thereby affecting the reliability of the system's search. Summary of the Invention

[0005] This invention provides a verifiable, traceable, and retrieval-based method for enterprise memory systems to solve the aforementioned technical problems. Specifically, the technical solution is as follows: A verifiable, traceable, and retrieval-based method for enterprise memory systems includes the following steps: The source content is stabilized and segmented, and the segment evidence fingerprint of each segment is calculated. All segment evidence fingerprints are aggregated to generate the source version fingerprint, forming a source object evidence containing the source object identifier, the version information of the database, the segmentation rule version number, the evidence fingerprint of each segment and the source version fingerprint. Extract memory entries and bind them to a set of evidence pointers. The evidence pointers contain the source object identifier and segment number or range information. During retrieval, candidate memory entries are recalled, the current evidence content is located and retrieved based on the evidence pointer, the current segmented evidence fingerprint is recalculated using the same stabilization rule version as when it was entered into the database, and compared with the segmented evidence fingerprint stored in the database to generate a proof node containing the comparison result, audit field, segmented evidence fingerprint and source version fingerprint.

[0006] Furthermore, the stabilization process includes the following specific steps: Unified encoding processing converts the character encoding of the source content to UTF-8 encoding format; Unified line break processing converts all types of line breaks in the source content into a single line break character. Unify whitespace rules, remove trailing spaces and merge consecutive spaces into a single space; Rich text stabilization process extracts the main text from HTML or rich text format content and retains headings, list symbols, and separators as fixed tags to maintain paragraph boundary stability. The table is serialized stably, outputting the table content in a stable text format according to row and column order, using fixed row and column separators; Location information retention processing retains information such as the number of speaking rounds, paragraph numbers, and timestamps used for segmented location; the version number of the above stabilization processing rules and the source object are also saved together.

[0007] Furthermore, the calculation method for segmented evidence fingerprints is as follows: the segment number, separator, and segment text are combined in sequence and then hashed to generate the fingerprint of the segment, and the segment number is included in the fingerprint calculation. The source version fingerprint is calculated as follows: all segmented evidence fingerprints are arranged in order of segment number, connected by a delimiter, and then a hash operation is performed to generate the overall fingerprint.

[0008] Furthermore, the extracted memory entries are also bound to segmented evidence fingerprints stored in the database.

[0009] Furthermore, the extracted memory entries are also bound to entry permissions. When retrieving and recalling candidate memory entries, the corresponding memory entries are matched according to the permissions of the retrieval personnel.

[0010] Furthermore, when the user's permissions are insufficient to access the target memory entry, the system returns a permission insufficient prompt message to the user, indicating that the memory entry exists but cannot be displayed due to permission restrictions; The system identifies relevant personnel with access rights that match the permissions of the memory entry based on the searcher's affiliation and collaboration network within the enterprise's organizational structure. The system recommends a list of relevant personnel with access rights to the searcher and provides an interface for initiating collaboration requests or commissioned queries, so that the searcher can obtain access rights to the memory entry or commission query results through authorized channels.

[0011] Furthermore, the verifiable traceability, evidence storage, and retrieval presentation methods for enterprise memory systems also include: Based on the verification results of the proof nodes, the credibility of the memory entries is evaluated and sorted, and the memory content and corresponding verification status markers are output to form a minimum disclosure evidence package.

[0012] Furthermore, the minimum disclosure evidence package contains the information necessary for verification: evidence pointers, segmented evidence fingerprints stored in the database, recalculated segmented evidence fingerprints and comparison results, and source version fingerprint comparison results, but does not include the full text.

[0013] Furthermore, the credibility assessment and ranking of memory entries based on the verification results of the proof nodes is based on the integrity of the evidence, the freshness of time, and the consistency of the processing version.

