A full life cycle digital archive dynamic construction method

Abstract

The application discloses a kind of full life cycle digital archives dynamic construction method, belong to digital archives management technical field, it can solve the technical problems such as semantic fault, cognitive lag, associated inactivation and cognitive inertia in the full life cycle management of digital archives.The application realizes the traceability, quantifiable, backtracking of archive content semantics in each stage of full life cycle by semantic fingerprint and its evolution chain, fundamentally solves the problem of "semantic fault".By context perception engine, the implicit automatic update of archive semantic state is realized, so that the "cognitive state" of the archives is always synchronized with the "real context".Through dynamic association calculation and decay / activation mechanism based on semantic fingerprint, it is ensured that the association relationship in knowledge graph always remains active and reliable.Through semantic reconstruction and cognitive consistency verification during stage conversion, intelligent semantic adaptation of archives in different life cycle stages is realized, and semantic misplacement caused by "cognitive inertia" is avoided.

Description

Technical Field

[0001] This invention belongs to the field of digital archives management technology, specifically relating to a method for dynamically constructing digital archives throughout their entire lifecycle. Background Technology

[0002] Digital archives lifecycle management refers to the systematic management of digital archives throughout their entire lifecycle, from generation, circulation, archiving, preservation, utilization to destruction. In recent years, with the deepening of digital transformation, the types of digital archives have become increasingly diversified (including multimodal data such as text, images, audio and video), the number of archives has grown exponentially, and the relationships between archives have become increasingly complex. Traditional archives management methods are no longer sufficient to meet the needs of intelligent management.

[0003] Existing digital archive lifecycle management technologies mainly focus on the process management of archives, such as evidence preservation, storage, transfer, and destruction, and have the following main technical shortcomings: (i) Existing technologies only establish procedural connections between different stages of an archive's lifecycle (generation, circulation, archiving, preservation, utilization, and destruction), but fail to establish semantic continuity at the content level. When archives are modified, supplemented, formatted, or migrated by different entities at different stages, their semantic meaning often drifts or even breaks, making it impossible for subsequent users to accurately understand the true meaning and evolution of the archives. Although existing technologies include methods for version management using hash values, hash values ​​can only reflect byte-level changes in content and cannot characterize semantic changes, let alone quantify the degree of semantic change.

[0004] (ii) Existing dynamic construction methods are mostly "event-triggered" updates—the status of an archive is only updated when an explicit operation occurs. However, the semantic state of an archive depends not only on the operations performed on it, but also on changes in the external context, such as updates to laws and policies, changes in industry standards, and changes in associated archives. Existing technologies cannot perceive the implicit impact of these external changes on the semantics of archives, causing the "cognitive state" of an archive to lag behind its "real-world context."

[0005] (iii) Existing knowledge graphs or association models are mostly established once in the early stages of archival construction, and lack a dynamic maintenance mechanism thereafter. When a certain archive changes, the semantic relationships in other archives related to it are not updated synchronously, causing the association relationships to gradually "lose their effectiveness" or "become invalid," ultimately making cross-archival knowledge reasoning unreliable.

[0006] (iv) In existing technologies, when archives are transferred from one lifecycle stage to another, only process approval and storage migration are performed, without considering the fundamental impact of stage transition on the semantic weight, interpretation framework, and utilization value of the archives. The same archive should have different semantic interpretation frameworks and access cognition models in the "generation stage" and the "archiving stage," and existing technologies lack this stage-aware semantic reconstruction capability.

[0007] Therefore, in order to address the aforementioned technical issues, it is necessary to provide a method for dynamically constructing digital archives throughout their entire lifecycle. Summary of the Invention

[0008] The purpose of this invention is to provide a dynamic construction method for digital archives throughout their entire lifecycle. This method, with "cognitive consistency" as its core, solves technical problems such as semantic discontinuity, cognitive lag, association inactivation, and cognitive inertia in the management of digital archives throughout their entire lifecycle through the synergistic effect of four core mechanisms: semantic fingerprint generation, dynamic knowledge graph construction, context awareness, and adaptive evolution.

