Sample review upload receiving link closure determination method based on deep learning

Abstract

The application relates to the technical field of sample processing link determination, in particular to a sample audit uploading receiving link closure determination method based on deep learning. The method obtains sample processing event records formed in the audit, packaging uploading and provincial artificial intelligence platform receiving backfilling processes of a centralized control station, constructs a sample processing dynamic graph containing a sample version node and a processing event association relationship, writes a processing event identifier and a last processing event identifier for the processing event, and in a historical event aggregation process, corrects event similarity and aggregation proportion according to the difference between the last processing event identifier and the processing event identifier of a candidate historical event, to obtain corrected historical event features. Then, the method combines processing stage undertaking relationship to output a closed, disconnected, wrong, jumped or unfilled link determination result. The method is suitable for sample link verification in the scenes of multiple versions, retransmission and backfilling delay.

Description

Technical Field

[0001] This invention relates to the field of sample processing link determination technology, and in particular to a deep learning-based method for determining the closure of the sample review upload and reception link. Background Technology

[0002] In the sample review, upload, and reception process, samples at the station level typically do not complete the entire process in one go. Instead, a review record is first created at the centralized control station, followed by a packaging and upload stage to generate a packaging batch or upload task. Finally, the provincial AI platform receives the data and backfills the processing results. Each stage leaves a business record in its corresponding system, but these records are often scattered across different logs, task tables, or receipt tables, and the source of fields and generation times are not entirely consistent. For samples with a regular single-transfer process, the processing can be roughly reconstructed using the sample number, upload task number, or time sequence. However, in actual business operations, the same sample may undergo re-review, re-packaging, failed re-transmission, or delayed backfilling. The same packaging batch, upload task, or platform receipt may also be associated with multiple sample versions. Therefore, link determination cannot rely solely on the existence of a certain status field; it is also necessary to determine whether the review, upload, and backfilling processes truly follow the same business processing chain.

[0003] Existing solutions similar to dynamic graph link analysis, such as the Chinese patent with publication number CN117972319A entitled "A Method and Apparatus for Call Chain Anomaly Detection Based on Dynamic Graph Self-Supervised Learning," are geared towards microservice call chain anomaly detection scenarios. They collect call chain data during operation, construct microservice objects as nodes, construct the call relationships between microservices as directed edges, and use call type and call time as edge attributes. Subsequently, the dynamic graph is perturbated and augmented, and a graph-level representation is obtained through dynamic graph attention mechanism, average pooling, and self-supervised node prediction. Then, the anomaly of a new call chain is determined based on this representation.

[0004] Another paper, "Temporal Graph Networks for Deep Learning on Dynamic Graphs," proposes a temporal graph network framework that represents dynamic graphs as sequences of events with temporal information. Node representations are updated through memory modules and graph operators, enabling the processing of relational data that changes over time. These approaches demonstrate that organizing continuously occurring events into dynamic graphs and performing feature learning is an effective method for handling complex temporal relationships.

[0005] However, regarding the specific business of sample review, uploading, and receiving, existing dynamic graph or call chain analysis solutions focus more on the adjacency relationships of nodes, edges, time, and structure themselves, and typically do not use the "previous processing event" declared in the business records as a constraint condition when aggregating within the model. Thus, when there are many sample versions, reused upload tasks, platform receipt delays, or failed retransmissions, the model is prone to treating historical events that are close in time, have the same objects, or are adjacent in graph structure as related events, even though these events may not actually be the previous events that the current processing event inherits from.

[0006] In other words, existing solutions can learn the temporal and structural relationships between events, but they lack constraints on the business chain relationship of "which existing event the current event follows" to directly enter the historical event aggregation process. Summary of the Invention

[0007] One objective of this invention is to propose a deep learning-based method for determining the closure of a sample review, upload, and receive link. The core technical problem this application aims to solve is: in the case of multiple versions of samples circulating, retransmissions and backfilling, and reuse of event-related objects at the station end, how to ensure that the review, encapsulation, upload, and backfilling events of the centralized control station and the provincial artificial intelligence platform are included in the same processing link according to their true declared relationship with the previous processing event, and thereby give a unique link closure status.

[0008] The deep learning-based sample review upload and receive link closure determination method according to embodiments of the present invention includes:

[0009] S1. Obtain the sample processing event records generated during the process of station-side sample review, packaging and uploading at the centralized control station and receiving and backfilling on the provincial artificial intelligence platform. The sample processing event records include the sample version, processing stage, event associated object, event time, and the previous processing event declared to be undertaken by the current processing event. The previous processing event declared to be undertaken by the current processing event includes the corresponding sample version, processing stage, and event associated object, thus obtaining the sample processing event sequence.

[0010] S2. Construct a dynamic graph of sample processing based on the sequence of sample processing events. The dynamic graph of sample processing includes sample version nodes and the relationship between processing events between sample version nodes.

[0011] S3. Based on the sample processing dynamic diagram, generate a processing event identifier and a previous processing event identifier for each processing event. The processing event identifier is formed by the sample version, processing stage, and event-related object. The previous processing event identifier is formed by the sample version, processing stage, and event-related object corresponding to the previous processing event declared by the current processing event. This results in a sample processing dynamic diagram with processing event identifiers and previous processing event identifiers.

[0012] S4. Based on the sample processing dynamic graph with processing event identifier and previous processing event identifier, execute the event-aware dynamic graph learning model to extract the features of the current processing event and the features of candidate historical events.

[0013] S5. In the historical event aggregation process of the event-aware dynamic graph learning model, based on the difference value between the identifier of the previous processed event of the current processed event and the processed event identifier of the candidate historical event, the corrected event similarity between the current processed event features and the candidate historical event features is calculated, and the candidate historical event features are aggregated based on the corrected event similarity to obtain the corrected historical event features.

[0014] S6. Determine the closed state of the sample review, upload, and receive link based on the corrected historical event characteristics, and output the closed state determination result of the link between sample review, encapsulation and upload, and platform reception and backfilling.

[0015] Optionally, S1 specifically refers to:

[0016] The system acquires sample processing event records generated during the process of station-side sample review, packaging and uploading at the centralized control station, and receiving and backfilling on the provincial artificial intelligence platform. It extracts the sample version, processing stage, event-related object, event time, and the previous processing event declared by the current processing event from each sample processing event record to form a sample processing event record containing unified fields.

[0017] Based on the processing stage, sample processing event records containing uniform fields are classified according to the processing stage, forming sample processing event records classified by processing stage;

[0018] Based on the sample version, the event associated object, and the previous processing event declared by the current processing event, the sample processing event records classified by processing stage are grouped according to the same station sample to form grouped sample processing event records.

[0019] The sample processing event records after grouping are sorted according to the relationship between event time and processing stage. If samples on the same station are subject to re-review, re-encapsulation, or failed retransmission, the sorting position is adjusted according to the sample version update order to obtain the sample processing event sequence.

[0020] Optionally, S2 specifically refers to:

[0021] Receive the sample processing event sequence, extract the sample version, processing stage and event-related object from each processing event in the sample processing event sequence, and group the processing events corresponding to the same sample version into the same sample version node to form an initial sample version node set;

[0022] Based on the order in which sample versions appear in the sample processing event sequence in the initial sample version node set, the phase advancement direction between sample version nodes is determined, forming a sample version node set with a phase advancement direction.

[0023] Based on the processing stages and associated objects in the sample processing event sequence, the processing event association relationship is constructed for the sample version node set with stage advancement direction, so that the processing event association relationship of the centralized control station's review, packaging and uploading and the provincial artificial intelligence platform's reception and backfilling is connected to the corresponding sample version node;

[0024] When multiple sample version nodes correspond to the same event-related object, the processing event association relationship of each sample version node is updated according to the sample version in the sample processing event sequence, forming a dynamic graph of sample processing without merging sample version nodes.

[0025] Optionally, S3 specifically refers to:

[0026] Receive the dynamic graph of sample processing, traverse the processing event relationships between sample version nodes, and read the sample version node corresponding to the starting point of the processing event relationship, the sample version node corresponding to the ending point of the processing event relationship, the processing stage and the event association object from each processing event relationship to form a set of processing events that need to be identified.

[0027] The sample version is determined by the sample version node corresponding to the start and end points of the processing event association relationship in the processing event set generated according to the requirements. The processing stage and event association object are bound to the corresponding sample version to form a processing event combination composed of sample version, processing stage and event association object.

[0028] According to the order of sample version, processing stage and event-related object in the sample review upload and receive link, the processing event combination generates a processing event identifier for each processing event, and maps the processing event identifier to the corresponding processing event in the processing event set that needs to generate the identifier;

[0029] Based on the processing event association relationship corresponding to the current processing event in the sample processing dynamic graph, read the previous processing event declared to be inherited by the current processing event;

[0030] Based on the sample version, processing stage, and event-related object of the previous processing event declared by the current processing event, generate the identifier of the previous processing event for the current processing event;

[0031] If there exists a processing event in the set of processing events that needs to generate an identifier that corresponds to the previous processing event declared by the current processing event, make the identifier of the previous processing event of the current processing event consistent with the processing event identifier of the corresponding processing event.

[0032] If there is no processing event in the set of processing events for which an identifier needs to be generated that corresponds to the previous processing event declared by the current processing event, the identifier of the previous processing event generated based on the sample version, processing stage and event associated object of the previous processing event declared by the current processing event is retained.

[0033] When multiple sample version nodes are connected to the same event-related object, or when the same sample version is re-approved, re-encapsulated, or retransmitted after failure, the previous processing event identifier of the current processing event is determined based on the sample version, processing stage, and event-related object of the previous processing event declared by the current processing event, thus forming a processing event with the processing event identifier and the previous processing event identifier.