[0014] Furthermore, credibility is evaluated based on the completeness of evidence, its freshness, and the consistency of the processed versions, specifically including: Verify the integrity of the evidence; Calculate the time freshness score, assuming the current retrieval verification time is T. now The source content has a timestamp of T. src Time difference Δt=T now -T src Using the exponential decay formula S t =exp(-λ×Δt) calculates the time freshness score, where λ is the decay coefficient, and the smaller Δt is, the higher the score. t The closer Δt is to 1, the larger S becomes. t The closer to 0; The verification pass score is calculated. If the recalculated segmented evidence fingerprint matches the stored segmented evidence fingerprint, the verification pass score is S. v =1, otherwise S v =0; The total credibility score is calculated using the weighted summation formula S=w t ×S t +w v ×S vCalculate the total credibility score, where w t As a weight for time freshness, w v To verify the passing weight, w t +w v =1; Memory entries are sorted and displayed in descending order of their total credibility score S, with priority given to those that have been verified and are relatively new.

[0015] The advantage of this invention lies in the verifiable traceability and retrieval presentation method of the provided enterprise memory system. It employs a dual verification mechanism combining segmented evidence fingerprints and source version fingerprints, achieving integrity verification at both the segment and whole-document levels. Any content change will cause alterations to the corresponding segmented evidence fingerprints and source version fingerprints, thus being detected during the verification process. Therefore, not only can the retrieval entries be verified to ensure the authenticity of the retrieval results, but the source content can also be verified in reverse to understand any modifications made to the source content.

[0016] The verifiable, traceable, and retrieval-based method of the enterprise memory system, without disclosing the full text, discloses evidence pointers. The fingerprint of the content pointed to by these pointers is recalculated and compared with the stored original segmented evidence fingerprints and the source version fingerprints to complete verification. This process avoids disclosing the full text during verification, thus ensuring the security of enterprise business information. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a schematic diagram of a verifiable traceability and retrieval presentation method for an enterprise memory system according to the present invention; Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0020] like Figure 1 The diagram illustrates a verifiable, traceable, and retrieval-based method for presenting and retrieving enterprise memory systems according to this application, comprising the following steps: The source content is stabilized and segmented, and the segment evidence fingerprint of each segment is calculated. All segment evidence fingerprints are aggregated to generate the source version fingerprint, forming a source object evidence containing the source object identifier, the version information of the database, the segmentation rule version number, the evidence fingerprint of each segment and the source version fingerprint. Extract memory entries and bind them to a set of evidence pointers. The evidence pointers contain the source object identifier and segment number or range information. During retrieval, candidate memory entries are recalled, the current evidence content is located and retrieved based on the evidence pointer, the current segmented evidence fingerprint is recalculated using the same stabilization rule version as when it was entered into the database, and compared with the segmented evidence fingerprint stored in the database to generate a proof node containing the comparison results and audit fields.

[0021] Based on the above steps, the operation of the enterprise memory system consists of three stages: Evidence Preservation Phase: The system first retrieves source content from meeting minutes, emails, work orders, and documents, then stabilizes the source content to eliminate the impact of format differences on subsequent verification. Next, the system divides the stabilized source content into multiple segments according to paragraph / title / speaker turn / fixed window. Then, the system calculates a segment evidence fingerprint (segment_digest) for each segment, and then concatenates and hashes all segment fingerprints in sequence to generate a source version fingerprint (root_digest). This fingerprint information, along with the source object identifier, database entry version time, segmentation rule version number, etc., is stored together to form source object evidence, which is an immutable baseline record.

[0022] Binding Phase: The system extracts memory entries (such as key conclusions, decisions, parameters, etc.) from the source content and binds a set of evidence pointers to each memory entry. The evidence pointers precisely point to the source location of the memory content, including the source object ID, segment number, and specific range, thereby achieving accurate association between the memory entry and the source segment.

