Lifecycle-based traceability hierarchical storage method, device, equipment, and media
By adopting a lifecycle-based traceability hierarchical storage method, the problem of inaccurate knowledge storage hierarchy division in existing technologies is solved, enabling dynamic adjustment and traceability of knowledge storage, and improving the efficiency of knowledge management in fintech and healthcare business scenarios.
Patent Information
- Application Number
- CN202611324756.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-17
AI Technical Summary
Existing knowledge management technologies struggle to comprehensively reflect compliance constraints on knowledge content, the effective period corresponding to each category, and changes in importance caused by business events. This leads to inaccurate knowledge storage hierarchy divisions, resulting in the improper downgrading of important knowledge or the long-term occupation of real-time storage resources by short-term knowledge.
The lifecycle-based traceability and hierarchical storage method generates knowledge entries by acquiring knowledge data, establishes knowledge traceability records, determines compliance constraint levels and forgetting half-life parameters, generates dynamic weights by combining business events, performs hierarchical storage, and generates change version nodes.
This improves the accuracy of hierarchical knowledge storage and the traceability of changes, ensuring the compliance of important knowledge and that the storage level matches its actual importance, thus avoiding resource waste.
Smart Images

Figure CN123115240A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of distributed storage technology, and in particular to a lifecycle-based traceable hierarchical storage method, apparatus, device, and medium. Background Technology
[0002] Existing knowledge management technologies typically determine the importance of knowledge based on access frequency, update time, or fixed-time decay methods. This makes it difficult to comprehensively reflect compliance constraints on knowledge content, the effective period corresponding to knowledge categories, and changes in importance caused by business events. As a result, the judgment of the importance of knowledge is inconsistent with the actual retention requirements, and further leads to inaccurate knowledge storage hierarchy division.
[0003] In the fintech sector, financial laws, regulations, business rules, and temporary business notices have varying degrees of compliance constraints and effective periods. Events such as regulatory policy adjustments, product launches, and business disputes can also alter the perceived importance of corresponding knowledge. Existing financial knowledge management technologies primarily archive knowledge based on access frequency or update time, which can prematurely relocate infrequently used but still strictly compliant financial knowledge to infrequent storage, while simultaneously causing short-term, high-frequency information to occupy real-time storage resources for extended periods.
[0004] In the healthcare sector, medical laws and regulations, treatment guidelines, drug use requirements, and temporary public health notices all have varying degrees of compliance constraints and applicable periods. Adjustments to treatment guidelines, drug risk notifications, and public health emergencies can also alter the perceived importance of corresponding knowledge. Existing medical knowledge management technologies struggle to accurately assess the ongoing retention needs of different types of knowledge, leading to issues such as the inappropriate downgrading of important medical knowledge and the prolonged occupation of real-time storage resources by short-term knowledge. Summary of the Invention
[0005] The main objective of this invention is to provide a lifecycle-based traceability hierarchical storage method, apparatus, device, and storage medium, aiming to solve the technical problem that existing knowledge management technologies mainly use a single indicator to evaluate the importance of knowledge, which is difficult to reflect the dynamic changes in the actual state of knowledge, resulting in inaccurate knowledge hierarchical storage.
[0006] To achieve the above objectives, the present invention provides a lifecycle-based traceability hierarchical storage method, comprising: Obtain raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate a knowledge identifier for each knowledge entry in the set of knowledge entries and configure knowledge metadata including knowledge category; A knowledge traceability record is established based on the knowledge identifier of each knowledge item. An initial version node is generated based on the knowledge content and knowledge metadata of each knowledge item at the time of its generation, and the initial version node is written into the knowledge traceability record. The knowledge content of each knowledge item is input into a text classification model for compliance constraint level classification processing to obtain the compliance constraint level of each knowledge item, and the mandatory retention coefficient is determined based on the compliance constraint level. Based on the knowledge category of each knowledge item, the forgetting half-life parameter is determined, the review status data of each knowledge item is obtained, and the knowledge life cycle parameter data is generated. Based on the knowledge identifier of each knowledge item, obtain the business events associated with each knowledge item, and generate the dynamic weight of each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient. Based on the dynamic weights of each knowledge item, each knowledge item is stored in a hierarchical manner to generate a hierarchical storage result. Based on the dynamic weights and the hierarchical storage result, a change version node is generated. Taking the initial version node as the version starting point, the change version node is appended to the corresponding knowledge traceability record according to the version generation order.
[0007] Furthermore, to achieve the above objectives, the present invention provides a lifecycle-based traceable hierarchical storage device, comprising: The knowledge entry construction module is used to acquire raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate knowledge identifiers for each knowledge entry in the knowledge entry set, and configure knowledge metadata including knowledge categories. The knowledge traceability establishment module is used to establish a knowledge traceability record based on the knowledge identifier of each knowledge item, generate an initial version node based on the knowledge content and knowledge metadata when each knowledge item is generated, and write the initial version node into the knowledge traceability record; The compliance constraint assessment module is used to input the knowledge content of each knowledge item into the text classification model for compliance constraint level classification processing, obtain the compliance constraint level of each knowledge item, and determine the mandatory retention coefficient based on the compliance constraint level; The lifecycle parameter generation module is used to determine the forgetting half-life parameter based on the knowledge category of each knowledge item, obtain the review status data of each knowledge item, and generate knowledge lifecycle parameter data. The dynamic weight generation module is used to obtain business events associated with each knowledge item based on the knowledge identifier of each knowledge item, and generate dynamic weights for each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient. The hierarchical storage update module is used to perform hierarchical storage of each knowledge item based on the dynamic weight of each knowledge item, generate hierarchical storage results, generate change version nodes based on the dynamic weight and the hierarchical storage results, and append the change version nodes to the corresponding knowledge traceability records according to the version generation order, with the initial version node as the version starting point.
[0008] Furthermore, to achieve the above objectives, the present invention also provides a computer device, the computer device including a memory, a processor, and a lifecycle-based traceability hierarchical storage program stored in the memory and executable on the processor, wherein when the lifecycle-based traceability hierarchical storage program is executed by the processor, it implements the steps of the lifecycle-based traceability hierarchical storage method as described above.
[0009] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a lifecycle-based traceability hierarchical storage program, wherein when the lifecycle-based traceability hierarchical storage program is executed by a processor, it implements the steps of the lifecycle-based traceability hierarchical storage method as described above.
[0010] Beneficial Effects: This invention relates to the field of distributed storage technology and discloses a lifecycle-based traceable hierarchical storage method, apparatus, device, and medium, comprising: acquiring original knowledge data and generating knowledge entries with knowledge identifiers, knowledge categories, and knowledge metadata; establishing knowledge traceability records and initial version nodes; determining compliance constraint levels and mandatory retention coefficients based on knowledge content, and generating knowledge lifecycle parameter data based on knowledge categories and review status; generating dynamic weights by combining current time, business events, knowledge lifecycle parameter data, and mandatory retention coefficients; performing hierarchical storage based on dynamic weights, and appending the dynamic weights and hierarchical storage results to the knowledge traceability records after forming a change version node. This invention can be applied to business scenarios such as fintech and healthcare, generating dynamic weights by comprehensively considering knowledge status, business events, and compliance constraints, and adjusting storage levels and record version changes accordingly, thereby improving the accuracy of hierarchical knowledge storage and the traceability of the change process. Attached Figure Description
[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a schematic diagram of an application environment for a lifecycle-based traceability hierarchical storage method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an embodiment of the lifecycle-based traceability hierarchical storage method of the present invention; Figure 3 for Figure 2 A detailed flowchart of step S10; Figure 4 for Figure 2 A detailed flowchart of step S20; Figure 5 for Figure 2 A detailed flowchart of step S30; Figure 6 for Figure 2 A detailed flowchart of step S40; Figure 7 for Figure 2 A detailed flowchart of step S50; Figure 8 for Figure 2 A detailed flowchart of step S60; Figure 9 This is a schematic diagram of the functional modules of a preferred embodiment of the lifecycle-based traceable hierarchical storage device of the present invention. Figure 10 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 11 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation
[0012] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.
[0013] The lifecycle-based traceability hierarchical storage method provided in this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server can obtain raw knowledge data from the client and generate knowledge entries with knowledge identifiers, knowledge categories, and knowledge metadata; establish knowledge traceability records and initial version nodes; determine compliance constraint levels and mandatory retention coefficients based on knowledge content; generate knowledge lifecycle parameter data based on knowledge categories and review status; generate dynamic weights by combining current time, business events, knowledge lifecycle parameter data, and mandatory retention coefficients; perform hierarchical storage based on dynamic weights; and append the dynamic weights and hierarchical storage results to the knowledge traceability records after forming a change version node. This invention can be applied to business scenarios such as fintech and healthcare. By comprehensively considering knowledge status, business events, and compliance constraints to generate dynamic weights, and adjusting storage levels and record version changes accordingly, it improves the accuracy of hierarchical knowledge storage and the traceability of the change process. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.
[0014] Please see Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the lifecycle-based traceability hierarchical storage method provided by the present invention. It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.
[0015] like Figure 2 As shown, the lifecycle-based traceability hierarchical storage method proposed in this invention includes the following steps: S10, acquire raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate a knowledge identifier for each knowledge entry in the set of knowledge entries and configure knowledge metadata including knowledge category; In this embodiment, data from different sources is obtained through data interfaces, file reading, message receiving, or database queries, and the data source identifier, data format, and collection time are recorded. Field-based records read the title, body, category, and time fields according to field correspondence; text files are divided into content units based on heading level, paragraph boundaries, or independent semantic scope; table content combines table headers with corresponding cells to form content units that can independently express meaning.