[0034] The processing event relationships in the sample processing dynamic graph are updated based on the processing events with processing event identifiers and previous processing event identifiers, so that each processing event relationship carries the processing event identifier and previous processing event identifier of the corresponding processing event, resulting in a sample processing dynamic graph with processing event identifiers and previous processing event identifiers.

[0035] Optionally, S4 specifically refers to:

[0036] Receive a sample processing animation with a processing event identifier and a previous processing event identifier, determine the current processing event for which features are to be extracted, and read the sample version, processing stage, event associated object, event time, processing event identifier, and previous processing event identifier from the current processing event to form the current processing event encoding input;

[0037] The event-aware dynamic graph learning model is executed based on the current event encoding input. The sample version, processing stage, event associated object, event time, processing event identifier, and previous processing event identifier are encoded to obtain the current event features.

[0038] Based on the sample version, event associated object, and event time in the current event processing code input, the event processing dynamic graph with the processing event identifier and the previous processing event identifier is read to form a candidate historical event set. The event time is earlier than the current event processing time and contains the same sample version node or the same event associated object.

[0039] Based on the sample version, processing stage, associated object, event time, processing event identifier, and previous processing event identifier of the candidate historical events in the candidate historical event set, the candidate historical events are encoded using the same encoding method as the current processing event in the event-aware dynamic graph learning model, resulting in candidate historical event features that are in the same feature space as the current processing event features.

[0040] Optionally, S5 specifically refers to:

[0041] During the historical event aggregation process of the event-aware dynamic graph learning model, the features of the currently processed event, the features of the candidate historical events, the identifier of the previous processed event of the currently processed event, and the processing event identifiers of the candidate historical events are received. Each candidate historical event feature is bound to the corresponding processing event identifier of the candidate historical event to form a set of historical event features participating in the aggregation.

[0042] Based on the feature similarity calculation between the current event features and the candidate historical event features in the set of historical event features participating in the aggregation, the initial event similarity corresponding to the candidate historical event is obtained.

[0043] Based on the identifier of the previous processing event of the current processing event and the processing event identifier of the candidate historical events in the feature set of historical events participating in the aggregation, the identifier difference value is calculated for the sample version, processing stage and event associated object respectively to form the identifier difference value;

[0044] The identification difference value is embedded into the correction calculation of the initial event similarity. The similarity correction coefficient is determined based on the identification difference value. If the identification difference value is not greater than the preset threshold, the similarity correction coefficient is determined as the coefficient that keeps the similarity of the corresponding initial event unchanged. If the identification difference value is greater than the preset threshold, the similarity correction coefficient is determined as the coefficient that reduces the similarity of the corresponding initial event. The similarity of the corresponding initial event is corrected based on the similarity correction coefficient to obtain the corrected event similarity.

[0045] The aggregation ratio of each candidate historical event feature in the set of historical event features participating in the aggregation is determined based on the corrected event similarity, and the candidate historical event features are updated according to the aggregation ratio to form the candidate historical event features updated according to the aggregation ratio.

[0046] Historical event aggregation is performed based on the updated candidate historical event features according to the aggregation ratio to obtain the corrected historical event features.

[0047] Optionally, S6 specifically refers to:

[0048] Receive the corrected historical event features, and determine the processing stage connection between sample review and encapsulation upload, as well as the processing stage connection between encapsulation upload and platform reception and feedback, based on the aggregation ratio of candidate historical event features used when forming the corrected historical event features and the processing stage of candidate historical events.

[0049] Based on the processing stage succession relationship and the corresponding situation of the previous processing event declared to be succeeded by the current processing event in the candidate historical events in the corrected historical event characteristics, link closure determination information is formed.

[0050] The link closure determination information is used to determine the closure status of the sample review upload and receive link. The closure status of the sample review upload and receive link includes closed, broken link, wrong link, skipped link, or not filled back.

[0051] Based on the closed status of the sample review, upload, and receive link, output the link closure determination result between sample review, encapsulation upload, and platform reception feedback.

[0052] Optionally, when generating the previous processing event identifier for the current processing event, if the sample processing dynamic graph contains multiple sample version nodes connected to the same event-related object, or if the same sample version corresponds to re-approval, re-encapsulation, or failure retransmission, then based on the previous processing event declared by the current processing event, the processing event corresponding to the processing stage of the previous processing event declared by the current processing event is determined in the sample processing dynamic graph. The sample version and event-related object are then matched again within the processing events corresponding to the processing stage, without merging multiple processing events corresponding to the same event-related object into the same previous processing event identifier. When a match is found... When the sample version, processing stage, and event associated object of the current processing event all correspond to the previous processing event declared to be handled by the current processing event, the processing event identifier of the matched processing event is used as the previous processing event identifier of the current processing event. When no processing event is matched that corresponds to the previous processing event declared to be handled by the current processing event, the previous processing event identifier generated based on the sample version, processing stage, and event associated object of the previous processing event declared to be handled by the current processing event is retained, and the previous processing event identifier of the current processing event is written into the corresponding processing event association relationship.

[0053] Optionally, when the identifier difference value is used in the correction calculation of the initial event similarity, the processing stage in the identifier of the previous processing event of the current processing event is first matched with the processing stage in the identifier of the processing event of the candidate historical event. If the processing stages do not correspond, the initial event similarity of the candidate historical event is reduced according to the similarity correction coefficient corresponding to the identifier difference value, and the features of the candidate historical event are retained for historical event aggregation. If the processing stages correspond, the sample version and event associated object in the identifier of the previous processing event of the current processing event are matched with the sample version and event associated object in the identifier of the processing event of the candidate historical event, respectively, and the identifier difference value is updated according to the matching results. When the sample version, processing stage, and event associated object all correspond, the initial event similarity of the candidate historical event is retained. When the sample version, processing stage, or event associated object does not correspond, the initial event similarity of the candidate historical event is reduced according to the similarity correction coefficient corresponding to the updated identifier difference value, and the corrected event similarity used to determine the aggregation ratio is obtained.

[0054] Optionally, when determining the link closure judgment result between sample review, package upload, and platform reception feedback, the link closure judgment result is determined according to the judgment order of closure, no feedback, skipped link, wrong link, and broken link. If the link closure judgment information indicates that there is a continuous connection between sample review and package upload, and between package upload and platform reception feedback, then the link is closed. If the link closure judgment information indicates that there is a continuous connection between sample review and package upload, and there is no processing event corresponding to platform reception feedback that forms a continuous connection with package upload, then the link is not feedback. If the processing stage connection relationship indicates that the processing stage does not form a continuous connection according to sample review, package upload, and platform reception feedback, and there is a processing event corresponding to platform reception feedback, then the link is skipped. If the link closure judgment information indicates that the candidate historical event participates in historical event aggregation but the sample version or event associated object does not correspond to the previous processing event declared to be received by the current processing event, then the link is wrong. If the link closure judgment information indicates that the previous processing event declared to be received by the current processing event does not form a corresponding relationship in the candidate historical events, then the link is broken.

[0055] The beneficial effects of this invention are:

[0056] (1) The main improvement of this invention is that it no longer relies solely on the temporal proximity, node connection, or common feature similarity between events in the dynamic graph to aggregate historical events. Instead, it transforms the previous processed event declared as being inherited by the current processed event into a previous processed event identifier, and calculates the difference between this identifier and the processed event identifier of the candidate historical events during the historical event aggregation process. This difference is not used as an additional rule for filtering after the judgment is completed, but directly participates in the correction of the initial event similarity and further affects the proportion of candidate historical events in the aggregation. As a result, historical events that are truly inherited by the current event can maintain a high contribution in the aggregation process, while records that are only close in time and similar in object but are not the real previous events will have a reduced impact, making the historical event features formed by the model closer to the actual inheritance relationship in the sample processing link.

[0057] (2) This invention also reorganizes the dynamic graph structure around the sample version as a business object. For situations such as re-review, re-packaging, and retransmission after failure, this application uses the sample version as a node to retain different processing states, rather than simply merging nodes based on the station-side sample identifier or event-related object; for situations where multiple sample versions are associated with the same packaging batch, the same upload task, or the same platform receipt, the corresponding processing event relationships are also retained respectively. The processing event identifier generated in this way simultaneously includes the sample version, processing stage, and event-related object, and the previous processing event identifier also uses the same field dimension to express it, so that each processing event edge has a verifiable receiving direction. This design can reduce the link mixing caused by batch uploads, task reuse, or retransmission backfilling, so that subsequent model aggregation and link determination are based on clearer in-graph identifiers.

[0058] (3) Further, the link closure determination of the present invention is not simply a matter of whether the review, upload, and backfill records have all appeared, but rather a determination of the final state by combining the corrected historical event characteristics, the aggregation ratio of candidate historical events, and the succession relationship of adjacent processing stages. For the encapsulated upload event, it is necessary to confirm whether its previous processing event corresponds to the review of the central control station; for the platform receiving backfill event, it is necessary to confirm whether its previous processing event corresponds to the encapsulated upload, and to determine whether the succession is valid by combining the contribution ratio of candidate historical events in the aggregation. Therefore, this method can distinguish different states such as closure, no backfill, skipped links, wrong links, and broken links within the same determination range, so that the output results reflect both whether the business record exists and whether a real and continuous sample processing link has been formed between the records. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 This is a flowchart of a deep learning-based sample review upload and reception link closure determination method proposed in this invention;

[0061] Figure 2 This is a network architecture diagram of a deep learning-based sample review upload and reception link closure determination method proposed in this invention. Detailed Implementation

[0062] In Example 1, reference Figures 1 to 2 A deep learning-based method for determining the closure of the sample review upload and receive link, comprising:

[0063] S1. Obtain the sample processing event records generated during the process of station-side sample review, packaging and uploading at the centralized control station and receiving and backfilling on the provincial artificial intelligence platform. The sample processing event records include the sample version, processing stage, event associated object, event time, and the previous processing event declared to be undertaken by the current processing event. The previous processing event declared to be undertaken by the current processing event includes the corresponding sample version, processing stage, and event associated object, thus obtaining the sample processing event sequence.