[0023] Verification Phase: When a user searches, the system recalls candidate memory entries, locates the current evidence content based on the evidence pointer, and then recalculates the segmented evidence fingerprint of the located current evidence content using the same stable rule version as when it was entered into the database. The recalculated segmented evidence fingerprint is then compared with the segmented evidence fingerprint stored in the database and the source version fingerprint. The evidence pointer, the stored segmented evidence fingerprint and source version fingerprint, the recalculated segmented evidence fingerprint, the comparison result, the rule version number used, and the audit field together constitute the proof node, providing the user with a verifiable chain of evidence. The audit field includes the visitor identifier, access time, and verification result.

[0024] The aforementioned steps for recalculating and comparing the fingerprint of the current segmented evidence include: marking the evidence segment as verified when the comparison matches, and marking it as verified as failed when the comparison does not match, triggering at least one of the following actions: indicating that the evidence has been changed and needs to be reviewed, triggering re-entry into the database, or generating a manual review task. This ensures that there is follow-up after verification failure, thereby guaranteeing the stability and reliability of the system.

[0025] The aforementioned enterprise memory system's verifiable traceability and retrieval presentation method employs a dual verification mechanism combining segmented evidence fingerprints and source version fingerprints. This achieves integrity verification at both the paragraph and overall level. Any content change will alter the corresponding segmented evidence fingerprint and source version fingerprint, allowing the verification process to pinpoint the paragraph level precisely without requiring full-text recalculation. Therefore, it not only verifies search entries to ensure the authenticity of search results but also allows for reverse verification of source content to understand any modifications made to it.

[0026] Furthermore, the verifiable traceability and retrieval method of this enterprise memory system, without disclosing the full text, completes verification by disclosing evidence pointers, recalculating the fingerprint of the content pointed to by the evidence pointers, and comparing it with the stored original segmented evidence fingerprints and source version fingerprints. This process avoids disclosing the full text during verification, thus ensuring the security of the enterprise's business information.

[0027] In other words, this solution constructs a complete and verifiable traceability system, solving the problem that existing enterprise memory systems cannot verify search entries or disclose document information. Verification is completed without revealing the full text, meeting the business information protection needs of compliant scenarios. Through a dual mechanism of segmented fingerprinting and overall fingerprinting, paragraph-level and document-level integrity verification is achieved, and any tampering with the source content can be detected.

[0028] As a preferred implementation method, the stabilization process includes the following specific steps: Unified encoding processing converts the character encoding of the source content to UTF-8 encoding format; Unified line break processing converts all types of line breaks in the source content into a single line break character. Unify whitespace rules, remove trailing spaces and merge consecutive spaces into a single space; Rich text stabilization process extracts the main text from HTML or rich text format content, retaining headings, list symbols, and separators as fixed tags to maintain paragraph boundary stability; table stabilization serialization process outputs table content in a stable text format according to row and column order, using fixed row and column separators. Location information retention processing: retain information such as the number of speaking rounds, paragraph numbers, and timestamps for segmented location; This solution saves the rule version number and source object of the aforementioned stabilization process together, ensuring that the same rule version is used for calculations during subsequent verification. Through refined stabilization, this solution addresses the problem of hash algorithms being overly sensitive to minor changes, ensuring a reliable correspondence where "if the content remains unchanged, the fingerprint remains unchanged; if the content changes, the fingerprint changes." Rule versioning management guarantees the reproducibility of the verification process, avoiding misjudgments due to rule changes. The stabilization of rich text and tables expands the system's applicability, covering common document types in enterprises. The retention of location information achieves both stabilization and traceability, eliminating format noise while preserving semantic structure.

[0029] In addition, the source object evidence storage also includes hash algorithm identifiers and permission policy references. The evidence storage information of segmented evidence fingerprints includes segment ranges and segment numbers, thereby ensuring the accuracy of segment refining and avoiding deviations in the use of calculation rules.