[0016] The character encoding, time format, field names, and whitespace characters of content units are standardized, and content with complete semantics and independent management capabilities is grouped into knowledge content. Knowledge content can correspond to a complete document, or to clauses, items, requirements, or prompts within a document. Each piece of knowledge content is encapsulated as a knowledge entry and organized into a collection of knowledge entries according to a unified field structure.
[0017] Generate a unique and stable knowledge identifier for each knowledge entry. Knowledge identifiers can be generated based on data source identifiers and source record identifiers, or they can be generated using a combination of namespaces and sequential encoding. When the knowledge content changes, the original knowledge identifier continues to be used, and duplicate identifiers are eliminated through conflict detection.
[0018] Knowledge categories are determined from data source fields, directory information, title text, or knowledge content. Knowledge categories can be determined based on field correspondence, keyword matching, or semantic similarity. The knowledge identifier, knowledge category, data source identifier, collection time, and content summary are combined into knowledge metadata and configured for the corresponding knowledge entry.
[0019] When the data source field structure is stable, a field mapping table is used to record the correspondence between source fields and unified fields. The title field, content field and time field in the same source record are combined into knowledge content. A knowledge identifier is generated based on the data source identifier and the source record identifier. After the category field is converted into a knowledge category, it is written into the knowledge metadata.
[0020] When text files constitute a large proportion of the data, layout parsing and semantic segmentation are used to read titles, body text, tables, and time information. Knowledge content is then categorized based on title level, paragraph numbering, and content theme. File path, file name, and location within the file form source record identifiers, which are then combined with data source identifiers to generate knowledge identifiers. Knowledge categories are determined based on directory information, title keywords, and body text semantics.
[0021] In the fintech business, regulatory documents, business systems and product information are obtained from regulatory information platforms, internal policy databases and product information databases. The knowledge content is divided into independent clauses, processing matters and product scope, knowledge entries and knowledge labels are generated, and knowledge categories such as regulatory norms, internal policies or product information are configured.
[0022] In the field of healthcare, treatment guidelines, drug usage requirements, and health management data are obtained from medical guideline databases, drug information platforms, and health management platforms. The knowledge content is then categorized according to treatment matters, applicable subjects, and risk warnings, generating knowledge entries and knowledge identifiers, and configuring corresponding knowledge categories, source identifiers, and time information.
[0023] This embodiment unifies the content structure of different sources and data formats, encapsulates independent semantic content into knowledge entries, and configures stable knowledge identifiers and knowledge metadata containing knowledge categories. This can reduce data confusion caused by differences in fields and content granularity, and improve the identifiability, classification accuracy, and data structure consistency of knowledge entries.
[0024] S20, establish a knowledge traceability record based on the knowledge identifier of each knowledge item, generate an initial version node based on the knowledge content and knowledge metadata when each knowledge item is generated, and write the initial version node into the knowledge traceability record; In this embodiment, knowledge tracing records are created according to the knowledge identifiers of each knowledge item, and a correspondence is established between the knowledge identifiers and the knowledge tracing records. Each knowledge tracing record is configured with a record identifier, a knowledge identifier, a set of version nodes, and a record creation time to distinguish version data for different knowledge items.
[0025] Read the knowledge content and metadata at the time of knowledge entry generation, and generate an initial content snapshot according to a preset field structure. The initial content snapshot retains the knowledge content, knowledge category, data source, time attribute, and content summary. Perform summary processing on the initial content snapshot to generate version summary data and record the node generation time.
[0026] The initial content snapshot, version summary data, node generation time, and knowledge identifier are combined into an initial version node, and the node identifier and version sequence are configured. Based on the knowledge identifier, the corresponding knowledge tracing record is located, and the initial version node is written into the version node set. The node identifier and version sequence are verified to avoid duplicate writing.
[0027] When using a relational database for storage, a knowledge tracing record table and a version node table are set up. The knowledge tracing record table stores record identifiers and knowledge identifiers, while the version node table stores initial content snapshots, version summary data, node generation time, and version sequence. The two tables are linked by the record identifier. The knowledge tracing record table is queried using the knowledge identifier; if no corresponding record is found, a new record is created, and the initial version node is then written to the version node table.
[0028] In document storage architectures, knowledge identifiers are used as document indexes, and knowledge tracing records are saved as data objects containing record attributes and an array of version nodes. Initial content snapshots are converted into structured documents, and version summary data and node identifiers are generated and written to the version node array. Atomic writes ensure that only one initial version node is generated for the same knowledge identifier.
[0029] In the fintech business, after financial regulatory documents are divided into multiple knowledge items, a knowledge traceability record is established based on each knowledge identifier. The document clause content, regulatory category, issuing entity, effective time and data source are combined into an initial content snapshot and written into the corresponding initial version node, so that the generation status of each financial regulatory knowledge can be saved independently.
[0030] In the field of healthcare, the health management platform forms treatment guidelines and drug usage requirements into knowledge entries, establishes knowledge traceability records based on knowledge identifiers, and saves the content of the guidelines, knowledge categories, publication dates, scope of application, and source platforms as initial version nodes, so that the initial content and attributes of different healthcare knowledge can be located separately.
[0031] This embodiment establishes knowledge traceability records by using knowledge identifiers and saves the knowledge content and knowledge metadata of the knowledge entry generation stage as initial version nodes, so that the knowledge entry and the initial data state form a stable correspondence. By identifying the initial version node through version summary data, node generation time and version sequence, duplicate writing and data confusion can be reduced, and the integrity and traceability of the initial state record of knowledge can be improved.
[0032] S30, input the knowledge content of each knowledge item into the text classification model for compliance constraint level classification processing, obtain the compliance constraint level of each knowledge item, and determine the mandatory retention coefficient based on the compliance constraint level; In this embodiment, textual information affecting compliance judgment is identified from knowledge content, including the publishing entity, effectiveness description, scope of application, effective period, and semantic meaning of the text, and the character format, date format, and synonym expressions are standardized. The publishing entity is used to reflect the constraint level of the content source, the effectiveness description is used to identify semantics such as mandatory execution, restricted use, or reference suggestions, and the scope of application and effective period are used to determine the constraint objects and their duration.
[0033] The identified text information is converted into categorized data, which is then input into a text classification model for compliance constraint level classification. This generates probability data for multiple candidate levels, and the candidate level with the highest probability value is determined as the compliance constraint level. The compliance constraint level represents the strength of the constraint on the knowledge content, and different levels are ordered according to their constraint strength. When the probability values of multiple candidate levels are close, the priority of the publishing entity, the semantic strength of the validity, and the scope of application can be used as auxiliary classification features input into the text classification model to improve the accuracy of compliance constraint level classification.
[0034] The system queries the data corresponding to the compliance constraint level and determines the matching value as the mandatory retention coefficient. The data corresponding to the level coefficient is set in ascending order of compliance constraint level, so that knowledge content with higher constraint strength receives a larger mandatory retention coefficient, and the compliance constraint level and mandatory retention coefficient are associated with the corresponding knowledge item.
[0035] When the knowledge content field structure is stable, model auxiliary features are generated by field matching and weight aggregation, and the level is determined by a text classification model. Matching words and judgment values are configured for the publishing entity, validity description, and scope of application, respectively. The judgment values of each field are aggregated to form model auxiliary features. The model auxiliary features and knowledge content are input into the text classification model to obtain the compliance constraint level, and then the mandatory retention coefficient is read from the data corresponding to the level coefficient.
[0036] In cases where the knowledge content structure varies significantly, a text classification model is used to determine the compliance level. The model comprises a text encoding layer, a semantic representation layer, and a classification output layer, converting the segmented and encoded knowledge content into probability data for multiple candidate levels. The model parameters are adjusted using training text labeled with compliance constraint levels, and the candidate level with the highest probability value is determined as the compliance constraint level. Then, the corresponding mandatory retention coefficient is determined.
[0037] In the fintech business sector, we analyzed financial regulatory provisions, internal control systems, product materials, and temporary business notices. Regulatory provisions with legally binding semantics and broad applicability were assigned a higher compliance constraint level and a larger mandatory retention coefficient; notices with shorter applicable periods and used only for business notifications were assigned a relatively smaller mandatory retention coefficient.
[0038] In the healthcare sector, analyses were conducted on drug safety requirements, treatment guidelines, health management standards, and short-term risk notices. Content containing mandatory usage restrictions or safety warnings was identified as having a higher compliance constraint level, while health tips were assigned a corresponding level and mandatory retention coefficient based on the issuing entity, scope of application, and validity period.
[0039] This embodiment identifies the source, validity semantics, and scope of application from knowledge content, inputs the identified information into a text classification model for compliance constraint level classification, and then converts the level into a mandatory retention coefficient. This can reduce the bias caused by judging the importance of knowledge solely based on access frequency or update time, and improve the accuracy of compliance attribute quantification results.
[0040] S40, determine the forgetting half-life parameter based on the knowledge category of each knowledge item, obtain the review status data of each knowledge item, and generate knowledge life cycle parameter data; In this embodiment, the corresponding forgetting half-life parameter is determined by querying the category period data based on the knowledge category of each knowledge item. The forgetting half-life parameter is used to represent the effective retention period of different categories of knowledge over time. The category period data can be configured according to the stability of the content, the update frequency, and the applicable period, so that different knowledge categories adopt different time parameters.
[0041] Based on the knowledge identifier, the corresponding review record is retrieved. The most recent review time and the cumulative number of reviews are determined from the review record, and combined to form the review status data. For knowledge entries without review records, the entry generation time can be used as the time base, and an initial number of reviews can be configured to avoid missing necessary fields in the knowledge lifecycle parameter data.
[0042] The forgetting half-life parameter, time base, and cumulative review count are combined into knowledge lifecycle parameter data according to a unified field structure, and a correspondence is established with the knowledge identifier. The knowledge lifecycle parameter data consists of category periodic attributes and actual review status, which can distinguish knowledge items with different review statuses within the same category.