[0064] S2. Construct a dynamic graph of sample processing based on the sequence of sample processing events. The dynamic graph of sample processing includes sample version nodes and the relationship between processing events between sample version nodes.

[0065] S3. Based on the sample processing dynamic diagram, generate a processing event identifier and a previous processing event identifier for each processing event. The processing event identifier is formed by the sample version, processing stage, and event-related object. The previous processing event identifier is formed by the sample version, processing stage, and event-related object corresponding to the previous processing event declared by the current processing event. This results in a sample processing dynamic diagram with processing event identifiers and previous processing event identifiers.

[0066] S4. Based on the sample processing dynamic graph with processing event identifier and previous processing event identifier, execute the event-aware dynamic graph learning model to extract the features of the current processing event and the features of candidate historical events.

[0067] S5. In the historical event aggregation process of the event-aware dynamic graph learning model, based on the difference value between the identifier of the previous processed event of the current processed event and the processed event identifier of the candidate historical event, the corrected event similarity between the current processed event features and the candidate historical event features is calculated, and the candidate historical event features are aggregated based on the corrected event similarity to obtain the corrected historical event features.

[0068] S6. Determine the closed state of the sample review, upload, and receive link based on the corrected historical event characteristics, and output the closed state determination result of the link between sample review, encapsulation and upload, and platform reception and backfilling.

[0069] In this embodiment, step S1 specifically includes:

[0070] The original business records generated during the process of station-side samples being reviewed, packaged, uploaded, and received and backfilled by the provincial artificial intelligence platform are recorded as follows: , Indicates the number of original business records. This represents the index of the original business records. Indicates the first The original business records are from the central control station's audit log, the encapsulation and upload log, and the provincial artificial intelligence platform's receive and feedback records. The central control station's audit log provides at least the audit record, sample version, audit time, and audit declaration recipient. The encapsulation and upload log provides at least the encapsulation batch, upload task, sample version, encapsulation and upload time, and encapsulation and upload declaration recipient. The provincial artificial intelligence platform's receive and feedback records provide at least the platform receipt, upload task, sample version, receive and feedback time, and receive and feedback declaration recipient.

[0071] For each original business record Extract sample version Processing stage Event-related objects Event Time The previous event that the current event is handled by. Sample version , Indicates the station-side sample identifier. This indicates the sample version update order. If the original business record stores the sample version as a sample version string, then the site-side sample identifier is parsed from the sample version string. and sample version update order If the sample version string includes a version generation time, then the sample version update order is determined according to the version generation time. The processing stage set is denoted as Processing stage From the processing stage set Values ​​are retrieved from the event-related object. Used to represent audit records, batch packaging, uploaded tasks, or platform receipts, and event times. The original business records are used to determine the time of the business transaction and are uniformly converted to the same time base.

[0072] The current event being processed is the one that the previous event is handled by. , This indicates the sample version corresponding to the previous processed event declared in the current event processing statement. This indicates the processing stage corresponding to the previous processing event declared in the current event processing declaration. This indicates the event-associated object of the previous event declared as being handled by the current event handling declaration. The declaration acceptance field from the original business records is not inferred post-hoc by the event-aware dynamic graph learning model. For encapsulated upload events, the declaration acceptance field points to the sample version, processing stage, and event-related object corresponding to the central control station's audit record. For provincial AI platform receiving and backfilling events, the declaration acceptance field points to the sample version, processing stage, and event-related object corresponding to the encapsulated upload record. For central control station audit events at the beginning of the link, if the central control station audit log declares acceptance of the station-side collected event, then... The value is collected at the station. The value is the identifier of the station-side data collection record. If the central control station's audit log does not declare that it will accept the station-side data collection event, then the initial previous processing event is set. , Indicates the preset initial processing stage marker. This indicates the preset initial event associated object marker. The preset initial processing stage marker is only used to form the previous processing event that the current processing event declaration inherits from, and does not participate in the processing stage set. Classification of processing stages;

[0073] After the field extraction is completed, a sample processing event record containing the unified fields is generated. And obtain sample processing event records. Record sample processing events By processing stage Classify the processing stages and then... Sample processing event records are categorized into ,Will Sample processing event records are categorized into ,Will Sample processing event records are categorized into During the processing stage, sample versions are classified but not deleted. Event-related objects Event Time The previous event that the current event is handled by. ;

[0074] After classifying the sample processing event records by processing stage, group the samples at the same station end, and assign each sample processing event record a group. Construct group keys , This represents the grouping key formed by the station-side sample identifier, the event-related object, and the previous processed event declared in the current event declaration. The value for any grouping key is... Forming a record of sample processing events after grouping. Sample version Station-side sample identifier Used to identify samples from the same station, sample version Sample version update order Used for sorting adjustments in scenarios involving re-auditing, repackaging, or retransmission after failure;

[0075] For each The sample processing event records within are sorted, with the sorting rule based on event time. Ascending order, processing stage progression, and sample version update order Ascending order and raw business record indexes In ascending order, the processing stages proceed as follows: centralized control station review precedes packaging and uploading, and packaging and uploading precedes provincial AI platform reception and feedback. If multiple sample processing event records have the same station-side sample identifier, and the same sample undergoes re-review, re-packaging, or failed retransmission,... And the order of sample version updates If they are different, the sample processing event records corresponding to the update order of the smaller sample version will be arranged before the sample processing event records corresponding to the update order of the larger sample version. If the sample version update order is the same, the order will continue to be stable according to the event time, the relationship of the processing stage and the original business record index.

[0076] Sort each Merge records according to the ascending order of event time of the first sample processed event record in each group. If the event time of the first sample processed event record in each group is the same, merge them according to the station-end sample identifier. and original business record index Merge in ascending order to form a sequence of sample processing events. , Indicates the number of events processed in the sample. Indicates the sequence index of sample processing events. Indicates the first Each processing event, the input to the sample processing dynamic graph construction layer is an ordered sequence of sample processing events. Each event processing For a five-field record, After expansion, each event is processed. It contains seven basic fields: sample version, processing stage, event associated object, event time, sample version corresponding to the previous processed event, processing stage corresponding to the previous processed event, and event associated object corresponding to the previous processed event, representing the sample processing event sequence. The event time and the previous processing event declared by the current processing event are retained, enabling the sample processing dynamic graph construction layer to build the sample processing dynamic graph based on the sample version, processing stage, and event-related objects.

[0077] In this embodiment, step S2 specifically includes:

[0078] The sample processing dynamic graph construction layer receives the sample processing event sequence. , Indicates the number of events processed in the sample. Indicates the sequence index of sample processing events. Indicates the first One event processing, sample version , Indicates the station-side sample identifier. Indicates the order of sample version updates and processing stages. From the processing stage set Values ​​are retrieved from the event-related object. Used to represent audit records, batch packaging, uploaded tasks, or platform receipts, and event times. Indicates the time when the event occurred, and the previous event that the current event declaration inherits from. , This indicates the sample version corresponding to the previous processed event. This indicates the processing stage corresponding to the previous event. This represents the event-associated object corresponding to the previously processed event;

[0079] The dynamic graph of sample processing is denoted as Sample processing animation Includes sample version node set Set of directions for phased advancement and the set of relationships related to event handling The sample version node is used to represent the different processing statuses of samples from the same site after re-review, re-encapsulation, or retransmission upon failure. Therefore, the sample processing dynamic graph construction layer is based on the sample version. Establish nodes instead of directly identifying them using station-side samples. Merging nodes and processing event relationships are represented by edges in the graph, not ordinary business field relationships. Each processing event relationship connects the starting sample version node and the ending sample version node, and carries the processing stage, the associated event object, the event time, and the previous processing event declared by the current processing event. This is part of the dynamic graph of sample processing. Updated in ascending order of the sample processing event sequence index, when the centralized control station reviews, packages and uploads, and the provincial artificial intelligence platform receives and backfills events, the corresponding sample version node and the association relationship of the processing event are read, and synchronously written into the sample processing dynamic graph. ;

[0080] Sample processing event sequence Perform a sequential scan and read each processing event. Sample version Processing stage and event-related objects The sequence of sample processing events The different sample versions appearing in the sample version set constitute the sample version set. For any sample version Establish sample version nodes , Indicates sample version The corresponding node will satisfy Event handling Write to sample version node event list , Indicates belonging to the sample version node The set of events to be processed is formed by all sample version nodes, creating the initial set of sample version nodes. The centralized control station review events, packaged upload events, and provincial artificial intelligence platform receive and fill events corresponding to the same sample version are grouped into the same sample version node, while the processing events corresponding to different sample versions are grouped into different sample version nodes respectively.