[0030] As a preferred implementation, the segmented evidence fingerprint is calculated by combining the segment number, separator, and segment text in sequence and then performing a hash operation to generate the fingerprint for that segment. The segment number is included in the fingerprint calculation to prevent verification failure caused by paragraph reordering. In other words, if only the segment text is hashed, a tamper could alter the overall semantics of the document by changing the order of the paragraphs (e.g., swapping the 3rd and 5th paragraphs) without changing the content of each segment. However, the inclusion of the segment number ensures that any change in order will inevitably lead to a change in the fingerprint, preventing such tampering.

[0031] The source version fingerprint is calculated as follows: all segmented evidence fingerprints are arranged in order of segment number, connected by a delimiter, and then hashed to generate the overall fingerprint. This fingerprint is naturally consistent with the segmented evidence fingerprints. Any change in any segment will cause the corresponding segmented evidence fingerprint and the overall source version fingerprint to change simultaneously.

[0032] Specifically, after segmenting the stable text of the source content according to predetermined segmentation rules, the segment text _i of the i-th segment is obtained. For each segment, a segmentation evidence fingerprint is calculated: segment_digest_i = hash(segment number i + separator + segment text_i) The source version fingerprint is obtained by concatenating and hashing all segmented evidence fingerprints: Root_digest = hash(segment_digest_1 + separator + segment_digest_2 + ... + separator + segment_digest_n) In this way, any change in any segment will cause a change in the corresponding segment's evidence fingerprint, which in turn will change the source version fingerprint. Furthermore, the source version fingerprint maintains consistency with the segment's evidence fingerprint, facilitating segment-level verification and document-level auditing.

[0033] As a preferred implementation, the extracted memory entries are also bound to segmented evidence fingerprints stored in the database. When extracting memory entries, not only are evidence pointers (location information) bound to them, but the stored segmented evidence fingerprints are also directly bound to the memory entries. Thus, during the retrieval and verification phase, the system can directly retrieve the stored segmented evidence fingerprints from the memory entries, eliminating the need to query the source object's evidence storage based on the evidence pointer. This design "redundantly" stores the fingerprint information required for verification on the memory entry side, reducing the number of database queries, enabling faster matching and comparison, and significantly improving verification efficiency.

[0034] In a preferred implementation, the extracted memory entries are also bound to entry permissions. When retrieving and recalling candidate memory entries, the corresponding memory entries are matched according to the permissions of the retrieving personnel. That is, when extracting memory entries, the system binds entry permissions to the memory entries based on the permission policies of the source content (such as confidentiality level, department affiliation, project scope, etc.). Permission information may include: a list of readable roles, accessible departments, confidentiality level identifier, validity period, etc.

[0035] During retrieval, when recalling candidate memory entries, the system obtains the identity information of the current retrieval personnel (role, department, permission level, etc.) and matches it with the entry permissions of the memory entries, returning only memory entries with matching permissions. Here, permission matching can employ a whitelist mechanism (explicit authorization) or a rule engine (attribute-based access control). Unlike traditional solutions that only control access to the source document, this solution directly controls access at the memory entry level, allowing different permissions to be set for multiple memory entries extracted from the same source document. Binding permissions to memory entries ensures that permission checks are completed during the retrieval phase, eliminating the need for repeated checks during the verification phase and improving system performance. This mechanism supports dynamic permission changes; when a personnel's role changes, the permission matching results take effect in real time.

[0036] Furthermore, when a searcher's permissions are insufficient to access the target memory entry, the system returns a permission-insufficient message, indicating that the memory entry exists but cannot be displayed due to permission restrictions. In other words, during a search, when permission matching fails, the system does not simply return "no results," but explicitly prompts "the memory entry exists, but you do not have permission to access it," thus resolving the information asymmetry issue of whether the memory entry cannot be found because it does not exist or because of insufficient permissions.