[0009] To achieve the above objectives, a specific embodiment of the present invention provides the following technical solution: A method for dynamically constructing digital archives throughout their entire lifecycle includes the following steps: S1: Obtain multimodal content data of digital archives, extract semantic feature vectors of each modality, generate a unified semantic fingerprint vector through cross-modal attention fusion mechanism, and construct an evolution chain of the semantic fingerprint vector as it evolves with each content change; S2: Based on the similarity, context co-occurrence, and temporal correlation between the semantic fingerprint vectors, dynamically calculate the correlation strength between the archives, and update the correlation strength in real time according to the evolution of the semantic fingerprint of each archive. S3: Continuously monitor changes in the external context, evaluate the impact coefficient of the changes in the external context on the semantic fingerprint of each file, generate a semantic offset proposal vector, and perform implicit semantic updates on the semantic fingerprint of the file based on the semantic offset proposal vector; S4: When a digital archive transitions from its current lifecycle stage to a target lifecycle stage, based on the semantic interpretation function corresponding to the target lifecycle stage, the current semantic fingerprint of the archive is semantically reconstructed, a stage-specific semantic description is generated, and the access cognitive model is updated. S5: Perform cognitive consistency verification at each lifecycle stage transition point and after each semantic update. If inconsistency is detected, automatically generate a conflict resolution solution and record the resolution process.

[0010] In one or more embodiments of the present invention, in step S1, the semantic fingerprint vector is a semantic-level compressed representation. Files with the same content but different formats have similar semantic fingerprint vectors. When the content undergoes substantial changes, the semantic fingerprint vector generates a quantifiable offset. The evolution chain records the displacement path of the semantic fingerprint vector in high-dimensional space, which is used to quantify the semantic distance between any two states.

[0011] In one or more embodiments of the present invention, in step S2, the update of the association strength includes: when the semantic fingerprint of any file is shifted, the association strength between the file and all associated files is automatically recalculated; associations with association strength below a preset threshold are automatically weakened or removed, and new strong associations are automatically established; the association strength is also dynamically adjusted by introducing a time decay factor and a context activation factor.

[0012] In one or more embodiments of the present invention, in step S3, the implicit semantic update is an automatic adjustment of the semantic state without explicit modification of the file content itself; the implicit semantic update is recorded in the evolution chain of the semantic fingerprint and is accompanied by a source marker indicating the reason for the context change.

[0013] In one or more embodiments of the present invention, in step S3, the external context changes include updates to laws and policies, changes in industry standards, organizational restructuring, and changes in the status of associated archives; the context-aware engine dynamically adjusts the scope and response intensity of context monitoring based on the current lifecycle stage of the archives.

[0014] In one or more embodiments of the present invention, in step S4, the semantic reconstruction reinterprets the current semantic fingerprint of the file based on the semantic interpretation function of the target lifecycle stage; the access cognition model includes retrieval weights, classification tags, and access permission semantics; different lifecycle stages correspond to different semantic interpretation functions and access cognition models.

[0015] In one or more embodiments of the present invention, in step S5, the cognitive consistency check examines whether there are conflicts or inconsistencies among the semantic fingerprint, stage-specific semantic description, association status, and contextual influence coefficient of the archive; the conflict resolution scheme includes at least one of adjusting the semantic fingerprint, correcting the association strength, and re-performing semantic reconstruction.

[0016] In one or more embodiments of the present invention, step S6 is further included: performing cluster analysis based on the semantic fingerprint vector of each file, automatically classifying the files into semantically similar categories, and automatically determining the importance level of the file in the category according to the position of the semantic fingerprint in the category space; the classification and level are dynamically adjusted as the semantic fingerprint evolves.

[0017] In one or more embodiments of the present invention, S1 to S5 achieve collaboration through a unified state vector: the semantic fingerprint generation layer outputs a semantic fingerprint vector and an evolution chain state to the dynamic knowledge graph layer and the context-aware engine; the context influence vector output by the context-aware engine is fused with the semantic fingerprint vector and input to the adaptive evolution layer; the next-stage state vector output by the adaptive evolution layer is fed back to the semantic fingerprint generation layer as the initial condition for the next round of evolution.