[0081] To determine the direction of progress between sample version nodes, the sample processing dynamic graph construction layer records each sample version. In the sample processing event sequence First appearance index , Indicates satisfaction The minimum sample processing event sequence index, for samples with the same station-end identifier. For multiple sample version nodes, first sort by the index of first occurrence. Sort in ascending order; if the first occurrence of the same index, sort according to the sample version update order. Sort in ascending order, with each sample version node pointing to a subsequent sample version node, forming a set of stage advancement directions. Set of directions for phased advancement Used to represent the processing order of samples from the same station across different sample versions, without changing the attribution relationship between sample version nodes and processing events; initial set of sample version nodes. Write the set of stages for advancement Subsequently, a set of sample version nodes with phased advancement directions is formed. ;

[0082] For each processed event Establish event handling relationships , Indicates that the event is handled The generated graph edges handle event relationships. The endpoint sample version node is set to If the sample identifier is from the same station The following arrangement exists in Previously and with For adjacent sample version nodes, set the adjacent sample version nodes as event associations. The starting sample version node, if it does not exist, is arranged in Previous sample version nodes will Also set to handle event association relationships The starting and ending sample version nodes are used to handle event relationships. Simultaneously written to the processing stage Event-related objects Event Time The previous event that the current event is handled by. ,when When reviewing data at the centralized control station, handle event correlations. Connect to the sample version node corresponding to the audit event at the central control station, when To handle event relationships during the upload process. Connect to the sample version node corresponding to the encapsulated upload event, when When the provincial artificial intelligence platform receives feedback, it processes event correlations. Connect to the provincial artificial intelligence platform to receive the sample version node corresponding to the backfill event;

[0083] All event processing relationships are combined into a set of event processing relationships. For any event-related object Construct the set of sample version nodes corresponding to the event-associated object. ,when When there are at least two sample version nodes, it indicates that multiple sample version nodes are connected by the same review record, the same packaging batch, the same upload task, or the same platform receipt. The sample processing dynamic graph construction layer is based on each processing event. Sample version Update the sample version node respectively List of Event Relationships Do not associate objects with the same event Multiple sample version nodes are merged into one node, and objects with the same event are not associated. Multiple corresponding processing event relationships are merged into a single edge. The processing method for sample version nodes that are not merged is used to preserve sample version differences in scenarios such as batch packaging, batch uploading, and platform receipt reuse, and to avoid the processing event relationships between different sample version nodes being written into the same link due to the same event-related object. Sample processing dynamic graph It consists of sample version nodes, phase advancement direction, and processing event relationships, forming a set of processing event relationships. Each edge in the process retains the processing stage, the associated object of the event, the event time, and the previous processing event declared by the current processing event declaration.

[0084] In this embodiment, step S3 specifically includes:

[0085] Dynamic diagram of sample processing received by the event identifier generation layer , Represents the set of sample version nodes. This represents the set of directions for advancement in a given stage. This represents a set of relationships between events being processed. Indicates the first Process the relationship between events. Indicates the number of event relationships processed. This represents the index of event processing relationships, with each event processing relationship being... Carrying the starting sample version node Endpoint Sample Version Node Processing stage Event-related objects Event Time The previous event that the current event is handled by. Starting point sample version node Indicates the handling of event relationships Source node, endpoint sample version node Indicates the handling of event relationships The arriving node, the currently processed event declares the previous processed event. , This indicates the sample version corresponding to the previous processed event. This indicates the processing stage corresponding to the previous event. This represents the event-associated object corresponding to the previously processed event;

[0086] The event identifier generation layer traverses the set of event associations. From each processing event relationship Read the starting sample version node Endpoint Sample Version Node Processing stage Event-related objects Event Time The previous event that the current event is handled by. This forms a set of processing events that require the generation of identifiers. A set of events that need to be processed and identified needs to be generated. From the dynamic diagram of sample processing The set of processed events retrieved from the event association relationship in the processing event set. Event handling in Relationship with handling events One-to-one correspondence, event handling Including the starting sample version node Endpoint Sample Version Node Processing stage Event-related objects Event Time The previous event that the current event is handled by. Processing event sets It is only used to generate the processing event identifier and the previous processing event identifier, and does not merge the processing event association relationships;

[0087] For any processed event The event identifier generation layer processes events based on the starting sample version node. and endpoint sample version node Determine the sample version When the starting sample version node With the endpoint sample version node When consistent, sample version Take the sample version corresponding to the sample version node, when the starting sample version node With the endpoint sample version node In case of inconsistency, sample version Get the endpoint sample version node The corresponding sample version, starting sample version node Used to confirm the handling of events At the current stage of the process, the event identifier generation layer will handle the stage. and event-related objects Binding to sample version Forming a combination of events to handle Handling event combinations Not an independent algorithm object, but a combination of events. The identifier generation input consists of sample version, processing stage, and event-related objects;

[0088] The event identifier generation layer generates event identifiers according to the positional order in the sample review upload-receive link. The positional order is represented by a fixed stage sequence number, with the central control station review corresponding to the stage sequence number. Encapsulate and upload the corresponding stage sequence. The provincial artificial intelligence platform receives and backfills the corresponding stage sequence. Combine the processing events Processing stage Convert to the corresponding stage sequence Then, write the event identifier according to the sample version, stage sequence, and event-related object. , Indicates handling events Event handling identifiers, Indicates processing stage Corresponding stage sequence, event identifier The event identifier is generated by combining three fields: sample version, processing stage, and event associated object. When stored as a character identifier, the event identifier generation layer concatenates the sample version field, stage sequence field, and event associated object field in a fixed order. When stored as a numeric identifier, the event identifier generation layer uses a one-to-one mapping table to map the same combination of three fields to the same numeric identifier, and different combinations of three fields to different numeric identifiers. Mapped to the event handling set Event handling in And write the event association relationship. ;

[0089] For the current event being processed The event identifier generation layer processes the corresponding event association relationships. Read the previous event that the currently processed event declares to be processed. The event identifier generation layer will handle the processing stage corresponding to the previous event. Convert to the corresponding stage sequence ,when When reviewing the centralized control station, Pick ,when For encapsulated upload, Pick ,when When the provincial artificial intelligence platform receives the feedback, Pick ,when When setting the initial processing stage marker or collecting data at the station, Pick The event identifier generation layer generates the previous event identifier according to the sample version, stage sequence, and associated object of the previous event. , Indicates the currently processed event The previous event identifier, the previous event identifier Used to express the previous processing event declared by the current processing event declaration, it is not required that the previous processing event declared by the current processing event declaration already exists in the sample processing dynamic graph. middle;

[0090] To reliably determine the previous processing event identifier when there are re-approval, re-packaging, failed retransmission, multiple sample version nodes connected to the same packaging batch, or multiple sample version nodes connected to the same upload task, the processing event identifier generation layer constructs a precise corresponding index set based on the previous processing event declared by the current processing event. :

[0091] ;

[0092] In the formula, Indicates the currently processed event The declaration inherits the previous processed event in the processed event set. This forms a set of candidate indices that correspond exactly to each other. Indicates candidate processing events index, Represents a set of events to be processed. Number of events processed in China Indicates the currently processed event index, This indicates that the candidate event index is different from the current event index. Indicates candidate processing events Sample version, This indicates the sample version corresponding to the previous processed event declared in the current event processing statement. Indicates candidate processing events The stage sequence corresponding to the processing stage. This indicates the stage sequence number corresponding to the previous event that the current event declaration is attached to. Indicates candidate processing events The event-related object, This indicates the event-related object that the current event declaration inherits from the previous event. Indicates candidate processing events The time of the event Indicates the currently processed event The time of the event Represents a logical OR relation. This indicates a logical AND relationship;

[0093] When the exact corresponding index set When at least one candidate index is included, the event identifier generation layer processes the data from... Select event time The largest candidate processing event is selected. If multiple candidate processing events have the same event time, the sample processing event sequence index is chosen. The largest candidate event to process is read by the event identifier generation layer. Event handling identifier and the currently processed event The previous event identifier Updated to When the exact corresponding index set When empty, the event identifier generation layer retains the information based on... , and The generated previous processed event identifier The identifier of the previous processed event is retained. This indicates the business object that the current event processing declaration undertakes, enabling the event processing identifier generation layer to still generate identifiers even when the previous event processing is missing, and enabling the historical event aggregation layer to calculate the identifier difference value between the identifier of the previous event processing of the current event processing and the event processing identifier of the candidate historical events.

[0094] When the sample processing animation When multiple sample version nodes are connected to the same event-related object, the event identifier generation layer constructs the event-related object. Corresponding sample version node set Event-related objects Corresponding sample version node set This includes handling event associations by carrying event-related objects. All sample version nodes, if If there are at least two sample version nodes, then for each currently processed event, the previous processed event declared as being handled by the current processed event is used. Determine the identifier of the previous processed event. Do not associate the same event with other objects. Multiple corresponding processing events are merged into the same previous processing event identifier. The processing event identifier generation layer first matches the processing stage. The sample version continues to be matched in the corresponding processing event during the processing phase. and event-related objects When the matched processing event is consistent with the sample version, processing stage, and event associated object... When the matching event is matched, the event identifier of the matching event is used as the previous event identifier of the current event. If no matching event is found that meets the above conditions, the previous event identifier generated based on the previous event declared by the current event is retained.

[0095] When there is a re-approval, re-encapsulation, or failed retransmission within the same sample version node, the event identifier generation layer still uses the previous processed event declared by the current processed event. To accurately identify the identifier of the previous processed event When there are more than two centralized control stations reviewing and processing events within the same sample version node, they are re-reviewed according to the event association objects corresponding to the review records. When there are more than two encapsulated upload processing events within the same sample version node, they are re-encapsulated or re-transmitted if they fail, according to the event association objects corresponding to the encapsulation batch or upload task. When there are more than two provincial artificial intelligence platforms receiving backfill processing events within the same sample version node, they are received backfill events according to the event association objects corresponding to the platform receipts. The event identifier generation layer does not select the previous processing event separately based on the event time, nor does it merge the previous processing event identifiers based on the same event association objects.