[0037] Furthermore, the system analyzes the position of the search personnel within the enterprise's organizational structure: constructing a personnel relationship diagram based on organizational affiliations (such as departmental hierarchy, project team members) and collaborative relationship networks (such as joint meetings, email exchanges, and work order collaboration). Based on the search personnel's affiliations and collaborative relationship networks within the enterprise's organizational structure, the system identifies relevant personnel with access permissions matching the permissions of the aforementioned memory entries that were previously denied access. Then, the system recommends a list of relevant personnel with access permissions to the search personnel and provides interfaces for initiating collaboration requests or delegated queries, allowing the search personnel to obtain access permissions for the memory entry or delegate query results through authorized channels. The collaboration request interface allows search personnel to initiate a permission application process, obtaining temporary or permanent access rights after approval. The delegated query interface allows search personnel to delegate queries to authorized personnel, with the delegate confirming and returning anonymized or complete results.

[0038] The above solutions improve system availability and collaboration efficiency, while clear permission deficiency prompts prevent user confusion and repeated attempts. The dual-channel collaboration design, allowing for collaborative requests or delegated queries, adapts to different urgency and sensitivity levels of search scenarios, promoting the flow of enterprise knowledge while maintaining compliance with access control regulations.

[0039] As a preferred implementation, the verifiable traceability and retrieval presentation method of the enterprise memory system further includes: assessing and ranking the credibility of memory entries based on the verification results of proof nodes, outputting the memory content and corresponding verification status markers, and forming a minimal disclosure evidence package. This minimal disclosure evidence package contains the necessary information for verification: evidence pointers, segmented evidence fingerprints stored in the database, recalculated segmented evidence fingerprints and comparison results, and source version fingerprint comparison results, but does not include the full text. Verification status includes: verified (fingerprint perfectly matched); unverified (no fingerprint calculation and matching verification performed); and verification failed (fingerprint mismatch or inconsistent rule versions).

[0040] The minimum disclosure evidence package does not include the full text of the original document, constituting a processing method that does not involve verifying the content of the original document at all. This method is more secure and suitable for highly sensitive scenarios, such as confidential meeting minutes and core business data.

[0041] Based on the aforementioned credibility scores, the system reorders the recalled candidate memory entries, prioritizing the display of results with high credibility. During output, a minimal disclosure evidence package is formed according to access permissions: for high-privilege users, the complete proof node is displayed; for low-privilege users, only the verification status marker and necessary fingerprint information are displayed, without exposing sensitive fingerprint details.

[0042] Thus, the overall solution forms a complete closed loop from source evidence storage to evidence pointer binding, to fingerprint verification, to the credibility assessment of verification results, significantly improving the credibility, usability, and compliance of the enterprise memory system.

[0043] The aforementioned memory entries include memory content and a set of evidence pointers. The evidence pointer set explicitly points to the source object identifier, segment number or round ID, and range, enabling the verifier to locate the specific segment. The segmented evidence fingerprints stored in the database serve as the baseline values ​​calculated during the evidence preservation stage. Recalculating the segmented evidence fingerprints is the result of recalculating them using the same rules during the verification stage. The comparison conclusion clearly identifies "consistent" or "inconsistent."

[0044] Furthermore, the credibility assessment and ranking of memory entries based on the verification results of the proof nodes is based on the integrity of evidence, time freshness, and consistency of processing versions to reduce the risk of misuse. Here, the integrity of evidence refers to the sufficiency of the evidence supporting the memory entry, time freshness refers to the timeliness of the source content, and consistency of processing versions refers to the standardization of the verification process.