[0018] Compared with the prior art, the present invention has the following advantages: This invention utilizes semantic fingerprints and their evolutionary chains to achieve traceability, quantification, and retrospective capability of archival content semantics throughout its entire lifecycle, fundamentally solving the problem of "semantic discontinuity." A context-aware engine enables implicit automatic updates to the semantic state of archives, ensuring that the archive's "cognitive state" remains synchronized with its "real-world context." Dynamic association calculations and decay / activation mechanisms based on semantic fingerprints ensure the continued activity and reliability of relationships within the knowledge graph. Semantic reconstruction and cognitive consistency verification during stage transitions enable intelligent semantic adaptation of archives at different lifecycle stages, avoiding semantic misalignment caused by "cognitive inertia." Attached Figure Description

[0019] 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 recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the overall steps of a method for dynamically constructing digital archives throughout their entire lifecycle, as described in one embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the dynamic knowledge graph association strength calculation and updating in a method for dynamically constructing digital archives throughout their entire lifecycle, according to an embodiment of the present invention. Figure 3 This is a schematic diagram of semantic reconstruction during the lifecycle stage transition of a dynamic construction method for full lifecycle digital archives in one embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions in this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0022] like Figure 1 As shown, the present invention provides a method for dynamically constructing digital archives throughout their entire lifecycle, comprising the following steps: Step S1: Semantic fingerprint generation.

[0023] The system acquires multimodal content data from digital archives. These digital archives include, but are not limited to, digital content in various modalities such as electronic documents, scanned images, audio recordings, video materials, and structured data. For each digital archive, the system extracts semantic feature vectors for each modality.

[0024] Specifically, for the text modality, pre-trained language models based on the Transformer architecture, such as BERT and RoBERTa, are used to extract semantic feature vectors of the text; for the image modality, convolutional neural networks, such as ResNet and EfficientNet, are used to extract visual semantic features of the image; for the audio modality, speech recognition models and acoustic feature extraction models are used to extract semantic features of the audio; and for the video modality, visual frame features and audio features are jointly extracted and temporally aligned.

[0025] More specifically, for a digital archive containing text, images, and audio: Text modality processing: After segmenting and encoding the text content, it is input into the pre-trained BERT model. The mean pooling result of the last hidden state is taken as the text semantic feature vector with a dimension of 768.

[0026] Image modality processing: After adjusting the image to a uniform size, it is input into the pre-trained ResNet-50 model. The output of the global average pooling layer is taken as the image semantic feature vector with a dimension of 2048. It is then mapped to a 768-dimensional space through a linear projection layer.

[0027] Audio modality processing: After extracting Mel frequency cepstral coefficient (MFCC) features from the audio data, the data is input into the pre-trained VGGish model. The mean of the output layer is taken as the audio semantic feature vector with a dimension of 128. The vector is then mapped to a 768-dimensional space through a linear projection layer.

[0028] Cross-modal fusion: Using the 768-dimensional feature vector of the text modality as the query vector Q, and the 768-dimensional feature vectors of the image and audio modalities as the key K and value V, the cross-modal attention weights are calculated. Attention(Q,K,V)=softmax(QK T / √d)V Where d is the feature dimension. The fused vector is the unified semantic fingerprint vector.

[0029] When the content of the archive changes, such as when the text is modified, the image is replaced, or the audio is updated, the system re-executes the above extraction and fusion process to generate a new semantic fingerprint vector and records the displacement path between the old and new vectors to construct an evolutionary chain.

[0030] Subsequently, a cross-modal attention fusion mechanism is used to fuse the semantic feature vectors of each modality into a unified semantic fingerprint vector. The specific implementation of this cross-modal attention fusion mechanism is as follows: using the feature vector of the text modality as the query vector and the feature vectors of other modalities as key-value pairs, cross-modal attention weights are calculated, and the resulting weighted fusion yields a unified semantic fingerprint vector. This semantic fingerprint vector is a semantic-level compressed representation; files with the same content but different formats have similar semantic fingerprint vectors, while the semantic fingerprint vector exhibits a quantifiable shift when the content undergoes substantial changes.