[0096] The event identifier generation layer will process the event identifiers. and the previous event identifier Write the corresponding event handling association. This forms a relationship between processed events, which includes the processed event identifier and the identifier of the previous processed event. , Indicates the relationship between the events being handled. The updated event relationship relationships, once all event relationship relationships have been written, form a set of event relationship relationships. , This represents the set of processing event associations carrying the processing event identifier and the previous processing event identifier; the sample version node set. and set of phased advancement directions The sample processing animation remains unchanged, but is updated to include the processing event identifier and the previous processing event identifier. Handling event identifiers and the previous event identifier As an in-graph identifier field, it enters the event feature encoding layer and the historical event aggregation layer.

[0097] In this embodiment, step S4 specifically includes:

[0098] The event-aware dynamic graph learning model runs on a sample processing dynamic graph with a processing event identifier and a previous processing event identifier. The event-aware dynamic graph learning model includes at least an event feature encoding layer, a candidate historical event reading layer, and a historical event aggregation layer. The event feature encoding layer is used to encode the same field for the current processing event and the candidate historical events. The candidate historical event reading layer is used to read processing events from the sample processing dynamic graph whose event time is earlier than the current processing event and contains the same sample version node or the same event-related object. The historical event aggregation layer reads the features of the current processing event, the features of the candidate historical events, the identifier of the previous processing event of the current processing event, and the processing event identifier of the candidate historical events.

[0099] The event feature encoding layer and the candidate historical event reading layer receive a sample processing dynamic graph with the processed event identifier and the previous processed event identifier. , Represents the set of sample version nodes. This represents the set of directions for advancement in a given stage. This represents the set of processing event associations, which includes both the processing event identifier and the identifier of the previous processing event. The event processing relationships in the process are denoted as , Indicates the number of event relationships processed. This represents the index of event processing relationships, with each event processing relationship being... Carrying the starting sample version node Endpoint Sample Version Node Sample version Processing stage Event-related objects Event Time Event handling identifier and the previous event identifier Starting point sample version node Indicates the handling of event relationships Source node, endpoint sample version node Indicates the handling of event relationships The arrival node;

[0100] The current processing event index of the feature to be extracted is denoted as , Current Event Index As specified by the processing event to be determined, in the dynamic graph of sample processing. When performing batch feature extraction on all processed events, according to The current processing events are determined sequentially, and the event feature encoding layer determines the relationship between the current processing event and the processing events. Read sample version Processing stage Event-related objects Event Time Event handling identifier and the previous event identifier This forms the input code for the current event being processed. Current event code input The record consists of six fields, each derived from a sample processing animation graph that includes a processing event identifier and the identifier of the previous processing event. ;

[0101] The event feature encoding layer encodes the currently processed event as input. Convert to 128-dimensional sample processing event vector Sample version features 32-dimensional, sample version , Indicates the station-side sample identifier. Indicates the order of sample version updates, and the site-side sample identifier. Input a 16-dimensional sample identifier embedding table, sample version update order Input a 16-dimensional version of the sequential embedding table and process stage features. 16-dimensional, processing stage According to the stage sequence input of the centralized control station's review, encapsulation and upload, and the provincial artificial intelligence platform's reception and feedback, input the 16-dimensional processing stage embedding table, and the event-related object characteristics. 32-dimensional, event-related object , Indicates the type of the event-associated object. This indicates the identifier of the event-related object. The types of event-related objects include review records, batch packaging, upload tasks, and platform receipts. Input a 16-dimensional object type embedding table, and an event-associated object identifier. Input a 16-dimensional object identifier embedding table, event time features It has 16 dimensions, of which 8 dimensions are derived from event time. The time bucket numbers, formed according to a unified time base and a 60-second time granularity, are obtained through a time bucket embedding table. The 8-dimensional element is obtained from the index position of the current processed event in the sample processed event sequence through a sequence position embedding table, and the processed event identification feature is also included. It is 16-dimensional, consisting of event identifiers. The previous processing event identifier feature is obtained from the input processing event identifier embedding table. It is 16-dimensional, identified by the previous processed event. The previous processed event identifier is embedded in the input table.

[0102] The event feature encoding layer is denoted as , The model parameters represent the event feature encoding layer. The event feature encoding layer consists of a field embedding table, first-layer fully connected weights, first-layer biases, second-layer fully connected weights, and second-layer biases. The field embedding table uses discrete field indices as keys and fixed-dimensional vectors as values. The field embedding table is updated during the training of the event-aware dynamic graph learning model and is directly looked up by field index during feature extraction. The event feature encoding layer comprises two fully connected layers. The first fully connected layer has 128 neurons that receive 128-dimensional samples and process event vectors. The first layer performs linear transformations, bias superposition, and nonlinear mapping to generate 128-dimensional intermediate event features. The second layer, a fully connected structure, consists of 128 neurons that receive these 128-dimensional intermediate event features, perform linear transformations and bias superposition, and generate 128-dimensional features for the currently processed event. Current event characteristics Input with the currently processed event encoding The six fields correspond to the records;

[0103] The candidate historical event reading layer inputs the event code of the currently processed event. Sample version in Event-related objects and event time In the sample processing animation with the processing event identifier and the previous processing event identifier. Read candidate historical events and denote the candidate historical event index set as follows: :

[0104] ;

[0105] In the formula, This represents the set of candidate historical event indices corresponding to the currently processed event. Indicates the index of candidate processing events. Indicates the number of event relationships processed. Indicates the event time of the candidate processing event. Indicates the event time at which the current event is being processed. The starting sample version node represents the association between candidate processing events and their corresponding processing events. The endpoint sample version node represents the association between candidate processing events and their corresponding processing events. Indicates sample version Corresponding sample version node, sample version node From the sample version node set By sample version Direct reading, The event associated object representing the candidate event to be processed. This represents the event-associated object currently handling the event. Indicates a logical OR relationship;

[0106] The candidate historical event set consists of the candidate historical event index set. The processing event association is composed of the following: the processing events in the candidate historical event set all satisfy the condition that the event time is earlier than the event time of the current processing event, and contain the same sample version node or the same event association object. The candidate historical event set may contain the actual previous processing event declared by the current processing event, or it may contain easily confused events formed by the same packaging batch, the same upload task, or platform receipt delay. The candidate historical event reading layer only determines the candidate range that can participate in the historical event aggregation, and does not determine whether the candidate historical event is the previous processing event declared by the current processing event in the candidate historical event reading layer.

[0107] For the candidate historical event index set Index of each candidate processing event in The candidate historical event reading layer handles event relationships. Read sample version Processing stage Event-related objects Event Time Event handling identifier and the previous event identifier This forms the input of candidate historical event codes. The event feature encoding layer uses the same field embedding table as the currently processed event, a first-layer fully connected structure, and a second-layer fully connected structure to encode and input candidate historical events. Convert to 128-dimensional sample processing event vector And generate 128-dimensional candidate historical event features. The set of features for candidate historical events is denoted as , This represents all candidate historical event features corresponding to the currently processed event, and the features of the currently processed event. and candidate historical event characteristics All are composed of event feature encoding layers The generated features are all 128-dimensional, and the current event features are being processed. and candidate historical event characteristics Event similarity calculations can be performed at the historical event aggregation layer.

[0108] In this embodiment, step S5 specifically includes:

[0109] The historical event aggregation layer operates within the existing historical event aggregation process of the event-aware dynamic graph learning model. It is not an independent filtering module added outside the event-aware dynamic graph learning model; rather, it receives features of the currently processed event. Candidate historical event feature set The identifier of the previous processed event for the currently processed event. and event identifiers for handling candidate historical events , Indicates the index of the currently processed event. Indicates the candidate historical event index. This represents the set of candidate historical event indices corresponding to the currently processed event, and the characteristics of the currently processed event. and candidate historical event characteristics All are 128-dimensional event features, generated by the same encoding method in the event feature encoding layer, and represent the identifier of the previous processed event of the currently processed event. , This indicates the sample version corresponding to the previous processed event declared in the current event processing statement. This indicates the stage sequence number corresponding to the previous event that the current event declaration is attached to. This indicates the event associated object corresponding to the previous processed event declared in the current processed event declaration, and is the processed event identifier of the candidate historical event. , This represents a sample version of a candidate historical event. This indicates the stage sequence corresponding to the processing stage of a candidate historical event. The event-related object representing the candidate historical event;

[0110] The historical event aggregation layer will aggregate the features of each candidate historical event. Event identifiers for corresponding candidate historical events Binding, forming a set of historical event features participating in the aggregation. The set of historical event features participating in the aggregation Preserve candidate historical event index This makes the candidate historical event characteristics Event identifiers for handling candidate historical events The processing stage of candidate historical events is read using the same index. The historical event aggregation layer does not delete candidate historical event features that are inconsistent in processing stage, sample version, or event associated object during the binding stage. Candidate historical event features that are inconsistent in processing stage, sample version, or event associated object still enter the historical event aggregation process.