[0043] When the number of knowledge categories is fixed and the update cycle is clear, a category-cycle mapping table is used to store the correspondence between each knowledge category and the forgetting half-life parameter. The forgetting half-life parameter is directly queried based on the knowledge category, and then the review records are read according to the knowledge identifier. The most recent review time and the number of records are written into the knowledge lifecycle parameter data.
[0044] When knowledge categories have significantly different update frequencies, a tiered periodic interval configuration for the forgetting half-life parameter is adopted. The periodic interval is selected based on the content stability level and applicable period corresponding to the knowledge category, and review status data is generated based on the most recent review time and cumulative review count in the review records. When review records are missing, the corresponding fields are supplemented using the entry generation time and the preset initial review count.
[0045] In the fintech business sector, different forgetting half-life parameters are configured for regulatory guidelines, internal systems, product materials, and temporary business notices. The most recent review time and cumulative number of reviews are determined based on the review records of each knowledge item. Category cycle information and review status are combined to form knowledge lifecycle parameter data, ensuring that long-term effective regulatory knowledge and short-term applicable business notices have different lifecycle parameters.
[0046] In the healthcare sector, treatment guidelines, drug usage requirements, medical research data, and short-term health notices are configured with different forgetting half-life parameters based on the stability of their content. Corresponding review records are retrieved based on knowledge identifiers, and the review time and number of reviews are written into the knowledge lifecycle parameter data, allowing long-term applicable medical knowledge and frequently updated health information to use different lifecycle parameters.
[0047] This embodiment reduces the bias in knowledge lifecycle judgment caused by uniform time parameters by configuring differentiated forgetting half-life parameters according to knowledge categories and incorporating actual review time and number of reviews into the knowledge lifecycle parameter data. This ensures that knowledge category attributes and review status form a consistent data expression, thereby improving the accuracy and completeness of knowledge lifecycle parameters.
[0048] S50: Obtain business events associated with each knowledge item based on the knowledge identifier of each knowledge item; generate dynamic weights for each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient. In this embodiment, corresponding events are matched in the business event record based on knowledge identifiers, thus establishing an association between each business event and a specific knowledge entry. A business event includes at least the event occurrence time and the event type. The event occurrence time is used to determine the timeliness of the event, and the event type is used to identify whether the event enhances, weakens, or maintains the importance of the knowledge.
[0049] The forgetting half-life parameter and review status data are read from the knowledge lifecycle parameter data. The time interval is determined based on the current time and the time base in the review status data, and the time change is generated by combining the forgetting half-life parameter. As the time interval increases, the time change gradually decreases; when the review status is updated or a related business event occurs, the validity of the corresponding knowledge item is adjusted.
[0050] Dynamic weights are generated by combining time-varying factors, event contribution from business events, and mandatory retention coefficients. The mandatory retention coefficient is used to suppress excessive weighting of highly constrained knowledge due to the passage of time or infrequent use, while business events reflect the real-time changes in the importance of knowledge, so that dynamic weights simultaneously reflect time status, event status, and retention requirements.
[0051] In scenarios with high business event update frequency, an event stream matching method is used to read new events carrying knowledge identifiers and update the event contribution based on the event occurrence time. For each valid event received, the corresponding knowledge lifecycle parameter data and mandatory retention coefficient are reread, and the dynamic weight is updated in combination with the current time. This method is suitable for data environments where regulatory information, risk warnings, or business status are constantly changing.
[0052] In scenarios where business events are updated infrequently, a periodic scanning approach is used to aggregate business events within a specified time range. Event contribution is generated for each knowledge item based on the number of events, event type, and the time of the most recent event, and then dynamic weights are generated in batches by combining time changes and mandatory retention coefficients. This approach is suitable for data environments with a large number of knowledge items and concentrated event updates.
[0053] In the fintech business sector, regulatory requirements adjustments, product launches, or dispute resolution events carry corresponding knowledge identifiers. Event contribution is calculated based on the event's occurrence time and type, and dynamically weighted by combining lifecycle parameters of regulatory knowledge with mandatory retention coefficients, ensuring that infrequently used but highly constrained financial knowledge maintains an appropriate weight.
[0054] In the healthcare sector, updates to treatment guidelines, drug risk notifications, or public health events are linked to corresponding knowledge entries. The degree of temporal change in the knowledge content is assessed based on the current time, and dynamic weights are generated by combining the event's impact and mandatory retention requirements. This assigns appropriate weights to healthcare knowledge with higher safety requirements or those recently affected by events.
[0055] This embodiment accurately associates business events with knowledge identifiers and generates dynamic weights by combining current time, knowledge lifecycle parameter data, and mandatory retention coefficients. This reduces weight bias caused by single-time decay, allowing the importance of knowledge to be adjusted according to time status and business changes. At the same time, it avoids excessive deweighting of highly constrained knowledge and improves the accuracy of dynamic weights.
[0056] S60, based on the dynamic weight of each knowledge item, perform hierarchical storage of each knowledge item, generate hierarchical storage results, generate change version nodes based on the dynamic weight and the hierarchical storage results, and append the change version nodes to the corresponding knowledge traceability records according to the version generation order, with the initial version node as the version starting point.
[0057] In this embodiment, the dynamic weight of each knowledge item is compared with the corresponding storage level conditions to determine the corresponding target storage area. Knowledge items with higher dynamic weights are allocated to storage areas with faster access response speeds, while knowledge items with lower dynamic weights are allocated to low-frequency access storage areas. The current storage area of a knowledge item is read; if the current storage area is inconsistent with the target storage area, the knowledge item is migrated; if they are consistent, the original storage location is retained.
[0058] The hierarchical storage results record the knowledge identifier, dynamic weight, current storage area, target storage area, and storage processing time, which are used to indicate the change in the position of the knowledge item in this hierarchical storage. After migration or retention is completed, the dynamic weight, hierarchical storage results, and node generation time are combined into a changed version node.
[0059] The initial version node serves as the first version node in the knowledge tracing record. Based on the version generation time, the current last version node in the knowledge tracing record is determined. The version identifier of the current last version node is written into the changed version node, and the changed version node is then appended to the current last version node, ensuring that all version nodes are arranged in the order of their generation.
[0060] When there are significant differences in the access time of knowledge items, a single archiving threshold is used to determine the target storage area. Knowledge items with a dynamic weight not lower than the archiving threshold are stored in the real-time data storage area, while knowledge items with a weight lower than the archiving threshold are migrated to the cold storage area, and the storage location and storage processing time before and after the migration are recorded.
[0061] In scenarios with finely divided storage resources, multiple weighted intervals are used to match different storage areas. High-weight knowledge items are configured in memory cache or high-speed databases, medium-weight knowledge items are configured in ordinary databases, and low-weight knowledge items are configured in compressed storage areas. Each time the storage level changes, a change version node containing the identifier of the previous version is generated and appended to the corresponding knowledge tracing record.
[0062] In the fintech business, regulatory provisions and risk control requirements with higher dynamic weights are stored in the real-time data storage area, while temporary business notifications with lower weights are migrated to the cold storage area. Each change in storage location generates a version change node to record the weights and storage status changes of financial knowledge.
[0063] In the healthcare business, frequently updated and frequently accessed drug risk requirements are stored in a high-speed storage area, while periodic health notifications with reduced applicability are moved to a low-frequency access storage area. The dynamic weights, migration locations, and processing times of knowledge entries are written to the version change node, enabling continuous querying of the hierarchical storage process for healthcare knowledge.
[0064] This embodiment determines the storage level of knowledge items through dynamic weights, which can match the storage location with the current importance of the knowledge and reduce the long-term occupation of high-performance storage resources by low-weight knowledge. By writing the dynamic weights and hierarchical storage results into the sequentially arranged change version nodes, the changes in the storage location of knowledge items and the basis for the changes can be retained, improving the traceability of the hierarchical storage process.
[0065] In one embodiment, such as Figure 3 As shown, step S10 above includes: S101, collect raw knowledge data from multiple data sources, and obtain data format identifier and data source identifier corresponding to the raw knowledge data; S102, based on the data form identifier, the original knowledge data is divided into structured knowledge data, unstructured knowledge data and time-series knowledge data. Field names and field values are extracted from the structured knowledge data, the text body is extracted from the unstructured knowledge data, and time field values are extracted from the time-series knowledge data. Candidate knowledge content is formed based on the field names and field values, the text body and the time field values. S103, unify the field names, character encoding and time base of the candidate knowledge content, combine the candidate knowledge content to generate knowledge content, and generate a content summary identifier based on the knowledge content; S104, Identify duplicate knowledge content based on the data source identifier, the content summary identifier and the time field value, merge the duplicate knowledge content, and encapsulate the merged knowledge content and the non-duplicate knowledge content into knowledge entries respectively, generating a knowledge entry set containing knowledge content; S105, generate a knowledge identifier for each knowledge item in the knowledge item set based on the data source identifier, the content summary identifier, and the time field value; S106, determine the knowledge category based on the knowledge content of each knowledge entry and the time field value, combine the knowledge identifier, the knowledge category, the data source identifier and the time field value into knowledge metadata, and configure the knowledge metadata to the corresponding knowledge entry.
[0066] In this embodiment, raw knowledge data is collected from multiple data sources through data interfaces, database queries, file reading, or message reception, and a data format identifier and a data source identifier are recorded for each piece of raw knowledge data. The data format identifier is used to distinguish between field-based records, text content, and data with time attributes, while the data source identifier is used to determine the location where the data was generated or to provide its source, so that the subsequently generated knowledge content can retain the source association.