[0031] Each time the file content changes, the system records the semantic fingerprint vectors before and after the change, constructing an evolution chain in which the semantic fingerprint vectors evolve with each change. The evolution chain records the displacement path of the semantic fingerprint vectors in high-dimensional space, including the starting vector, ending vector, displacement direction, and displacement distance for each change, thereby quantifying the semantic distance between any two states.

[0032] Step S2: Dynamic knowledge graph construction and updating.

[0033] Based on the semantic fingerprint vectors of each file, the system dynamically calculates the association strength between files. The calculation of association strength comprehensively considers the following three dimensions: First, the cosine similarity between semantic fingerprint vectors reflects the degree of semantic similarity at the level of archive content; Second, context co-occurrence reflects the frequency with which files are co-cited in the same or similar contexts; Third, temporal correlation reflects the sequential or causal relationship of archives in the time dimension.

[0034] The calculation results of the above three dimensions are summed according to preset weights to obtain the correlation strength value between the files.

[0035] When the semantic fingerprint of any file shifts, the system automatically triggers a recalculation of the association strength, recalculating the association strength between that file and all related files. Associations with a strength below a preset threshold are automatically weakened or dissolved, and new strong associations are automatically established.

[0036] Furthermore, the association strength is dynamically adjusted by introducing a time decay factor and a context activation factor. The time decay factor allows the association to decay naturally over time, preventing historical associations from excessively influencing current semantic judgments; the context activation factor allows the association to be reactivated and strengthened when triggered by specific external contexts, achieving dynamic balance of the knowledge graph.

[0037] like Figure 2 As shown, the specific implementation of dynamic calculation of association strength is as follows: For archives A and B, the association strength R(A,B) is calculated according to the following formula: R(A,B)=α·Sim(V A V B )+β·Coc(A,B)+γ·Temp(A,B) in: Sim(V A V B V is the semantic fingerprint vector. A and V B The cosine similarity is calculated using the following formula: Sim(V A V B )=(V A ·V B ) / (||V A ||·||V B ||) Coc(A,B) represents the contextual co-occurrence of files A and B, calculated using the following formula: Coc(A,B)=N AB / N total Where N AB N represents the number of times file A and file B appear together within the same context window. total This represents the total number of occurrences for all file pairs.

[0038] Temp(A,B) represents the temporal correlation between file A and file B, calculated using the following formula: Temp(A,B)=exp(-|t A -t B | / τ) Where t A and t B τ represents the timestamps of file A and file B, respectively, and τ is the time decay constant.

[0039] α, β, and γ are preset weighting coefficients that satisfy α + β + γ = 1.

[0040] The association strength R(A,B) ranges from [0,1]. When R(A,B) is lower than the preset threshold θ, the association is automatically weakened or dissolved; when R(A,B) is higher than the preset threshold, the association is recognized as a valid association.

[0041] In addition, the correlation strength is dynamically adjusted in the following ways: R'(A,B)=R(A,B)·D(t)·C(ε) Where D(t)=exp(-λt) is the time decay factor, λ is the decay rate, and t is the time since the last update; C(ε) is the context activation factor, which is C(ε)>1 when the relevant context changes occur, otherwise C(ε)=1.

[0042] Step S3: Context awareness and implicit semantic update.

[0043] The system continuously monitors changes in the external context. These changes include, but are not limited to, updates to laws and policies, changes in industry standards, organizational restructuring, and changes in the status of associated files.

[0044] When an external context change is detected, the system evaluates the impact coefficient of that change on the semantic fingerprint of each active archive. Specifically, the system pre-constructs a mapping model between context changes and semantic impact. This model is trained based on historical data and can predict the direction and degree of impact on the semantic fingerprints of different categories of archives based on the type, magnitude, and scope of the context change.

[0045] Based on the impact coefficients obtained from the evaluation, the system generates a semantic offset proposal vector. The semantic offset proposal vector indicates the direction and magnitude in which the semantic fingerprint of the archive should be offset in the semantic space.

[0046] Subsequently, the system performs implicit semantic updates on the semantic fingerprint of the archive based on the semantic offset suggestion vector. The core feature of this implicit semantic update is that the semantic fingerprint of the archive can be automatically adjusted according to changes in the external context without any explicit modification to the archive content itself. This implicit update is recorded in the evolution chain of the semantic fingerprint and is accompanied by a traceability marker indicating the "reason for context change," ensuring that each implicit update is auditable and traceable.