[0111] To ensure that the identifier difference value uses the underlying numerical value in the calculation, the historical event aggregation layer will include sample versions. and sample version Convert to integer encoding according to the sample version encoding table and Associate the event with the object and event-related objects Convert to integer encoding according to the event-related object encoding table and The sample version encoding table and the event-related object encoding table are generated using a one-to-one mapping method. The same field value is mapped to the same integer code, and different field values ​​are mapped to different integer codes, with stage sequence... and stage sequence It's already an integer sequence number; the central control station will verify the corresponding information. Encapsulate and upload the corresponding The provincial artificial intelligence platform receives and returns the corresponding information. Preset initial processing stage markers or corresponding station-side data collection. ;

[0112] When the candidate historical event index set When not empty, the historical event aggregation layer completes the generation of initial event similarity, identifier difference value, similarity correction coefficient, corrected event similarity, aggregation ratio, candidate historical event features updated according to the aggregation ratio, and corrected historical event features during the same historical event aggregation process:

[0113] ;

[0114] In the formula, Indicates candidate historical events The corresponding initial event similarity is the similarity obtained from the features of the current event being processed and the features of candidate historical events before the introduction of inheritance relationship constraints. Indicates the candidate historical event index. Indicates the index of the currently processed event. This represents the event feature dimension index, with a value range of [value range missing]. to , Indicates the characteristics of the currently processed event. The Dimensional value, Indicates the characteristics of candidate historical events The Dimensional value, This represents the denominator protection constant, in this embodiment , Indicates candidate historical events The corresponding identifier difference value is the identifier of the previous processed event for the currently processed event. Event identifiers for handling candidate historical events The differences between them Indicates the weight of differences in the processing stage. Indicates the weight of differences between sample versions. In this embodiment, the difference weights of the event-related objects are represented. , , , This indicates an indicator function; the value is taken when the condition within the parentheses is true. If the condition in parentheses is not met, then take... , This indicates the stage sequence corresponding to the processing stage of a candidate historical event. This indicates the stage sequence number corresponding to the previous event that the current event declaration is attached to. Integer encoding representing the sample version of a candidate historical event. This represents the integer encoding of the sample version corresponding to the previous processed event declared in the current processed event declaration. Integer encoding of the event association object representing the candidate historical event. This represents the integer code of the event-associated object that the current event declaration inherits from. Indicates candidate historical events The corresponding similarity correction coefficient, This represents a preset threshold, in this embodiment. , This represents the similarity decay coefficient. greater than and less than The dimensionless value, in this embodiment , Indicates candidate historical events The corresponding corrected event similarity, Indicates candidate historical events The corresponding polymerization ratio, Indicates an exponential mapping. Represents the candidate historical event index set any candidate historical event index in, This represents the set of candidate historical event indices corresponding to the currently processed event. Indicates candidate historical events The corresponding corrected event similarity, This represents the updated candidate historical event features based on the aggregation ratio. Indicates the currently processed event The corresponding revised characteristics of historical events, Indicate the characteristics of candidate historical events;

[0115] Identify the difference value The calculations are performed in the order of priority of the processing stages, with the historical event aggregation layer comparing them first. and In cases of inconsistency at different processing stages, identify the differences. The initial event similarity corresponding to the candidate historical event is determined by the indicator corresponding to the difference weight in the processing stage. Correction coefficient based on similarity Reduce, candidate historical event characteristics Retained in the set of historical event features participating in the aggregation In the process, assuming consistency across processing stages, the historical event aggregation layer continues to compare the integer encodings of sample versions. and And compare the integer encodings of the event-related objects. and Update the identifier difference value based on the sample version matching results and the event-related object matching results. ;

[0116] When the sample version, processing stage, and event-related object are all consistent, the difference value is identified. Not greater than the preset threshold Similarity correction coefficient Pick Initial event similarity The difference remains unchanged; however, when there are inconsistencies in sample version, processing stage, or event-related objects, the difference value is identified. Greater than the preset threshold Similarity correction coefficient Less than Initial event similarity The similarity of events is reduced and corrected. Corrected event similarity This is not a post-selection score after the historical event aggregation process is completed; it is a corrected event similarity score. Features directly used to determine candidate historical events during the historical event aggregation process. polymerization ratio ;

[0117] The historical event aggregation layer is based on the corrected event similarity. Candidate historical event index set The candidate historical events are normalized exponentially to obtain the aggregation ratio. polymerization ratio Determine the characteristics of candidate historical events The proportion of historical events in the aggregation layer, with the aggregation ratio as the basis. Update candidate historical event features This generates candidate historical event features updated according to the aggregation ratio. The historical event aggregation layer provides a set of candidate historical event indexes. All Perform element-wise summation to form the historical event characteristics after correction of 128. ;

[0118] When the candidate historical event index set When empty, the historical event aggregation layer will use the corrected historical event features. Set it to a 128-dimensional zero vector, and mark the previous processing event of the current processing event. Retained as a graph identifier field readable by the link closure determination layer, and corrected historical event features. Candidate historical event features updated according to aggregation ratio Aggregated and corrected historical event features And the aggregation ratio retained when forming the revised historical event characteristics. Upon entering the link closure determination layer, the link closure determination layer does not replace the corrected historical event features with the original log status fields. Revised characteristics of historical events This serves as the sole historical feature determination basis received by the link closure determination layer from the historical event aggregation process.

[0119] In this embodiment, step S6 specifically includes:

[0120] The link closure determination layer receives the corrected historical event features generated by the historical event aggregation layer. The aggregation ratio used when forming the revised historical event characteristics The processing stage of candidate historical events Event identifiers for handling candidate historical events The identifier of the previous processed event for the currently processed event. and the current processing stage of the event. , Indicates the index of the currently processed event. Indicates the candidate historical event index. This represents the set of candidate historical event indices corresponding to the currently processed event, along with the corrected historical event features. The corrected historical event features are represented by a 128-dimensional vector. The 128 dimensions utilize the unified feature space of the event feature encoding layer, formed by aggregating candidate historical event features according to an aggregation ratio. These candidate historical event features originate from the encoding results of sample version, processing stage, event-related objects, event time, processed event identifier, and the identifier of the previous processed event. The aggregation ratio... Indicates the characteristics of candidate historical events The proportion of historical event aggregation process, and the handling of event identifiers. ,in This represents a sample version of a candidate historical event. This indicates the stage sequence corresponding to the processing stage of a candidate historical event. This represents the event-related object of the candidate historical event, and the identifier of the previous processed event. ,in This indicates the sample version corresponding to the previous processed event declared in the current event processing statement. This indicates the stage sequence number corresponding to the previous event that the current event declaration is attached to. This indicates the event-related object that the current event declaration inherits from;

[0121] The link closure determination layer uses the processing events corresponding to the same station-end sample as the determination range, and the determination range index set. It consists of an index of processing event associations for samples belonging to the same station in the sample processing dynamic graph, which includes a processing event identifier and the identifier of the previous processing event. Samples from the same station are associated based on their sample version. Station-side sample identifier Sure, Indicates the sample version update order; the link closure decision layer is based on the decision range index set. Read the corrected historical event characteristics corresponding to each currently processed event Candidate historical event index set Aggregation ratio The processing stage of candidate historical events Event identifiers for handling candidate historical events The identifier of the previous processed event for the currently processed event. and the current processing stage of the event. The centralized control station's review and processing event serves as the starting point of the sample review upload and receive link, and it is not required that the centralized control station's review and processing event have a corresponding relationship with the previous processing event;

[0122] The link closure determination layer consists of a fully connected layer with 64 neurons and an output layer with 5 neurons. The fully connected layer with 64 neurons receives 128 corrected historical event features. The system generates 64-dimensional intermediate features for link determination. The output layer of five neurons receives these features and sets five state output components, corresponding to closure, broken link, incorrect link, skipped link, and unfilled link, respectively. The link closure determination layer does not read the "approved," "uploaded," or "received" status fields from the original log; instead, it uses modified historical event features. Aggregation ratio The relationship between the processing stage and the sample review, upload, and reception links is determined to maintain a closed loop.

[0123] The processing stage succession relationship indicates whether there is a continuous succession between adjacent processing stages. The processing stage succession relationship is derived from the aggregation ratio of candidate historical event features. and the processing stage of candidate historical events For the processing stage To encapsulate the currently processed uploaded event, the link closure determination layer uses the candidate historical event index set. Search for and process event identifiers Compared with the previous event identifier For the same candidate historical event, if the processing stage of the found candidate historical event is... The centralized control station reviews the candidate historical events, and the revised historical event characteristics are formed. The corresponding polymerization ratio Greater than Therefore, the processing stages between sample review and packaging / uploading are determined to be continuous. For the processing stages... To receive the currently processed events from the provincial artificial intelligence platform, the link closure determination layer uses the candidate historical event index set. Search for and process event identifiers Compared with the previous event identifier For the same candidate historical event, if the processing stage of the found candidate historical event is... To encapsulate the upload, and to form corrected historical event features for candidate historical events. The corresponding polymerization ratio Greater than Therefore, the processing stage relationship between the encapsulation upload and the platform receiving and filling is determined to be continuous.