[0067] The raw knowledge data is categorized based on its data format identifier. For structured knowledge data, field names and values are read according to field mapping relationships, and semantically related field values within the same data record are grouped into content units. For unstructured knowledge data, titles, body text, paragraphs, and table text are read through text parsing, and the text portions that can independently express meaning are identified as the text body. For time-series knowledge data, publication time, effective time, expiration time, update time, or data record time are read to form time field values. Field names, field values, text body, and time field values are combined according to their source relationships to form candidate knowledge content.
[0068] Synonymous field names in candidate knowledge content are converted to unified field names according to a field mapping table, character content is converted to unified character encoding, and time field values are converted to unified time zones, date formats, and time precision. After format unification, field values, text bodies, and time field values belonging to the same semantic object are combined to generate knowledge content. The format-unified knowledge content undergoes summarization processing to generate content summary identifiers that reflect the content status; identical content results in the same summary, while content changes result in different summary results.
[0069] Duplicate knowledge content is identified based on the data source identifier, content summary identifier, and time field value. Knowledge content with the same data source, the same content summary identifier, and a time field value that meets the preset consistency criteria is identified as duplicate content and merged into a single knowledge content. Knowledge content that does not meet the duplication criteria is retained separately. The merged knowledge content and the non-duplicate knowledge content are encapsulated into knowledge entries using a unified data structure, and multiple knowledge entries form a knowledge entry set.
[0070] The data source identifier, content summary identifier, and time field value are combined according to a preset coding order to generate a knowledge identifier for each knowledge entry. After generation, the knowledge identifier undergoes uniqueness verification; if an identifier conflict occurs, a distinguishing code is added and the identifier is regenerated. Each knowledge identifier is permanently associated with its corresponding knowledge entry, used to differentiate between different knowledge entries within the knowledge entry set.
[0071] Knowledge categories are determined based on the theme, purpose, and semantic scope reflected in the knowledge content, combined with the validity period, update cycle, or timeliness attribute reflected in the time field value. Category determination can employ field matching, keyword matching, or semantic similarity comparison. Knowledge identifiers, knowledge categories, data source identifiers, and time field values are combined according to a unified field structure to form knowledge metadata, which is then written into the corresponding knowledge entries, creating a searchable data object from knowledge content, source information, category information, and time information.
[0072] This embodiment reduces inconsistencies caused by differences in field names, character encoding, and time formats by classifying, extracting, and unifying the format of multi-source heterogeneous data; it reduces the number of duplicate knowledge entries by identifying and merging duplicate content through content summary identifiers and time field values; and it improves the accuracy of knowledge data organization, identification, and classification by generating unique knowledge identifiers and configuring knowledge metadata containing knowledge category, data source, and time information, giving each knowledge entry a clear data identity and structured attributes.
[0073] In one embodiment, such as Figure 4 As shown, step S20 above includes: S201, Obtain the knowledge identifier of each knowledge item, as well as the knowledge content and knowledge metadata when each knowledge item is generated, generate a traceability record identifier based on the knowledge identifier, and establish a knowledge traceability record based on the traceability record identifier; S202, organize the knowledge content and knowledge metadata of each knowledge entry according to the field type, generate an initial content snapshot, generate version summary data based on the initial content snapshot, and obtain the node generation time; S203, combine the initial content snapshot, the version summary data, the node generation time, and the source record identifier into an initial version node, and mark the initial version node as the version starting point of the knowledge source record; S204, write the initial version node into the knowledge traceability record based on the traceability record identifier, and establish the correspondence between the knowledge identifier, the traceability record identifier and the initial version node.
[0074] In this embodiment, the knowledge identifier of each knowledge entry is read, and the knowledge content and knowledge metadata saved when the knowledge entry was generated are retrieved. The knowledge identifier is used to distinguish different knowledge entries, and the knowledge content and knowledge metadata retain the data state at the time of generation to prevent subsequent content changes from overwriting the initial state. The knowledge metadata may include knowledge category, data source identifier, time field value, and content summary identifier.
[0075] A source traceability record identifier is generated based on the knowledge identifier. The source traceability record identifier can be generated by combining the knowledge identifier and the namespace tag, or it can be assigned through a unique mapping table. After generation, uniqueness verification is performed, and a knowledge source traceability record containing the source traceability record identifier, the knowledge identifier, the version node set, and the record creation time is established, giving each knowledge entry an independent record space.
[0076] Knowledge content and metadata are organized according to content, category, source, and time fields. Field names, character encoding, null value formats, and time precision are standardized, and the data is serialized according to a fixed field order to generate an initial content snapshot. This initial content snapshot preserves the complete data state at the time of knowledge entry generation. A summary transformation is performed on the initial content snapshot to generate version summary data, while clock data is read to form the node generation time. The version summary data is used to identify the snapshot content, and the node generation time is used to determine the generation sequence of version nodes.
[0077] The initial content snapshot, version summary data, node generation time, and source record identifier are combined into an initial version node, which is then configured with a node identifier and the first version sequence number. Based on the source record identifier, the knowledge source record is located, and the initial version node is saved to the version node set via atomic write, establishing a mapping relationship between the knowledge identifier, the source record identifier, and the initial version node. Before writing, the node identifier and version sequence number are verified; if the same initial version node already exists, duplicate writing is stopped.
[0078] This embodiment establishes a knowledge traceability record for each knowledge entry and solidifies the knowledge content and knowledge metadata at the time of generation into an initial version node with version summary data and node generation time, which can completely preserve the initial data state of the knowledge entry. By controlling node writing through unique identifiers, version sequence and mapping relationships, it can reduce data confusion between different knowledge entries and duplicate writing of initial nodes, and improve the integrity and traceability of the initial state record of knowledge.
[0079] In one embodiment, such as Figure 5 As shown, step S30 above includes: S301, extract the main text, effectiveness description text, scope of application text and body text from the knowledge content of each knowledge item, and combine them into classified text data according to the preset field order; S302, perform text segmentation and tag encoding on the classified text data to generate a classified text sequence; S303, input the classified text sequence into the text classification model trained by classification to perform compliance constraint level classification processing, and generate level probability data corresponding to multiple candidate compliance constraint levels respectively; S304. Compare the probability values of the probability data at each level, and determine the candidate compliance constraint level corresponding to the probability data of the level with the highest probability value as the compliance constraint level. S305, based on the preset level coefficient correspondence, determine the mandatory retention coefficient corresponding to the compliance constraint level, write the compliance constraint level and the mandatory retention coefficient into the knowledge metadata of the corresponding knowledge entry, and establish the correspondence between the knowledge identifier, the compliance constraint level and the mandatory retention coefficient.
[0080] In this embodiment, the main body text, effectiveness description text, scope of application text, and body text are identified from the knowledge content of each knowledge entry. For data with fixed fields, the corresponding text is read based on the field name and field position; for continuous text, the text interval is determined based on the heading level, paragraph boundaries, main body name thesaurus, effectiveness semantic thesaurus, and scope limiting statements. The main body text records the source of the content publication, the effectiveness description text reflects semantics such as mandatory, restrictive, permitted, or guiding, the scope of application text records the object, region, business scope, and effective period, and the body text retains specific requirements. Missing fields are marked with null values, duplicate text is merged, and character encoding, punctuation, and whitespace characters are standardized.
[0081] The text is arranged according to a preset field order: main text, effectiveness description text, scope of application text, and body text. Field separators are inserted between adjacent fields to generate categorized text data. Body text exceeding a preset length is segmented according to paragraph boundaries, and shorter text is padded with padding. Each text segment is converted into tokens using a token segmenter, and tokens are assigned token numbers, field category numbers, position numbers, and valid position markers to generate a categorized text sequence. Field category numbers are used to distinguish fields from different sources, position numbers are used to preserve token order, and valid position markers are used to exclude the influence of padded content on the classification results.
[0082] The text classification model comprises a text embedding layer, a multi-layer semantic encoding layer, a feature aggregation layer, and a classification output layer. The text embedding layer converts word IDs, field category IDs, and position IDs into vector representations. The multi-layer semantic encoding layer extracts semantic relationships between fields and within text segments. The feature aggregation layer converts the encoding results of each word ID into classification features. The classification output layer outputs the probability data for each candidate compliance constraint level. Candidate compliance constraint levels are set into multiple ordered categories according to constraint strength.
[0083] The model training data consists of knowledge text and corresponding level labels. The knowledge text is generated into training sequences according to the same field order, segmentation method, and label encoding method. The training sequences are input into the text classification model, and the parameters of the text embedding layer, semantic encoding layer, and classification output layer are adjusted based on the cross-entropy loss between the classification output and the level labels. The training data is divided into training and validation datasets. Model parameters are selected based on the validation loss or classification accuracy, and training stops when the validation results no longer improve after several consecutive training epochs. The model input is the classification text sequence, and the output is the probability value corresponding to each candidate compliance constraint level.
[0084] The text sequence to be classified is input into the trained text classification model. The probability values of each level of probability data are compared, and the candidate compliance constraint level corresponding to the highest probability value is determined as the compliance constraint level. When multiple probability values are the same, one of the candidate compliance constraint levels is selected according to the preset priority order, so that the same input can obtain consistent classification results.
[0085] The correspondence between compliance constraint levels and their corresponding coefficients is stored using a level mapping table. Each compliance constraint level is configured with a mandatory retention coefficient, with higher constraint strength levels receiving larger coefficient values. The level mapping table is queried based on the compliance constraint level to retrieve the corresponding mandatory retention coefficient, and the compliance constraint level and mandatory retention coefficient are written into the knowledge metadata of the corresponding knowledge entry. Knowledge identifiers are used to establish the association between knowledge entries, compliance constraint levels, and mandatory retention coefficients, avoiding confusion in the classification results of different knowledge entries.