[0047] The context-aware engine dynamically adjusts the scope and intensity of context monitoring based on the current lifecycle stage of the file. For example, "active" files are highly sensitive to changes in external context, and the system monitors their relevant context more frequently and responds more significantly; "frozen" files are less sensitive, and the system monitors them less frequently and responds less significantly.

[0048] The specific implementation of implicit semantic updates is as follows: The system continuously monitors changes in the external context. For example, updates to regulations and policies: When the system detects that a regulation related to records management has been revised, the system first identifies the subject areas involved in the revision, such as "data privacy", "confidentiality period", and "scope of disclosure".

[0049] The system then uses a pre-built "context change-semantic impact" mapping model to assess the impact coefficient of the regulatory revision on each active file. For example, the impact coefficient is higher for files involving personal privacy data, and lower or zero for files not involving personal privacy data.

[0050] For files with an impact coefficient higher than a preset threshold, the system generates a semantic offset suggestion vector, indicating that the semantic fingerprint of the file should be shifted towards "compliance" by a certain amount.

[0051] Finally, the system performs an implicit semantic update, automatically adjusting the semantic fingerprint vector of the file without modifying its content, to reflect the compliance requirements following the regulatory revisions. This implicit update is recorded in the semantic fingerprint evolution chain, accompanied by a traceability marker for the "regulatory revision" and information on the specific clauses of the revision.

[0052] Step S4: Semantic reconstruction of lifecycle stage transition.

[0053] Digital archives go through multiple stages throughout their lifecycle, including but not limited to the generation stage, circulation stage, archiving stage, preservation stage, utilization stage, and destruction stage. When a digital archive transitions from its current lifecycle stage to a target lifecycle stage, such as from the "generation stage" to the "archiving stage," the system performs a semantic reconstruction operation.

[0054] Specifically, the system predefines corresponding semantic interpretation functions for each lifecycle stage. The semantic interpretation functions differ across lifecycle stages, meaning the same semantic fingerprint is interpreted with different semantic meanings at different stages. For example, the semantic interpretation of the same technical document in the "generation stage" focuses on the technical content itself, while in the "archiving stage," it focuses on metadata integrity and credential value.

[0055] When a phase transition occurs, the system reinterprets the current semantic fingerprint of the archive based on the semantic interpretation function corresponding to the target lifecycle phase, generating a phase-specific semantic description for that phase. Simultaneously, the system updates the archive's access cognition model, which includes, but is not limited to: retrieval weights (determining the archive's ranking position in search results), classification tags (determining the archive's affiliation within a classification system), and access permission semantics (determining the archive's accessibility in different scenarios).

[0056] like Figure 3 As shown, the specific implementation of semantic reconstruction during lifecycle phase transition is as follows: Take the transition of a technical report from the "generation stage" to the "archiving stage" as an example: During the "generation phase," the semantic interpretation function f of the technical report... generate Its semantic fingerprint is interpreted as: technical content description, author information, creation time, and other technical semantics.

[0057] When the technical report completes the approval process and enters the "archiving stage," the system automatically triggers semantic reconstruction. The semantic interpretation function f for the target stage, the "archiving stage," is... archive The semantic fingerprint is reinterpreted as: file number, filing date, retention period, security classification, metadata integrity, voucher value, and other file management semantics.

[0058] After semantic reconstruction, the system updated the access cognitive model of the archive: the retrieval weight was adjusted from "technical relevance" to "archive integrity"; the category tag was adjusted from "technical document" to "archived archive"; and the access permission semantics were adjusted from "editable" to "read-only".

[0059] Step S5: Cognitive consistency verification and conflict resolution.

[0060] At each lifecycle stage transition point and after each semantic update, the system performs a cognitive consistency check. Semantic updates include explicit updates and implicit updates.

[0061] The cognitive consistency check examines whether there are any conflicts or inconsistencies among the following four aspects: (1) The current semantic fingerprint of the archive; (2) Stage-specific semantic descriptions of the archives; (3) The status of the relationship between the archives, including the strength of the relationship with other archives; (4) The context influence coefficient of the archive reflects the degree of influence of the external context on the archive.