[0124] When the candidate historical event index set There are multiple event handling identifiers in it. Compared with the previous event identifier When the same candidate historical events occur, the link closure determination layer selects the aggregation ratio. The largest candidate historical event is used as the corresponding event among the candidate historical events for the previous processing event declared by the current processing event. If multiple candidate historical events have the same aggregation ratio, the candidate historical event with the latest event time is selected. If multiple candidate historical events have the same event time, the candidate historical event with the largest sample processing event sequence index is selected. This selection rule is only used to determine the correspondence in the link closure determination information and does not change the corrected historical event features already formed by the historical event aggregation layer. ;

[0125] The link closure determination layer generates link closure determination information. Link closure determination information This includes the processing stage relationships between sample review and packaged upload, the processing stage relationships between packaged upload and platform-received feedback, the presence of platform-received feedback processing events, the correspondence of the previous processing event declared by the current processing event in the candidate historical events, the sample version correspondence of the dominant candidate historical events participating in historical event aggregation, and the correspondence of event-related objects, as well as the presence of platform-received feedback processing events in the judgment scope index set. The processing stage involves the provincial artificial intelligence platform receiving and processing events during the backfilling process. In the decision range index set When there is no processing stage in the process of receiving and filling back the event on the provincial artificial intelligence platform, the processing stage is not taken. The corresponding information of the previous processed event declared in the candidate historical events is obtained by comparison. and The corresponding sample versions were obtained through comparison. and The corresponding information about the event-related objects is obtained through comparison. and The dominant candidate historical events participating in the historical event aggregation are obtained as the candidate historical event index set. Medium polymerization ratio The biggest candidate historical event;

[0126] The sample review, upload, and receive link closure status includes closed, broken, incorrect, skipped, and unfilled. Closed indicates a continuous connection between sample review and encapsulation upload, and a continuous connection between encapsulation upload and the provincial AI platform's receiving and filling. Broken link indicates that the previous processing event declared by the current processing event does not have a corresponding relationship in the candidate historical events, i.e., the candidate historical event index set. There is no event handling identifier in the middle. Compared with the previous event identifier For the same candidate historical event, a broken link indicates that the candidate historical event participates in historical event aggregation, but the sample version or event-related object of the dominant candidate historical event participating in the historical event aggregation does not correspond to the previous processed event declared by the current processed event. A skipped link indicates that the processing stage did not form a continuous succession according to the centralized control station's review, packaging and uploading, and the provincial artificial intelligence platform's receiving and backfilling, and the judgment range index set is also affected. There are processing events corresponding to the provincial artificial intelligence platform receiving feedback. The absence of feedback indicates a continuous connection between sample review and packaging upload, but there are no processing events corresponding to the provincial artificial intelligence platform receiving feedback that form a continuous connection with packaging upload.

[0127] The link closure determination layer determines the link closure status of the sample review upload and reception link in the order of closure, no backfilling, skipped link, broken link, and broken link. The determination order is set to index the same set within the determination range. When multiple state conditions are met simultaneously, a unique state is output. The "no backfill" condition and the "broken link" condition may occur simultaneously. This is because when the previous processing event corresponding to the backfill received by the provincial artificial intelligence platform is missing, there is no continuous connection between the encapsulation upload and the platform's backfill received. The judgment order places "no backfill" before "broken link" to prioritize expressing the absence of platform backfill received when there is already a continuous connection between sample review and encapsulation upload. The "no backfill" condition and the "skipped link" condition may also occur simultaneously when there is a platform backfill processing event but the platform backfill processing event does not form a continuous connection with the encapsulation upload. The judgment order places "no backfill" before "skipped link" to prioritize expressing that no effective platform backfill connection has been formed after encapsulation upload. The "skipped link" condition is placed before "wrong link" and "broken link" to prioritize expressing stage skipping when the provincial artificial intelligence platform's backfill processing event has occurred but the processing stage order is not continuous.

[0128] If the link closure determination information This indicates a continuous connection between sample review and packaging / uploading, and between packaging / uploading and platform reception / response. Therefore, the sample review / upload / reception link is considered closed. (The link closure determination information is missing from the original text.) This indicates a continuous connection between sample review and encapsulation upload, and there is no corresponding processing event for the provincial AI platform's feedback mechanism that forms a continuous connection with the encapsulation upload. Therefore, the closed-loop status of the sample review-upload-receive link is determined to be "no feedback." If the link closure determination information... This indicates that the processing phase did not follow the continuous process of review by the centralized control station, packaging and uploading, and receiving and backfilling by the provincial artificial intelligence platform, and that the platform's receiving and backfilling processing events were marked and retrieved. If the sample review upload and receive link closure status is determined to be a hop link, then the link closure determination information... This indicates the sample versions of the dominant candidate historical events participating in the historical event aggregation. Sample version corresponding to the previous event that the current event handling declaration inherits from. Inconsistent, or event-related objects of the dominant candidate historical event participating in historical event aggregation. The event associated object corresponding to the previous event that the current event declaration inherits from. If there is a discrepancy, the sample review upload and receive link closure status is determined to be a faulty link. If the link closure determination information... This indicates that the previous event declared for the current event being processed is not in the candidate historical event index set. If a corresponding relationship is formed, the closed state of the sample review upload and receive link is determined to be a broken link. Based on the determined closed state of the sample review upload and receive link, the link closure determination layer selects the state output component corresponding to closure, broken link, wrong link, skipped link, or no backfill from the output layer of the 5 neurons, forming the link closure determination result between sample review, encapsulation upload, and platform reception backfill.

[0129] Example 2

[0130] Example 2 is used to supplement the explanation of the training method and parameter acquisition method of the event-aware dynamic graph learning model, without changing the link closure determination process in Example 1;

[0131] This embodiment performs supervised training on the event-aware dynamic graph learning model. The dataset includes event details, link summary, encapsulated upload object and field description. Event details are used to construct the sample processing dynamic graph, processing event identifier, previous processing event identifier and candidate historical event set. Link summary is used to form link closure state labels.

[0132] Table 1. Examples of training data and iteration results

[0133]

[0134] During training, the sample processing events under the same link are first grouped into the same training sample according to the link number. Then, a sample processing dynamic graph with an identifier field is constructed based on the sample version, processing stage, event associated object, event time, processing event identifier, and previous processing event identifier. For each current processing event, the event-aware dynamic graph learning model extracts the features of the current processing event and the features of candidate historical events. The event similarity is then corrected based on the identifier difference between the previous processing event identifier and the processing event identifier of the candidate historical event to obtain the corrected historical event features.

[0135] Link state labels are determined only by link aggregation during the training phase. When the same link meets multiple state conditions at the same time, a unique training label is determined in the order of closed, unfilled, skipped, wrong, and broken. This order is only used for training sample labeling and is not used as an external decision rule during the model inference phase.

[0136] The link closure determination layer receives the corrected historical event features, which are then processed by a fully connected layer of 64 neurons to form intermediate features for link determination. The output layer of 5 neurons then produces output components for the states of closure, broken link, wrong link, skipped link, and unfilled link, respectively. During training, the five output components are normalized to obtain the predicted probabilities of the five states. Cross-entropy loss is then used to update the weight and bias parameters of the event feature encoding layer, historical event aggregation layer, and link closure determination layer.

[0137] After the model training is completed, only the sample processing event records of the link to be judged are input during the inference phase. The trained event-aware dynamic graph learning model outputs five state components, and the state corresponding to the state component with the largest value is determined as the link closure judgment result.

[0138] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A deep learning-based method for determining the closure of a sample review upload and reception link, characterized in that, include: S1. Obtain the sample processing event records generated during the process of station-side sample review, packaging and uploading at the centralized control station and receiving and backfilling on the provincial artificial intelligence platform. The sample processing event records include the sample version, processing stage, event associated object, event time, and the previous processing event declared to be undertaken by the current processing event. The previous processing event declared to be undertaken by the current processing event includes the corresponding sample version, processing stage, and event associated object, thus obtaining the sample processing event sequence. S2. Construct a dynamic graph of sample processing based on the sequence of sample processing events. The dynamic graph of sample processing includes sample version nodes and the relationship between processing events between sample version nodes. S3. Based on the sample processing dynamic diagram, generate a processing event identifier and a previous processing event identifier for each processing event. The processing event identifier is formed by the sample version, processing stage, and event-related object. The previous processing event identifier is formed by the sample version, processing stage, and event-related object corresponding to the previous processing event declared by the current processing event. This results in a sample processing dynamic diagram with processing event identifiers and previous processing event identifiers. S4. Based on the sample processing dynamic graph with processing event identifier and previous processing event identifier, execute the event-aware dynamic graph learning model to extract the features of the current processing event and the features of candidate historical events. S5. In the historical event aggregation process of the event-aware dynamic graph learning model, based on the difference value between the identifier of the previous processed event of the current processed event and the processed event identifier of the candidate historical event, the corrected event similarity between the current processed event features and the candidate historical event features is calculated, and the candidate historical event features are aggregated based on the corrected event similarity to obtain the corrected historical event features. S6. Determine the closed state of the sample review, upload, and receive link based on the corrected historical event characteristics, and output the closed state determination result of the link between sample review, encapsulation and upload, and platform reception and backfilling.

2. The method for determining the closure of the sample review upload and reception link based on deep learning according to claim 1, characterized in that, S1 specifically refers to: The system acquires sample processing event records generated during the process of station-side sample review, packaging and uploading at the centralized control station, and receiving and backfilling on the provincial artificial intelligence platform. It extracts the sample version, processing stage, event-related object, event time, and the previous processing event declared by the current processing event from each sample processing event record to form a sample processing event record containing unified fields. Based on the processing stage, sample processing event records containing uniform fields are classified according to the processing stage, forming sample processing event records classified by processing stage; Based on the sample version, the event associated object, and the previous processing event declared by the current processing event, the sample processing event records classified by processing stage are grouped according to the same station sample to form grouped sample processing event records. The sample processing event records after grouping are sorted according to the relationship between event time and processing stage. If samples on the same station are subject to re-review, re-encapsulation, or failed retransmission, the sorting position is adjusted according to the sample version update order to obtain the sample processing event sequence.