[0086] This embodiment extracts and uniformly encodes the publishing subject, validity semantics, scope of application, and main text content into separate fields, and outputs level probability data using a text classification model trained with labeled data. This reduces level deviations caused by missing fields, differences in text structure, and differences in human judgment. By mapping the compliance constraint level corresponding to the highest probability to a mandatory retention coefficient and writing it into knowledge metadata based on knowledge identifiers, the accuracy and consistency of compliance constraint information quantification and associated storage can be improved.
[0087] In one embodiment, such as Figure 6 As shown, step S40 above includes: S401, Read the knowledge category from the knowledge metadata of each knowledge item, and determine the forgetting half-life parameter corresponding to the knowledge category based on the preset category period corresponding data; S402, Obtain review records based on the knowledge identifiers of each knowledge item; S403, when a review record is obtained, the review time is read from the review record, and the review record is arranged according to the review time. The latest review time is extracted from the arranged review record, the latest review time is determined as the weighted benchmark time, and the number of the review record is determined as the cumulative number of reviews. S404, when no review record is obtained, the entry generation time is read from the knowledge metadata of the corresponding knowledge entry, the entry generation time is determined as the weight benchmark time, and the preset initial review number is determined as the cumulative review number; S405, combine the weighted benchmark time and the cumulative number of reviews into review status data; S406, combine the forgetting half-life parameter and the review status data into knowledge lifecycle parameter data, and establish the correspondence between the knowledge lifecycle parameter data and the knowledge identifier of the corresponding knowledge item.
[0088] In this embodiment, knowledge categories are read from the knowledge metadata of each knowledge entry, and the knowledge category is used as the query key to access the data corresponding to the preset category period. The preset category period data records the mapping relationship between different knowledge categories and period parameters, and the period parameters are represented using a uniform time unit. After finding a unique match, the period value in the match is determined as the forgetting half-life parameter; when multiple matches exist, the period value with the highest degree of matching with the knowledge category is selected according to the category level or applicable scope; when no match is found, a preset default period value is used to avoid missing fields in the knowledge life cycle parameter data.
[0089] The review records are queried based on the knowledge identifiers of each knowledge item. Each review record includes at least a record identifier, review time, and review status. The knowledge identifier is used to distinguish the review data corresponding to different knowledge items. If the query results contain records with invalid status, duplicate record identifiers, or missing review times, invalid records are excluded and duplicate records are merged to ensure that the number of records reflects the number of valid reviews.
[0090] When valid review records exist, they are arranged from earliest to latest review time, with the latest review time determined as the weighted benchmark time. The cumulative number of reviews is represented by the number of valid review records, reflecting the frequency of reviews completed for each knowledge item. When multiple review records have the same review time, their arrangement can be determined based on the record identifier or the record entry time, without affecting the selection of the latest record.
[0091] When no valid review record exists, the entry generation time is read from the knowledge metadata and set as the weighted benchmark time. The cumulative review count uses a preset initial review count to ensure that knowledge entries that have not yet been reviewed still have complete review status data. If the entry generation time is missing, the corresponding field can be supplemented using the time configured in the knowledge metadata.
[0092] The weighted baseline time and cumulative number of reviews are combined into review status data using a unified field structure. Then, the forgetting half-life parameter is combined with the review status data to form knowledge lifecycle parameter data. The knowledge lifecycle parameter data includes at least the knowledge identifier, forgetting half-life parameter, weighted baseline time, and cumulative number of reviews, and a unique correspondence is established between the knowledge identifier and the corresponding knowledge item.
[0093] This embodiment determines differentiated forgetting half-life parameters by mapping the relationship between knowledge categories and cycle parameters, and generates weighted benchmark time and cumulative number of reviews based on valid review records. This can reduce life cycle status deviations caused by the lack of unified cycle parameters and review data. Necessary fields are supplemented by default cycle values, entry generation time and preset initial number of reviews, so that the knowledge life cycle parameter data remains complete and can accurately distinguish knowledge entries of different categories and different review statuses.
[0094] In one embodiment, such as Figure 7 As shown, step S50 above includes: S501, based on the knowledge identifier of each knowledge item, filter the event records in the business event records whose event knowledge identifier is the same as the knowledge identifier of each knowledge item, obtain the business events associated with each knowledge item, and read the event occurrence time and event type from the business events; S502, read the forgetting half-life parameter and review status data from the knowledge lifecycle parameter data, read the weighted baseline time and cumulative review count from the review status data, determine the event filtering period based on the weighted baseline time and the current time, and filter target business events whose event occurrence time is within the event filtering period from the business events; S503, when a target business event is selected, the event influence direction corresponding to the event type of each target business event is determined based on the preset event influence corresponding data. When the influence directions of each event are consistent, the consistent event influence direction is determined as the target event influence direction. When the influence directions of each event are inconsistent, the event influence direction corresponding to the target business event with the latest event occurrence time is determined as the target event influence direction, and the event status data is determined based on the target event influence direction. S504: When no target business event is found, the "no event" status will be determined as the event status data. S505, determine the event contribution data corresponding to the event state data based on the preset state contribution data, generate time decay data based on the current time, the weighted benchmark time, and the forgetting half-life parameter, and generate review incentive data based on the cumulative review count; S506, the time decay data and the event contribution data are weighted based on a preset balance coefficient to generate time event weights, and candidate dynamic weights are generated based on the time event weights, the forced retention coefficient and the review incentive data; S507, compare the candidate dynamic weights with the preset weight upper limit, determine the smaller value between the candidate dynamic weights and the preset weight upper limit as the dynamic weight of each knowledge item, and establish the correspondence between the dynamic weights and the knowledge identifiers of the corresponding knowledge items.
[0095] In this embodiment, the knowledge identifier of each knowledge entry is used as the matching key to retrieve the event knowledge identifier in the business event records, and event records with matching identifiers are grouped into the corresponding knowledge entries. A business event includes at least the event occurrence time and the event type. The event occurrence time is used to determine whether the event is within a valid filtering range, and the event type is used to identify whether the event enhances, weakens, or maintains the knowledge state.
[0096] The forgetting half-life parameter and review status data are read from the knowledge lifecycle parameter data. Then, the weighted baseline time and cumulative review count are read from the review status data. The weighted baseline time and the current time form an event filtering period. Business events occurring within this period are identified as target business events, ensuring that dynamic weights are only affected by events within the current lifecycle stage. The forgetting half-life parameter controls the rate at which knowledge status decreases over time, and the cumulative review count reflects the frequency with which knowledge content has been verified.
[0097] Based on the preset event impact data, the event type of each target business event is converted into an event impact direction. When the impact directions of all events are consistent, the consistent event impact direction is directly adopted; when the impact directions of all events are inconsistent, the most recently occurring target business event is selected based on the event occurrence time, and the event impact direction corresponding to that event is determined as the target event impact direction. When multiple target business events have the same most recently occurring time, the target event impact direction can be determined based on the event record write time or the preset event priority.
[0098] Event status data is generated based on the direction of the target event's impact. This data can include enhanced, weakened, and maintained states. If no target business event is found, a "no event" state is generated, ensuring that knowledge entries without specific business events still have complete event status data. Based on preset state contribution data, the event status data is converted into event contribution data, with different event states configured with positive, negative, or neutral contribution values.
[0099] Time decay data is generated based on the time interval between the current time and the weighted baseline time, combined with the forgetting half-life parameter. As the time interval increases, the time decay data decreases according to the decay rate corresponding to the forgetting half-life parameter. Review incentive data is generated based on the cumulative number of reviews. As the cumulative number of reviews increases, the review incentive data increases according to a preset growth range, and an incentive cap can be set to limit excessive weight increase caused by repeated reviews.
[0100] The time decay data and event contribution data are weighted and summed separately according to preset balancing coefficients to generate time event weights. The balancing coefficients are used to adjust the proportion of the influence of time changes and business events on the weight results. Each balancing coefficient can be configured according to knowledge management needs, but the same coefficient caliber is used for knowledge items in the same batch.
[0101] The gain of time event weights is adjusted using review incentive data to generate state weights. These state weights are then compared with a mandatory retention coefficient, and the larger of the two is selected as the candidate dynamic weight, ensuring the mandatory retention coefficient constitutes the minimum retention value for the candidate dynamic weights. The candidate dynamic weights are then compared with a preset weight upper limit, and the smaller of the two is taken as the dynamic weight, preventing the combined contribution of time events or review incentives from exceeding the set range. Each dynamic weight is uniquely associated with a corresponding knowledge item through a knowledge identifier, preventing confusion between weight data from different knowledge items.
[0102] This embodiment accurately collects business events through knowledge identification and filters target business events within the current lifecycle stage using weighted benchmark time and current time, which can reduce the interference of expired events on knowledge status judgment. Time event weights are generated by combining event impact direction, time decay data, and review incentive data. A minimum retention value is set with a mandatory retention coefficient, and the weight range is limited by a preset weight upper limit. This allows dynamic weights to simultaneously reflect time changes, event impact, review status, and retention requirements, improving the accuracy, completeness, and stability of knowledge item weight generation.
[0103] In one embodiment, such as Figure 8 As shown, step S60 above includes: S601, compare the dynamic weight of each knowledge item with the preset archiving threshold, determine the target storage area corresponding to the knowledge item whose dynamic weight is lower than the preset archiving threshold as the cold storage area, and determine the target storage area corresponding to the knowledge item whose dynamic weight is not lower than the preset archiving threshold as the real-time data storage area. S602, obtain the current storage area of each knowledge item; when the current storage area is inconsistent with the target storage area, migrate the corresponding knowledge item to the target storage area and update the storage location of the corresponding knowledge item to the target storage area. S603, when the current storage area is consistent with the target storage area, the corresponding knowledge entry is retained in the current storage area; S604, obtain the storage processing time, and generate a hierarchical storage result based on the knowledge identifier of each knowledge item, the dynamic weight, the current storage area, the target storage area, and the storage processing time; S605, taking the initial version node as the version starting point, determine the current last version node in the corresponding knowledge tracing record according to the version generation order, and obtain the version identifier of the current last version node. S606, obtain the node generation time, combine the dynamic weight, the hierarchical storage result, the version identifier of the current last version node, and the node generation time into a changed version node, and write the changed version node as the next version node of the current last version node into the corresponding knowledge tracing record.