[0062] If an inconsistency is detected, the system automatically generates a conflict resolution plan. The conflict resolution plan includes at least one of the following operations: adjusting the semantic fingerprint (fine-tuning the semantic fingerprint vector to align it with the other three), correcting the association strength (recalculating and adjusting the association strength value), and re-performing semantic reconstruction (regenerating a stage-specific semantic description based on the adjusted semantic fingerprint). The system records the entire conflict resolution process for audit traceability.

[0063] The specific implementation of cognitive consistency verification and conflict resolution is as follows: After each explicit or implicit semantic update, the system performs a cognitive consistency check. The check process is as follows: First, the system retrieves the four status parameters of the file: (1) Current semantic fingerprint vector V current ; (2) Stage-specific semantic description D stage ; (3) Correlation strength matrix R matrix ; (4) Contextual influence coefficient C factor .

[0064] Then, the system checks the following consistency conditions one by one: Condition 1: Semantic fingerprint V current With stage-specific semantic description D stage Are there any logical conflicts between them? For example, if the semantic fingerprint indicates that the file content is "public information," while the stage-specific semantic description is marked as "top secret," then a conflict exists.

[0065] Condition 2: Semantic fingerprint V current With the correlation strength matrix R matrix Are there any inconsistencies between them? For example, if the semantic fingerprint has shifted significantly, but the association strength has not been updated accordingly, then there is an inconsistency.

[0066] Condition 3: Semantic fingerprint V current Contextual influence coefficient C factor Whether there is consistency between them. For example, if the context influence coefficient indicates that the file should have undergone a large semantic shift, but the semantic fingerprint actually shifts very little, then there is a lack of consistency.

[0067] If an inconsistency is detected, the system automatically generates a conflict resolution solution. For example, for a conflict under condition one, the system can adjust the semantic fingerprint vector or regenerate a phase-specific semantic description; for an inconsistency under condition two, the system can trigger a recalculation of the association strength; and for an incoordination under condition three, the system can re-execute the implicit semantic update.

[0068] The complete conflict resolution process includes conflict type, detection time, resolution plan, and execution results being recorded.

[0069] Step S6: Automatic classification and grading.

[0070] Cluster analysis is performed on the semantic fingerprint vectors of each file, and the system automatically categorizes files into semantically similar groups. Known clustering algorithms such as K-means, DBSCAN, or hierarchical clustering can be used.

[0071] For each category, the system automatically determines the importance level of each archive within that category based on its semantic fingerprint position in the category space. Archives located at the center of the category are identified as representative archives with higher importance, while archives located at the edge of the category are identified as marginal archives with lower importance. The classification and ranking are dynamically adjusted as the semantic fingerprint evolves.

[0072] The above steps S1 to S5 of this invention are achieved through the collaborative work of four core modules: semantic fingerprint generation layer, dynamic knowledge graph layer, context awareness engine, and adaptive evolution layer. The four modules interact with each other through a unified state vector.

[0073] Specifically: The semantic fingerprint generation layer outputs the semantic fingerprint vector and evolution chain state of the current file to the dynamic knowledge graph layer and the context-aware engine. The dynamic knowledge graph layer calculates the association strength matrix based on the received semantic fingerprint vector and feeds the association strength matrix back to the semantic fingerprint generation layer and the adaptive evolution layer.

[0074] The context-aware engine generates a context influence vector based on the external context. This context influence vector is fused with the semantic fingerprint vector to generate a modified semantic fingerprint, which is then input into the adaptive evolution layer.

[0075] The adaptive evolution layer performs semantic reconstruction based on the corrected semantic fingerprint, association strength matrix, and current lifecycle stage, and outputs the state vector for the next stage. This state vector is fed back to the semantic fingerprint generation layer as the initial condition for the next round of evolution.

[0076] Semantic fingerprints drive association computation, context corrects semantic fingerprints, stage transitions trigger semantic reconstruction, and reconstruction results are fed back to the fingerprint generation layer, achieving dynamic, adaptive, and cognitively consistent maintenance throughout the entire lifecycle.