[0045] In this scheme, each memory entry corresponds to an evidence pointer and a proof node. The proof node for that memory entry already includes audit fields such as: the segmented evidence fingerprint stored in the database, the currently recalculated segmented evidence fingerprint, the comparison conclusion, the processing rule version number, the segmentation rule version number, and the hash algorithm identifier. Therefore, the credibility expression of a single memory entry can be simplified from the above evidence integrity to the integrity of the single memory entry itself, i.e., verifying the integrity of the evidence pointer and proof node of the single memory entry. After the system verifies the integrity of the evidence pointer and proof node of a single memory entry, it then calculates the time freshness and processing version consistency. Here, processing version consistency is divided into two cases: fingerprint verification consistency and fingerprint verification inconsistency. Therefore, when evaluating credibility based on evidence integrity, time freshness, and processing version consistency, the system directly checks and verifies the evidence integrity of the memory entry. After confirming the evidence integrity, it calculates the time freshness and processing version consistency, and the calculation result is the credibility. The credibility is then ranked according to the calculated credibility value of the memory entry.

[0046] Specifically, calculations are performed on time freshness and processing version consistency, including: Calculate the time freshness score, assuming the current retrieval verification time is T. now The source content has a timestamp of T. src Time difference Δt=T now -T src Using the exponential decay formula S t =exp(-λ×Δt) calculates the time freshness score, where λ is the decay coefficient, and the smaller Δt is, the higher the score.t The closer St is to 1, the larger Δt is, and the closer St is to 0. The verification pass score is calculated. If the recalculated segmented evidence fingerprint matches the stored segmented evidence fingerprint, the verification pass score is S. v =1, otherwise S v =0; The total credibility score is calculated using the weighted summation formula S=w t ×S t +w v ×S v Calculate the total credibility score, where w t As a weight for time freshness, w v To verify the passing weight, w t +w v =1; Memory entries are sorted and displayed in descending order of their total credibility score S, with priority given to those that have been verified and are relatively new.

[0047] In summary, this invention significantly enhances the credibility and security of enterprise knowledge management by constructing a verifiable and tamper-proof enterprise memory system. It employs a dual verification mechanism of segmented evidence fingerprints and source version fingerprints, achieving integrity verification at both the paragraph and document levels. Any content tampering can be accurately detected, solving the problem in existing systems where the source can be opened but the authenticity of the source content cannot be proven. By providing a minimum disclosure evidence package, verification can be completed through evidence pointer and fingerprint comparison without revealing the full text, meeting the requirements of transparent verification information and protection of commercial information, thus ensuring the security of enterprise commercial information.

[0048] This invention also employs a scientific credibility assessment system, comprehensively considering multiple dimensions such as time freshness and verification consistency, to provide objective ranking criteria for search results, thereby reducing the risk of data acceptance by enterprises. Furthermore, through access control and intelligent collaboration, this invention achieves more precise access management and efficient permission transfer, promoting compliant knowledge sharing within the enterprise. Stabilization processing and rule versioning management ensure the long-term reproducibility of the verification process, providing a foundation for the sustainable and reliable operation of the system.

[0049] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for verifiable traceability, evidence storage, and retrieval presentation of an enterprise memory system, characterized in that, Includes the following steps: The source content is stabilized and segmented, and the segment evidence fingerprint of each segment is calculated. All segment evidence fingerprints are aggregated to generate the source version fingerprint, forming a source object evidence containing the source object identifier, the version information of the database, the segmentation rule version number, the evidence fingerprint of each segment and the source version fingerprint. Extract memory entries and bind them to a set of evidence pointers, wherein the evidence pointers contain source object identifiers and segment numbers or range information; During retrieval, candidate memory entries are recalled, the current evidence content is located and retrieved based on the evidence pointer, the current segmented evidence fingerprint is recalculated using the same stabilization rule version as when it was entered into the database, and compared with the segmented evidence fingerprint stored in the database to generate a proof node containing the comparison result, audit field, segmented evidence fingerprint and source version fingerprint.

2. The verifiable traceability and retrieval method for enterprise memory systems according to claim 1, characterized in that, The stabilization process includes the following specific steps: Unified encoding processing converts the character encoding of the source content to UTF-8 encoding format; Unified line break processing converts all types of line breaks in the source content into a single line break character. Unify whitespace rules, remove trailing spaces and merge consecutive spaces into a single space; Rich text stabilization process extracts the main text from HTML or rich text format content and retains headings, list symbols, and separators as fixed tags to maintain paragraph boundary stability. The table is serialized stably, outputting the table content in a stable text format according to row and column order, using fixed row and column separators; Location information retention processing: retain information such as the number of speaking rounds, paragraph numbers, and timestamps for segmented location; Save the version number of the above stabilization rules along with the source object.