[0104] In this embodiment, the dynamic weight of each knowledge item is numerically compared with a preset archiving threshold. When the dynamic weight is lower than the preset archiving threshold, the corresponding target storage area is designated as a cold storage area; when the dynamic weight is not lower than the preset archiving threshold, the corresponding target storage area is designated as a real-time data storage area. The preset archiving threshold uses the same numerical range and precision as the dynamic weight to avoid comparison deviations caused by different data formats.
[0105] The storage location record of a knowledge entry is retrieved based on its knowledge identifier to obtain the current storage area. If the current storage area differs from the target storage area, the knowledge content, knowledge metadata, and storage association information are copied to the target storage area. After integrity verification, the storage location record is updated, and the corresponding data in the original storage area is deleted or marked. The knowledge identifier is retained during the migration process to ensure that the knowledge entry maintains the same data identity before and after the migration. If the current storage area and the target storage area are identical, the physical storage location of the knowledge entry remains unchanged.
[0106] After storage migration is complete or the original storage location is retained, the storage processing time is read, and the knowledge identifier, dynamic weight, current storage area, target storage area, and storage processing time are combined into a hierarchical storage result according to a unified field structure. The current storage area records the position before adjustment, and the target storage area records the position determined according to the dynamic weight. If the two are the same, it means that no migration has occurred; if they are different, it means that a change in storage level has occurred.
[0107] The system locates the corresponding knowledge source record based on the knowledge identifier, using the initial version node as the first version node. Version nodes are then arranged according to their generation time or version sequence, with the last version node in the sequence being determined as the current last version node. The version identifier of this last version node is then read. The version identifier is used to record the succession relationship between the changed version node and the previous version node.
[0108] The node generation time is read, and the dynamic weight, hierarchical storage result, version identifier of the current last version node, and node generation time are combined to form a changed version node. The changed version node records the dynamic weight and its corresponding storage adjustment result, and configures the version identifier of the current last version node as the previous version identifier. Based on the knowledge identifier, the changed version node is written to the corresponding knowledge tracing record, and its version order is set after the current last version node, ensuring that version nodes are arranged consecutively according to their generation order.
[0109] This embodiment determines the target storage area by comparing dynamic weights with preset archiving thresholds, and controls the migration of knowledge items based on the differences between the current storage area and the target storage area, so that the storage level of knowledge items can match the current weight status. By writing dynamic weights, storage locations before and after migration, and storage processing time into the change version node, and saving them to the knowledge traceability record in the order of version generation, the judgment basis and location changes of hierarchical storage can be completely recorded, improving the accuracy of storage resource allocation and the traceability of the hierarchical storage process.
[0110] In one embodiment, an insurance technology group collects raw knowledge data from its core system, underwriting platform, claims case database, product database, and regulatory information sources. Rate tables and underwriting conditions are marked as structured knowledge data; insurance terms and regulatory letters are marked as unstructured knowledge data; and product launch dates, discontinuation dates, and policy validity periods are marked as time-series knowledge data. Field names and values, text content, and time field values are extracted separately, and field names, character encodings, and time bases are standardized to generate knowledge content and content summary identifiers. Duplicate content is identified based on the data source identifier, content summary identifier, and time field value. The merged content and non-duplicate content are encapsulated into knowledge entries, and knowledge identifiers are generated. Regulatory legal provisions, insurance product terms, and temporary business notices are configured as core principle categories, product term categories, and temporary notice categories, respectively. Knowledge identifiers, knowledge categories, data source identifiers, and time field values are combined into knowledge metadata.
[0111] For each knowledge identifier, a traceability record identifier is generated, and a knowledge traceability record is established. The knowledge content and knowledge metadata at the time of knowledge entry generation are organized into an initial content snapshot, version summary data is generated, and the node generation time is recorded. These are then combined to form an initial version node, which is written into the corresponding knowledge traceability record. For example, when an insurance product clause is first entered into the database, the clause text, product name, effective date, and data source are all stored in the initial version node.
[0112] The system extracts the main text, effectiveness description text, scope of application text, and body text from the knowledge content. After segmentation and tagging encoding, these are input into BERT-base-Chinese (a Chinese basic bidirectional encoder representation model). The model outputs probability data for multiple candidate compliance constraint levels and determines the level corresponding to the highest probability as the compliance constraint level. Legal provisions, regulatory rules, industry practices, and business information can each be assigned a different level, with corresponding mandatory retention coefficients configured. These are then written into the knowledge metadata and associated with the knowledge identifier.
[0113] The forgetting half-life parameter is determined based on the knowledge category: a longer period is used for core principles, an annual period for product terms, and a shorter period for temporary notices. Review records are retrieved based on the knowledge identifier, with the most recent review time determined as the weighted baseline time, and the number of review records determined as the cumulative review count. If no review records exist, the entry generation time and the preset initial review count are used. The forgetting half-life parameter, the weighted baseline time, and the cumulative review count together form the knowledge lifecycle parameter data.
[0114] Business event logs are associated with knowledge entries through event knowledge identifiers. Regulatory policy releases, product launches, and clause revisions can be configured as enhanced events, while product discontinuation, document repeal, and notice expiration can be configured as mitigated events. Target business events are filtered based on weighted baseline time and current time, and the direction of event impact is determined according to event type. Event status is converted into event contribution data; current time, weighted baseline time, and forgetting half-life parameters generate time decay data, and cumulative review counts generate review incentive data. The time decay data and event contribution data are weighted to form a time event weight, which is then combined with a mandatory retention coefficient and review incentive data to generate candidate dynamic weights. The dynamic weight is obtained by setting a preset weight cap.
[0115] The dynamic weight of each knowledge item is compared with a preset archiving threshold. Knowledge items with a weight not lower than the threshold are stored in a Redis (remote dictionary service) cluster or a real-time database, while those with a weight lower than the threshold are migrated to a cold storage area. The knowledge identifier, dynamic weight, current storage area, target storage area, and storage processing time together form the hierarchical storage result. The dynamic weight, hierarchical storage result, version identifier of the current last version node, and node generation time are combined into a changed version node and appended to the corresponding knowledge tracing record, so that the weight changes and storage locations of the same knowledge item can be continuously saved according to the version generation order.
[0116] In one embodiment, a lifecycle-based traceability hierarchical storage device is provided, which corresponds one-to-one with the lifecycle-based traceability hierarchical storage method described in the above embodiments. (Refer to...) Figure 9 , Figure 9This is a functional module diagram of a preferred embodiment of the lifecycle-based traceability hierarchical storage device of the present invention. The lifecycle-based traceability hierarchical storage device includes: a knowledge entry construction module 10, a knowledge traceability establishment module 20, a compliance constraint assessment module 30, a lifecycle parameter generation module 40, a dynamic weight generation module 50, and a hierarchical storage update module 60. Detailed descriptions of each functional module are as follows: The knowledge entry construction module 10 is used to acquire raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate a knowledge identifier for each knowledge entry in the knowledge entry set, and configure knowledge metadata including knowledge category. The knowledge tracing establishment module 20 is used to establish a knowledge tracing record based on the knowledge identifier of each knowledge item, generate an initial version node based on the knowledge content and knowledge metadata when each knowledge item is generated, and write the initial version node into the knowledge tracing record. The compliance constraint assessment module 30 is used to input the knowledge content of each knowledge item into the text classification model for compliance constraint level classification processing, obtain the compliance constraint level of each knowledge item, and determine the mandatory retention coefficient based on the compliance constraint level; The life cycle parameter generation module 40 is used to determine the forgetting half-life parameter based on the knowledge category of each knowledge item, obtain the review status data of each knowledge item, and generate knowledge life cycle parameter data. The dynamic weight generation module 50 is used to obtain business events associated with each knowledge item based on the knowledge identifier of each knowledge item, and generate dynamic weights for each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events and the mandatory retention coefficient. The hierarchical storage update module 60 is used to perform hierarchical storage of each knowledge item based on the dynamic weight of each knowledge item, generate hierarchical storage results, generate change version nodes based on the dynamic weight and the hierarchical storage results, and append the change version nodes to the corresponding knowledge traceability records according to the version generation order, with the initial version node as the version starting point.
[0117] In one embodiment, the knowledge entry construction module 10 is specifically used for: Collect raw knowledge data from multiple data sources and obtain data format identifiers and data source identifiers corresponding to the raw knowledge data; Based on the data format identifier, the original knowledge data is divided into structured knowledge data, unstructured knowledge data, and time-series knowledge data. Field names and field values are extracted from the structured knowledge data, the text body is extracted from the unstructured knowledge data, and time field values are extracted from the time-series knowledge data. Candidate knowledge content is formed based on the field names and field values, the text body, and the time field values. Unify the field names, character encodings, and time bases of the candidate knowledge content, combine the candidate knowledge content to generate knowledge content, and generate a content summary identifier based on the knowledge content; Based on the data source identifier, the content summary identifier, and the time field value, duplicate knowledge content is identified, the duplicate knowledge content is merged, and the merged knowledge content and the non-duplicate knowledge content are encapsulated into knowledge entries to generate a set of knowledge entries containing knowledge content. A knowledge identifier is generated for each knowledge item in the knowledge item set based on the data source identifier, the content summary identifier, and the time field value. The knowledge category is determined based on the knowledge content of each knowledge entry and the time field value. The knowledge identifier, the knowledge category, the data source identifier, and the time field value are combined into knowledge metadata, and the knowledge metadata is configured to the corresponding knowledge entry.