[0077] It will be apparent to those skilled in the art that this disclosure is not limited to the details of the exemplary embodiments described above, and that this disclosure can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of this disclosure is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this disclosure. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0078] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for dynamically constructing digital archives throughout their entire lifecycle, characterized in that, Includes the following steps: S1: Obtain multimodal content data of digital archives, extract semantic feature vectors of each modality, generate a unified semantic fingerprint vector through cross-modal attention fusion mechanism, and construct an evolution chain of the semantic fingerprint vector as it evolves with each content change; S2: Based on the similarity, context co-occurrence, and temporal correlation between the semantic fingerprint vectors, dynamically calculate the correlation strength between the archives, and update the correlation strength in real time according to the evolution of the semantic fingerprint of each archive. S3: Continuously monitor changes in the external context, evaluate the impact coefficient of the changes in the external context on the semantic fingerprint of each file, generate a semantic offset proposal vector, and perform implicit semantic updates on the semantic fingerprint of the file based on the semantic offset proposal vector; S4: When a digital archive transitions from its current lifecycle stage to a target lifecycle stage, based on the semantic interpretation function corresponding to the target lifecycle stage, the current semantic fingerprint of the archive is semantically reconstructed, a stage-specific semantic description is generated, and the access cognitive model is updated. S5: Perform cognitive consistency verification at each lifecycle stage transition point and after each semantic update. If inconsistency is detected, automatically generate a conflict resolution solution and record the resolution process.

2. The method according to claim 1, characterized in that, In S1, the semantic fingerprint vector is a semantic-level compressed representation. Files with the same content but different formats have similar semantic fingerprint vectors. When the content undergoes substantial changes, the semantic fingerprint vector generates a quantifiable offset. The evolution chain records the displacement path of the semantic fingerprint vector in high-dimensional space, which is used to quantify the semantic distance between any two states.

3. The method according to claim 1, characterized in that, In step S2, the update of the association strength includes: when the semantic fingerprint of any file is shifted, the association strength between the file and all associated files is automatically recalculated; associations with association strength below a preset threshold are automatically weakened or removed, and new strong associations are automatically established; the association strength is also dynamically adjusted by introducing a time decay factor and a context activation factor.

4. The method according to claim 1, characterized in that, In S3, the implicit semantic update is an automatic adjustment of the semantic state without explicit modification of the file content itself; the implicit semantic update is recorded in the evolution chain of the semantic fingerprint and is accompanied by a source marker indicating the reason for the context change.

5. The method according to claim 1, characterized in that, In S3, the external context changes include updates to laws and policies, changes in industry standards, organizational restructuring, and changes in the status of associated archives; the context-aware engine dynamically adjusts the scope and response intensity of context monitoring based on the current lifecycle stage of the archives.

6. The method according to claim 1, characterized in that, In S4, the semantic reconstruction reinterprets the current semantic fingerprint of the file based on the semantic interpretation function of the target lifecycle stage; the access cognition model includes retrieval weights, classification labels, and access permission semantics; different lifecycle stages correspond to different semantic interpretation functions and access cognition models.

7. The method according to claim 1, characterized in that, In step S5, the cognitive consistency check examines whether there are conflicts or inconsistencies among the semantic fingerprint, stage-specific semantic description, association status, and contextual influence coefficient of the archive; the conflict resolution scheme includes at least one of adjusting the semantic fingerprint, correcting the association strength, and re-performing semantic reconstruction.

8. The method according to claim 1, characterized in that, It also includes step S6: performing cluster analysis based on the semantic fingerprint vector of each file, automatically classifying the files into semantically similar categories, and automatically determining the importance level of the file in the category according to the position of the semantic fingerprint in the category space; the classification and level are dynamically adjusted as the semantic fingerprint evolves.

9. The method according to claim 1, characterized in that, S1 to S5 achieve collaboration through a unified state vector: the semantic fingerprint generation layer outputs the semantic fingerprint vector and evolution chain state to the dynamic knowledge graph layer and the context-aware engine; The context influence vector and semantic fingerprint vector output by the context-aware engine are fused and input into the adaptive evolution layer; the next-stage state vector output by the adaptive evolution layer is fed back to the semantic fingerprint generation layer as the initial condition for the next round of evolution.

Images