3. The verifiable traceability and retrieval method for enterprise memory systems according to claim 1, characterized in that, The calculation method for the segmented evidence fingerprint is as follows: the segment number, separator and segment text are combined in sequence and then hashed to generate the fingerprint of the segment, and the segment number is included in the fingerprint calculation. The calculation method for the source version fingerprint is as follows: all segmented evidence fingerprints are arranged in order of segment number, connected by a separator, and then a hash operation is performed to generate the overall fingerprint.

4. The verifiable traceability and retrieval method for enterprise memory systems according to claim 1, characterized in that, The extracted memory entries are also bound to the segmented evidence fingerprints stored in the database.

5. The verifiable traceability and retrieval method for enterprise memory systems according to claim 1, characterized in that, The extracted memory entries are also bound to entry permissions. When retrieving and recalling candidate memory entries, the corresponding memory entries are matched according to the permissions of the person retrieving the memory.

6. The verifiable traceability and retrieval method for enterprise memory systems according to claim 5, characterized in that, When the user's permissions are insufficient to access the target memory entry, the system returns a permission insufficient prompt message to the user, indicating that the memory entry exists but cannot be displayed due to permission restrictions; The system identifies relevant personnel with access rights that match the permissions of the memory entry based on the searcher's affiliation and collaboration network within the enterprise's organizational structure. The system recommends the list of relevant personnel with access rights to the searcher and provides an interface for initiating collaboration requests or commissioned queries, so that the searcher can obtain access rights to the memory entry or commission query results through authorized channels.

7. The verifiable traceability and retrieval method for enterprise memory systems according to claim 1, characterized in that, The verifiable traceability, evidence storage, and retrieval presentation method of the enterprise memory system also includes: Based on the verification results of the proof nodes, the credibility of the memory entries is evaluated and sorted, and the memory content and corresponding verification status markers are output to form a minimum disclosure evidence package.

8. The verifiable traceability and retrieval method for enterprise memory systems according to claim 7, characterized in that, The minimum disclosure evidence package contains the information necessary for verification: evidence pointers, segmented evidence fingerprints stored in the database, recalculated segmented evidence fingerprints and comparison results, and source version fingerprint comparison results, but does not include the full text.

9. The verifiable traceability and retrieval method for an enterprise memory system according to claim 7, characterized in that, The credibility assessment and ranking of memory entries based on the verification results of the proof nodes is based on the integrity of the evidence, the freshness of time, and the consistency of the processing version.

10. The verifiable traceability and retrieval method for an enterprise memory system according to claim 9, characterized in that, The credibility evaluation based on the completeness of evidence, freshness of time, and consistency of processing versions specifically includes: Verify the integrity of the evidence; Calculate the time freshness score, assuming the current retrieval verification time is T. now The source content has a timestamp of T. src Time difference Δt=T now -T src Using the exponential decay formula S t =exp(-λ×Δt) calculates the time freshness score, where λ is the decay coefficient, and the smaller Δt is, the higher the score. t The closer Δt is to 1, the larger S becomes. t The closer to 0; The verification pass score is calculated. If the recalculated segmented evidence fingerprint matches the stored segmented evidence fingerprint, the verification pass score is S. v =1, otherwise S v =0; The total credibility score is calculated using the weighted summation formula S=w t ×S t +w v ×S v Calculate the total credibility score, where w t As a weight for time freshness, w v To verify the passing weight, w t +w v =1; Memory entries are sorted and displayed in descending order of their total credibility score S, with priority given to those that have been verified and are from more recent sources.

Images