[0118] In one embodiment, the knowledge tracing module 20 is specifically used for: Obtain the knowledge identifier of each knowledge item, as well as the knowledge content and knowledge metadata when each knowledge item is generated; generate a traceability record identifier based on the knowledge identifier; and establish a knowledge traceability record based on the traceability record identifier. Organize the knowledge content and knowledge metadata of each knowledge entry according to the field type, generate an initial content snapshot, generate version summary data based on the initial content snapshot, and obtain the node generation time; The initial content snapshot, the version summary data, the node generation time, and the source record identifier are combined into an initial version node, and the initial version node is marked as the version starting point of the knowledge source record; The initial version node is written into the knowledge traceability record based on the traceability record identifier, and a correspondence is established between the knowledge identifier, the traceability record identifier, and the initial version node.
[0119] In one embodiment, the compliance constraint assessment module 30 is specifically used for: Extract the main text, effectiveness description text, scope of application text, and body text from the knowledge content of each knowledge entry, and combine them into categorized text data according to the preset field order; The categorized text data is segmented and coded to generate a categorized text sequence; The classified text sequence is input into a text classification model trained for classification to perform compliance constraint level classification processing, generating level probability data corresponding to multiple candidate compliance constraint levels; Compare the probability values of the probability data at each level, and determine the candidate compliance constraint level corresponding to the probability data of the level with the highest probability value as the compliance constraint level; Based on the preset level coefficient correspondence, the mandatory retention coefficient corresponding to the compliance constraint level is determined, the compliance constraint level and the mandatory retention coefficient are written into the knowledge metadata of the corresponding knowledge entry, and the correspondence between the knowledge identifier, the compliance constraint level and the mandatory retention coefficient is established.
[0120] In one embodiment, the lifecycle parameter generation module 40 is specifically used for: The knowledge category is read from the knowledge metadata of each knowledge item, and the forgetting half-life parameter corresponding to the knowledge category is determined based on the data corresponding to the preset category period. Review records are obtained based on the knowledge identifiers of each knowledge item; When a review record is obtained, the review time is read from the review record, and the review record is arranged according to the review time. The latest review time is extracted from the arranged review record, and the latest review time is determined as the weighted benchmark time. The number of the review record is determined as the cumulative number of reviews. When no review record is obtained, the entry generation time is read from the knowledge metadata of the corresponding knowledge entry, the entry generation time is determined as the weighted benchmark time, and the preset initial review count is determined as the cumulative review count; The weighted benchmark time and the cumulative number of reviews are combined to form review status data; The forgetting half-life parameter and the review status data are combined into knowledge lifecycle parameter data, and a correspondence is established between the knowledge lifecycle parameter data and the knowledge identifier of the corresponding knowledge item.
[0121] In one embodiment, the dynamic weight generation module 50 is specifically used for: Based on the knowledge identifier of each knowledge item, the event records with the same event knowledge identifier as each knowledge item are filtered from the business event records to obtain the business events associated with each knowledge item, and the event occurrence time and event type are read from the business events; The forgetting half-life parameter and review status data are read from the knowledge lifecycle parameter data. The weighted baseline time and cumulative review count are read from the review status data. The event filtering period is determined based on the weighted baseline time and the current time. Target business events whose occurrence time is within the event filtering period are filtered from the business events. When a target business event is selected, the event impact direction corresponding to the event type of each target business event is determined based on the preset event impact corresponding data. When the impact directions of each event are consistent, the consistent event impact direction is determined as the target event impact direction. When the impact directions of each event are inconsistent, the event impact direction corresponding to the target business event with the latest event occurrence time is determined as the target event impact direction, and the event status data is determined based on the target event impact direction. When no target business event is found, the "no event" status will be defined as the event status data. Based on the preset state contribution data, the event contribution data corresponding to the event state data is determined. Time decay data is generated based on the current time, the weighted baseline time, and the forgetting half-life parameter. Review incentive data is generated based on the cumulative number of reviews. The time decay data and the event contribution data are weighted based on a preset balance coefficient to generate time event weights, and candidate dynamic weights are generated based on the time event weights, the forced retention coefficient, and the review incentive data. The candidate dynamic weights are compared with the preset weight upper limit. The smaller value between the candidate dynamic weights and the preset weight upper limit is determined as the dynamic weight of each knowledge item, and a correspondence is established between the dynamic weights and the knowledge identifiers of the corresponding knowledge items.
[0122] In one embodiment, the hierarchical storage update module 60 is specifically used for: The dynamic weight of each knowledge item is compared with a preset archiving threshold. The target storage area corresponding to the knowledge item whose dynamic weight is lower than the preset archiving threshold is determined as a cold storage area, and the target storage area corresponding to the knowledge item whose dynamic weight is not lower than the preset archiving threshold is determined as a real-time data storage area. Obtain the current storage area of each knowledge entry. When the current storage area is inconsistent with the target storage area, migrate the corresponding knowledge entry to the target storage area and update the storage location of the corresponding knowledge entry to the target storage area. When the current storage area is consistent with the target storage area, the corresponding knowledge entries are retained in the current storage area; The storage processing time is obtained, and a hierarchical storage result is generated based on the knowledge identifier of each knowledge item, the dynamic weight, the current storage area, the target storage area, and the storage processing time. Taking the initial version node as the version starting point, determine the current last version node in the corresponding knowledge tracing record according to the version generation order, and obtain the version identifier of the current last version node; Obtain the node generation time, combine the dynamic weight, the hierarchical storage result, the version identifier of the current last version node, and the node generation time into a changed version node, and write the changed version node as the next version node of the current last version node into the corresponding knowledge tracing record.
[0123] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides determination and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a lifecycle-based traceable hierarchical storage method on the server side.
[0124] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides determination and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of a lifecycle-based traceable hierarchical storage method.
[0125] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate a knowledge identifier for each knowledge entry in the set of knowledge entries and configure knowledge metadata including knowledge category; A knowledge traceability record is established based on the knowledge identifier of each knowledge item. An initial version node is generated based on the knowledge content and knowledge metadata of each knowledge item at the time of its generation, and the initial version node is written into the knowledge traceability record. The knowledge content of each knowledge item is input into a text classification model for compliance constraint level classification processing to obtain the compliance constraint level of each knowledge item, and the mandatory retention coefficient is determined based on the compliance constraint level. Based on the knowledge category of each knowledge item, the forgetting half-life parameter is determined, the review status data of each knowledge item is obtained, and the knowledge life cycle parameter data is generated. Based on the knowledge identifier of each knowledge item, obtain the business events associated with each knowledge item, and generate the dynamic weight of each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient. Based on the dynamic weights of each knowledge item, each knowledge item is stored in a hierarchical manner to generate a hierarchical storage result. Based on the dynamic weights and the hierarchical storage result, a change version node is generated. Taking the initial version node as the version starting point, the change version node is appended to the corresponding knowledge traceability record according to the version generation order.
[0126] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, and a computer program is stored thereon, which, when executed by a processor, performs the following steps: Obtain raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate a knowledge identifier for each knowledge entry in the set of knowledge entries and configure knowledge metadata including knowledge category; A knowledge traceability record is established based on the knowledge identifier of each knowledge item. An initial version node is generated based on the knowledge content and knowledge metadata of each knowledge item at the time of its generation, and the initial version node is written into the knowledge traceability record. The knowledge content of each knowledge item is input into a text classification model for compliance constraint level classification processing to obtain the compliance constraint level of each knowledge item, and the mandatory retention coefficient is determined based on the compliance constraint level. Based on the knowledge category of each knowledge item, the forgetting half-life parameter is determined, the review status data of each knowledge item is obtained, and the knowledge life cycle parameter data is generated. Based on the knowledge identifier of each knowledge item, obtain the business events associated with each knowledge item, and generate the dynamic weight of each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient. Based on the dynamic weights of each knowledge item, each knowledge item is stored in a hierarchical manner to generate a hierarchical storage result. Based on the dynamic weights and the hierarchical storage result, a change version node is generated. Taking the initial version node as the version starting point, the change version node is appended to the corresponding knowledge traceability record according to the version generation order.
[0127] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.
[0128] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0129] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0130] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.
[0131] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A lifecycle-based traceable hierarchical storage method, characterized in that, Includes the following steps: Obtain raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate a knowledge identifier for each knowledge entry in the set of knowledge entries and configure knowledge metadata including knowledge category; A knowledge traceability record is established based on the knowledge identifier of each knowledge item. An initial version node is generated based on the knowledge content and knowledge metadata of each knowledge item at the time of its generation, and the initial version node is written into the knowledge traceability record. The knowledge content of each knowledge item is input into a text classification model for compliance constraint level classification processing to obtain the compliance constraint level of each knowledge item, and the mandatory retention coefficient is determined based on the compliance constraint level. Based on the knowledge category of each knowledge item, the forgetting half-life parameter is determined, the review status data of each knowledge item is obtained, and the knowledge life cycle parameter data is generated. Based on the knowledge identifier of each knowledge item, obtain the business events associated with each knowledge item, and generate the dynamic weight of each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient. Based on the dynamic weights of each knowledge item, each knowledge item is stored in a hierarchical manner to generate a hierarchical storage result. Based on the dynamic weights and the hierarchical storage result, a change version node is generated. Taking the initial version node as the version starting point, the change version node is appended to the corresponding knowledge traceability record according to the version generation order.