3. The method for determining the closure of the sample review upload and reception link based on deep learning according to claim 1, characterized in that, S2 specifically refers to: Receive the sample processing event sequence, extract the sample version, processing stage and event-related object from each processing event in the sample processing event sequence, and group the processing events corresponding to the same sample version into the same sample version node to form an initial sample version node set; Based on the order in which sample versions appear in the sample processing event sequence in the initial sample version node set, the phase advancement direction between sample version nodes is determined, forming a sample version node set with a phase advancement direction. Based on the processing stages and associated objects in the sample processing event sequence, the processing event association relationship is constructed for the sample version node set with stage advancement direction, so that the processing event association relationship of the centralized control station's review, packaging and uploading and the provincial artificial intelligence platform's reception and backfilling is connected to the corresponding sample version node; When multiple sample version nodes correspond to the same event-related object, the processing event association relationship of each sample version node is updated according to the sample version in the sample processing event sequence, forming a dynamic graph of sample processing without merging sample version nodes.

4. The method for determining the closure of the sample review upload and reception link based on deep learning according to claim 1, characterized in that, S3 specifically refers to: Receive the dynamic graph of sample processing, traverse the processing event relationships between sample version nodes, and read the sample version node corresponding to the starting point of the processing event relationship, the sample version node corresponding to the ending point of the processing event relationship, the processing stage and the event association object from each processing event relationship to form a set of processing events that need to be identified. The sample version is determined by the sample version node corresponding to the start and end points of the processing event association relationship in the processing event set generated according to the requirements. The processing stage and event association object are bound to the corresponding sample version to form a processing event combination composed of sample version, processing stage and event association object. According to the order of sample version, processing stage and event-related object in the sample review upload and receive link, the processing event combination generates a processing event identifier for each processing event, and maps the processing event identifier to the corresponding processing event in the processing event set that needs to generate the identifier; Based on the processing event association relationship corresponding to the current processing event in the sample processing dynamic graph, read the previous processing event declared to be inherited by the current processing event; Based on the sample version, processing stage, and event-related object of the previous processing event declared by the current processing event, generate the identifier of the previous processing event for the current processing event; If there exists a processing event in the set of processing events that needs to generate an identifier that corresponds to the previous processing event declared by the current processing event, make the identifier of the previous processing event of the current processing event consistent with the processing event identifier of the corresponding processing event. If there is no processing event in the set of processing events for which an identifier needs to be generated that corresponds to the previous processing event declared by the current processing event, the identifier of the previous processing event generated based on the sample version, processing stage and event associated object of the previous processing event declared by the current processing event is retained. When multiple sample version nodes are connected to the same event-related object, or when the same sample version is re-approved, re-encapsulated, or retransmitted after failure, the previous processing event identifier of the current processing event is determined based on the sample version, processing stage, and event-related object of the previous processing event declared by the current processing event, thus forming a processing event with the processing event identifier and the previous processing event identifier. The processing event relationships in the sample processing dynamic graph are updated based on the processing events with processing event identifiers and previous processing event identifiers, so that each processing event relationship carries the processing event identifier and previous processing event identifier of the corresponding processing event, resulting in a sample processing dynamic graph with processing event identifiers and previous processing event identifiers.

5. The method for determining the closure of the sample review upload and reception link based on deep learning according to claim 1, characterized in that, S4 specifically refers to: Receive a sample processing animation with a processing event identifier and a previous processing event identifier, determine the current processing event for which features are to be extracted, and read the sample version, processing stage, event associated object, event time, processing event identifier, and previous processing event identifier from the current processing event to form the current processing event encoding input; The event-aware dynamic graph learning model is executed based on the current event encoding input. The sample version, processing stage, event associated object, event time, processing event identifier, and previous processing event identifier are encoded to obtain the current event features. Based on the sample version, event associated object, and event time in the current event processing code input, the event processing dynamic graph with the processing event identifier and the previous processing event identifier is read to form a candidate historical event set. The event time is earlier than the current event processing time and contains the same sample version node or the same event associated object. Based on the sample version, processing stage, associated object, event time, processing event identifier, and previous processing event identifier of the candidate historical events in the candidate historical event set, the candidate historical events are encoded using the same encoding method as the current processing event in the event-aware dynamic graph learning model, resulting in candidate historical event features that are in the same feature space as the current processing event features.

6. The method for determining the closure of the sample review upload and reception link based on deep learning according to claim 1, characterized in that, S5 specifically refers to: During the historical event aggregation process of the event-aware dynamic graph learning model, the features of the currently processed event, the features of the candidate historical events, the identifier of the previous processed event of the currently processed event, and the processing event identifiers of the candidate historical events are received. Each candidate historical event feature is bound to the corresponding processing event identifier of the candidate historical event to form a set of historical event features participating in the aggregation. Based on the feature similarity calculation between the current event features and the candidate historical event features in the set of historical event features participating in the aggregation, the initial event similarity corresponding to the candidate historical event is obtained. Based on the identifier of the previous processing event of the current processing event and the processing event identifier of the candidate historical events in the feature set of historical events participating in the aggregation, the identifier difference value is calculated for the sample version, processing stage and event associated object respectively to form the identifier difference value; The identification difference value is embedded into the correction calculation of the initial event similarity. The similarity correction coefficient is determined based on the identification difference value. If the identification difference value is not greater than the preset threshold, the similarity correction coefficient is determined as the coefficient that keeps the similarity of the corresponding initial event unchanged. If the identification difference value is greater than the preset threshold, the similarity correction coefficient is determined as the coefficient that reduces the similarity of the corresponding initial event. The similarity of the corresponding initial event is corrected based on the similarity correction coefficient to obtain the corrected event similarity. The aggregation ratio of each candidate historical event feature in the set of historical event features participating in the aggregation is determined based on the corrected event similarity, and the candidate historical event features are updated according to the aggregation ratio to form the candidate historical event features updated according to the aggregation ratio. Historical event aggregation is performed based on the updated candidate historical event features according to the aggregation ratio to obtain the corrected historical event features.

7. The deep learning-based sample review upload and reception link closure determination method according to claim 1, characterized in that, S6 specifically refers to: Receive the corrected historical event features, and determine the processing stage connection between sample review and encapsulation upload, as well as the processing stage connection between encapsulation upload and platform reception and feedback, based on the aggregation ratio of candidate historical event features used when forming the corrected historical event features and the processing stage of candidate historical events. Based on the processing stage succession relationship and the corresponding situation of the previous processing event declared to be succeeded by the current processing event in the candidate historical events in the corrected historical event characteristics, link closure determination information is formed. The link closure determination information is used to determine the closure status of the sample review upload and receive link. The closure status of the sample review upload and receive link includes closed, broken link, wrong link, skipped link, or not filled back. Based on the closed status of the sample review, upload, and receive link, output the link closure determination result between sample review, encapsulation upload, and platform reception feedback.

8. The deep learning-based sample review upload and reception link closure determination method according to claim 4, characterized in that, When generating the previous processing event identifier for the current processing event, if the sample processing dynamic graph contains multiple sample version nodes connected to the same event-related object, or if the same sample version corresponds to re-review, re-encapsulation, or failure retransmission, then based on the previous processing event declared by the current processing event, the processing event corresponding to the processing stage of the previous processing event declared by the current processing event is determined in the sample processing dynamic graph. The sample version and event-related object are then matched again within the processing events corresponding to the processing stage, without merging multiple processing events corresponding to the same event-related object into the same previous processing event identifier. When the sample version, processing stage, and event associated object of the processed event all correspond to the previous processed event declared by the current processed event, the processed event identifier of the matched processed event is used as the previous processed event identifier of the current processed event. When no processed event is matched that corresponds to the previous processed event declared by the current processed event, the previous processed event identifier generated based on the sample version, processing stage, and event associated object of the previous processed event declared by the current processed event is retained, and the previous processed event identifier of the current processed event is written into the corresponding processed event association relationship.

9. The method for determining the closure of the sample review upload and reception link based on deep learning according to claim 6, characterized in that, When the identifier difference value is used in the calculation of the initial event similarity correction, the processing stage in the identifier of the previous processing event of the current processing event is first matched with the processing stage in the identifier of the processing event of the candidate historical event. If the processing stages do not correspond, the initial event similarity of the candidate historical event is reduced according to the similarity correction coefficient corresponding to the identifier difference value, and the features of the candidate historical event are retained for historical event aggregation. If the processing stages correspond, the sample version and event associated object in the identifier of the previous processing event of the current processing event are matched with the sample version and event associated object in the identifier of the processing event of the candidate historical event, respectively, and the identifier difference value is updated according to the matching results. When the sample version, processing stage, and event associated object all correspond, the initial event similarity of the candidate historical event is retained. When the sample version, processing stage, or event associated object does not correspond, the initial event similarity of the candidate historical event is reduced according to the similarity correction coefficient corresponding to the updated identifier difference value, and the corrected event similarity used to determine the aggregation ratio is obtained.

10. The method for determining the closure of the sample review upload and reception link based on deep learning according to claim 7, characterized in that, When determining the link closure results between sample review, package upload, and platform reception feedback, the link closure determination results are determined in the following order: closed, no feedback, skipped link, wrong link, and broken link. If the link closure determination information indicates that there is a continuous connection between sample review and package upload, and between package upload and platform reception feedback, then the output is closed. If the link closure determination information indicates that there is a continuous connection between sample review and package upload, and there is no processing event corresponding to platform reception feedback that forms a continuous connection with package upload, then the output is no feedback. If the processing stage connection relationship indicates that the processing stage does not form a continuous connection according to sample review, package upload, and platform reception feedback, and there is a processing event corresponding to platform reception feedback, then the output is skipped link. If the link closure determination information indicates that the candidate historical event participates in historical event aggregation, but the sample version or event associated object does not correspond to the previous processing event declared as being received by the current processing event, then the output is wrong link. If the link closure determination information indicates that the previous processing event declared as being received by the current processing event does not form a corresponding relationship in the candidate historical events, then the output is broken link.

Images