2. The lifecycle-based traceability hierarchical storage method as described in claim 1, characterized in that, Acquire raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate a knowledge identifier for each knowledge entry in the knowledge entry set, and configure knowledge metadata including knowledge category, including: Collect raw knowledge data from multiple data sources and obtain data format identifiers and data source identifiers corresponding to the raw knowledge data; Based on the data format identifier, the original knowledge data is divided into structured knowledge data, unstructured knowledge data, and time-series knowledge data. Field names and field values are extracted from the structured knowledge data, the text body is extracted from the unstructured knowledge data, and time field values are extracted from the time-series knowledge data. Candidate knowledge content is formed based on the field names and field values, the text body, and the time field values. Unify the field names, character encodings, and time bases of the candidate knowledge content, combine the candidate knowledge content to generate knowledge content, and generate a content summary identifier based on the knowledge content; Based on the data source identifier, the content summary identifier, and the time field value, duplicate knowledge content is identified, the duplicate knowledge content is merged, and the merged knowledge content and the non-duplicate knowledge content are encapsulated into knowledge entries to generate a set of knowledge entries containing knowledge content. A knowledge identifier is generated for each knowledge item in the knowledge item set based on the data source identifier, the content summary identifier, and the time field value. The knowledge category is determined based on the knowledge content of each knowledge entry and the time field value. The knowledge identifier, the knowledge category, the data source identifier, and the time field value are combined into knowledge metadata, and the knowledge metadata is configured to the corresponding knowledge entry.
3. The lifecycle-based traceability hierarchical storage method as described in claim 1, characterized in that, A knowledge tracing record is established based on the knowledge identifier of each knowledge entry. An initial version node is generated based on the knowledge content and knowledge metadata of each knowledge entry at the time of its generation, and the initial version node is written into the knowledge tracing record, including: Obtain the knowledge identifier of each knowledge item, as well as the knowledge content and knowledge metadata when each knowledge item is generated; generate a traceability record identifier based on the knowledge identifier; and establish a knowledge traceability record based on the traceability record identifier. Organize the knowledge content and knowledge metadata of each knowledge entry according to the field type, generate an initial content snapshot, generate version summary data based on the initial content snapshot, and obtain the node generation time; The initial content snapshot, the version summary data, the node generation time, and the source record identifier are combined into an initial version node, and the initial version node is marked as the version starting point of the knowledge source record; The initial version node is written into the knowledge traceability record based on the traceability record identifier, and a correspondence is established between the knowledge identifier, the traceability record identifier, and the initial version node.
4. The lifecycle-based traceability hierarchical storage method as described in claim 1, characterized in that, The knowledge content of each knowledge item is input into a text classification model for compliance constraint level classification, resulting in a compliance constraint level for each knowledge item. Based on these compliance constraint levels, a mandatory retention coefficient is determined, including: Extract the main text, effectiveness description text, scope of application text, and body text from the knowledge content of each knowledge entry, and combine them into categorized text data according to the preset field order; The categorized text data is segmented and coded to generate a categorized text sequence; The classified text sequence is input into a text classification model trained for classification to perform compliance constraint level classification processing, generating level probability data corresponding to multiple candidate compliance constraint levels; Compare the probability values of the probability data at each level, and determine the candidate compliance constraint level corresponding to the probability data of the level with the highest probability value as the compliance constraint level; Based on the preset level coefficient correspondence, the mandatory retention coefficient corresponding to the compliance constraint level is determined, the compliance constraint level and the mandatory retention coefficient are written into the knowledge metadata of the corresponding knowledge entry, and the correspondence between the knowledge identifier, the compliance constraint level and the mandatory retention coefficient is established.
5. The lifecycle-based traceability hierarchical storage method as described in claim 1, characterized in that, Based on the knowledge category of each knowledge item, the forgetting half-life parameter is determined, the review status data of each knowledge item is obtained, and knowledge lifecycle parameter data is generated, including: The knowledge category is read from the knowledge metadata of each knowledge item, and the forgetting half-life parameter corresponding to the knowledge category is determined based on the data corresponding to the preset category period. Review records are obtained based on the knowledge identifiers of each knowledge item; When a review record is obtained, the review time is read from the review record, and the review record is arranged according to the review time. The latest review time is extracted from the arranged review record, and the latest review time is determined as the weighted benchmark time. The number of the review record is determined as the cumulative number of reviews. When no review record is obtained, the entry generation time is read from the knowledge metadata of the corresponding knowledge entry, the entry generation time is determined as the weighted benchmark time, and the preset initial review count is determined as the cumulative review count; The weighted benchmark time and the cumulative number of reviews are combined to form review status data; The forgetting half-life parameter and the review status data are combined into knowledge lifecycle parameter data, and a correspondence is established between the knowledge lifecycle parameter data and the knowledge identifier of the corresponding knowledge item.
6. The lifecycle-based traceability hierarchical storage method as described in claim 1, characterized in that, Based on the knowledge identifier of each knowledge entry, business events associated with each knowledge entry are obtained. Dynamic weights for each knowledge entry are generated based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient, including: Based on the knowledge identifier of each knowledge item, the event records with the same event knowledge identifier as each knowledge item are filtered from the business event records to obtain the business events associated with each knowledge item, and the event occurrence time and event type are read from the business events; The forgetting half-life parameter and review status data are read from the knowledge lifecycle parameter data. The weighted baseline time and cumulative review count are read from the review status data. The event filtering period is determined based on the weighted baseline time and the current time. Target business events whose occurrence time is within the event filtering period are filtered from the business events. When a target business event is selected, the event impact direction corresponding to the event type of each target business event is determined based on the preset event impact corresponding data. When the impact directions of each event are consistent, the consistent event impact direction is determined as the target event impact direction. When the impact directions of each event are inconsistent, the event impact direction corresponding to the target business event with the latest event occurrence time is determined as the target event impact direction, and the event status data is determined based on the target event impact direction. When no target business event is found, the "no event" status will be defined as the event status data. Based on the preset state contribution data, the event contribution data corresponding to the event state data is determined. Time decay data is generated based on the current time, the weighted baseline time, and the forgetting half-life parameter. Review incentive data is generated based on the cumulative number of reviews. The time decay data and the event contribution data are weighted based on a preset balance coefficient to generate time event weights, and candidate dynamic weights are generated based on the time event weights, the forced retention coefficient, and the review incentive data. The candidate dynamic weights are compared with the preset weight upper limit. The smaller value between the candidate dynamic weights and the preset weight upper limit is determined as the dynamic weight of each knowledge item, and a correspondence is established between the dynamic weights and the knowledge identifiers of the corresponding knowledge items.
7. The lifecycle-based traceability hierarchical storage method as described in claim 1, characterized in that, Based on the dynamic weights of each knowledge item, hierarchical storage is performed on each knowledge item to generate a hierarchical storage result. Change version nodes are generated based on the dynamic weights and the hierarchical storage result. Starting from the initial version node, the change version nodes are appended to the corresponding knowledge tracing records according to the version generation order, including: The dynamic weight of each knowledge item is compared with a preset archiving threshold. The target storage area corresponding to the knowledge item whose dynamic weight is lower than the preset archiving threshold is determined as a cold storage area, and the target storage area corresponding to the knowledge item whose dynamic weight is not lower than the preset archiving threshold is determined as a real-time data storage area. Obtain the current storage area of each knowledge entry. When the current storage area is inconsistent with the target storage area, migrate the corresponding knowledge entry to the target storage area and update the storage location of the corresponding knowledge entry to the target storage area. When the current storage area is consistent with the target storage area, the corresponding knowledge entries are retained in the current storage area; The storage processing time is obtained, and a hierarchical storage result is generated based on the knowledge identifier of each knowledge item, the dynamic weight, the current storage area, the target storage area, and the storage processing time. Taking the initial version node as the version starting point, determine the current last version node in the corresponding knowledge tracing record according to the version generation order, and obtain the version identifier of the current last version node; Obtain the node generation time, combine the dynamic weight, the hierarchical storage result, the version identifier of the current last version node, and the node generation time into a changed version node, and write the changed version node as the next version node of the current last version node into the corresponding knowledge tracing record.
8. A lifecycle-based traceable hierarchical storage device, characterized in that, The lifecycle-based traceability hierarchical storage device includes: The knowledge entry construction module is used to acquire raw knowledge data, generate a set of knowledge entries containing knowledge content based on the raw knowledge data, generate knowledge identifiers for each knowledge entry in the knowledge entry set, and configure knowledge metadata including knowledge categories. The knowledge traceability establishment module is used to establish a knowledge traceability record based on the knowledge identifier of each knowledge item, generate an initial version node based on the knowledge content and knowledge metadata when each knowledge item is generated, and write the initial version node into the knowledge traceability record; The compliance constraint assessment module is used to input the knowledge content of each knowledge item into the text classification model for compliance constraint level classification processing, obtain the compliance constraint level of each knowledge item, and determine the mandatory retention coefficient based on the compliance constraint level; The lifecycle parameter generation module is used to determine the forgetting half-life parameter based on the knowledge category of each knowledge item, obtain the review status data of each knowledge item, and generate knowledge lifecycle parameter data. The dynamic weight generation module is used to obtain business events associated with each knowledge item based on the knowledge identifier of each knowledge item, and generate dynamic weights for each knowledge item based on the current time, the knowledge lifecycle parameter data, the business events, and the mandatory retention coefficient. The hierarchical storage update module is used to perform hierarchical storage of each knowledge item based on the dynamic weight of each knowledge item, generate hierarchical storage results, generate change version nodes based on the dynamic weight and the hierarchical storage results, and append the change version nodes to the corresponding knowledge traceability records according to the version generation order, with the initial version node as the version starting point.
9. A computer device, characterized in that, The computer device includes a memory, a processor, and a lifecycle-based hierarchical storage program stored on the memory and executable on the processor, wherein the lifecycle-based hierarchical storage program, when executed by the processor, implements the steps of the lifecycle-based hierarchical storage method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The storage medium stores a lifecycle-based traceability hierarchical storage program, which, when executed by a processor, implements the steps of the lifecycle-based traceability hierarchical storage method as described in any one of claims 1-7.
