A cloud computing-based user social interaction data management system and method

Abstract

The application discloses a user social interaction data management system and method based on cloud computing and relates to the technical field of data processing.The user social interaction data management method based on cloud computing comprises the following steps: S1, collecting user social interaction data of multiple platforms, and storing and constructing an event timeline database after preprocessing; S2, performing homologous merging analysis through time sequence offset features, field matching features and link association feature data; S3, performing event chain integration mapping analysis by comprehensively merging result parameters and task queue operation data; and S4, performing time sequence stability analysis based on link segment integration results and task execution states, and performing timeline recovery, maintenance inhibition and instability alarm control operations according to the time sequence stability analysis results.The problems of homologous event misplacement projection, event chain semantic destruction and unstable downstream processing caused by multi-platform asynchronous writing, cache back flushing and breakpoint supplement transmission are solved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically a user social interaction data management system and method based on cloud computing. Background Technology

[0002] With the continuous development of cloud computing, big data processing, and multi-platform collaborative applications, the collection, aggregation, and management of user social interaction data have become fundamental data processing steps in internet platforms, campus platforms, content platforms, and integrated business platforms. In existing technologies, user comments, replies, forwards, messages, voice and video interactions, and status change records generated across different terminals, platform nodes, and interaction links are typically uploaded to the cloud via source platform links, content links, and task processing links. The cloud then performs unified storage, unified numbering and mapping, unified timeline organization, and unified status management for multi-platform events, thereby supporting downstream business processing such as profile updates, statistical updates, anomaly detection, and traceability writing.

[0003] For example, the invention patent CN117472640B discloses a cross-platform event processing method and system. It analyzes the event processing logs of the user terminal to obtain event processing attribute information, thereby identifying abnormal event blocks in the event processing process of the server currently connected to the user terminal. It also sends the abnormal event block processing status to all servers within the network platform, determines the target server suitable for processing the abnormal event blocks, and selects a reliable server for processing the abnormal event blocks. It extracts the event data to be processed corresponding to the abnormal event blocks from the event database of the user terminal and sends it to the target server, while adjusting the event process of the server currently connected to the user terminal. It returns the valid processing results generated by the target server to the time processing data stream of the currently connected server, utilizing other servers to assist the current server's event task processing work, thus realizing cross-platform processing of event tasks across different servers.

[0004] For example, the invention patent CN121502615B discloses a method and system for identifying key anomaly events under cloud-based multimodal consistency constraints. Targeting multimodal time-series data generated by devices such as cameras and environmental sensors in smart city and industrial IoT scenarios, it proposes a fully unsupervised event-level anomaly understanding framework: accurately locating collaborative anomaly periods through a multimodal consistency constraint mechanism; forming event clusters through online clustering of anomaly periods; extracting the most anomalous feature dimensions from each cluster based on Z-score saliency analysis and mapping them to human-readable semantic descriptions; quantifying the criticality of events by combining three indicators: the number of affected devices, duration, and state deviation, and outputting a ranked list. This invention does not require predefined event types or labeled data, and can simultaneously discover, name, and rank known and unknown anomaly events, significantly improving the efficiency of discovering and responding to key anomalies. It is suitable for lightweight cloud deployment in large-scale sensing networks.

[0005] However, in actual operation, social interaction data from multiple platforms often exhibits significant asynchronous writing characteristics. Affected by factors such as network jitter, node caching delays, link retries, cache refreshes, breakpoint retransmissions, and inconsistent write rhythms among distributed nodes, a single real interaction event involving the same user, the same interaction object, and the same session may result in multiple records in the cloud with similar fields but different time positions. These records have offsets in the time of event occurrence, event writing, and event cloud upload. They may also be accompanied by similar digest prefixes, identical status codes, similar business serial numbers, or related batch links. When out-of-order writes, refresh writes, and retransmission writes occur simultaneously, relying solely on time windows and field similarity for merging can easily lead to the incorrect segmentation of multiple records belonging to the same real event into multiple independent events, and can also easily misclassify multiple interaction events that should exist independently into the same event chain. This will further affect the continuity of the event timeline and the stability of downstream business results. On the one hand, if the event chain is incorrectly segmented or merged, it can easily lead to deviations in anomaly detection results, inconsistent profile update criteria, duplicate accumulation of statistical results, and broken traceability links, resulting in inconsistent representations of the same real event in different business tables. On the other hand, mirrored records, duplicate triggered records, and recalculation tasks repeatedly enter the statistical update, profile update, and anomaly detection processes in the queue, which can also lead to duplicate calculations, queue accumulation, and resource waste, increasing the depth of the computation task queue, increasing task processing latency, and further exacerbating the instability problems after the event timeline is reconstructed.

[0006] Therefore, there is an urgent need for a cloud-based user social interaction data management system and methodology. Summary of the Invention

[0007] To address the above problems, this invention provides a cloud-based user social interaction data management system and method to solve the problems of misaligned projection of the same source event, semantic destruction of the event chain, and instability of downstream processing caused by asynchronous writing across multiple platforms, cache flushing, and breakpoint retransmission.

[0008] To achieve the above objectives, the technical solution adopted by this invention is as follows: S1, collect user social interaction data from multiple platforms, preprocess the user social interaction data, store it, and construct an event timeline database; S2, perform source merging analysis using time series offset features, field matching features, and link association features, and perform master record selection, mirror record attachment, independent event retention, and merging result parameter statistics operations based on the merging analysis results; S3, perform event chain integration mapping analysis by combining merging result parameters with task queue running data, perform segment merging, task deduplication and folding, and mirror record marking operations based on the integration mapping analysis results, and output the chain segment integration results; S4, perform time series stability analysis based on the chain segment integration results and task execution status, and perform timeline recovery, maintenance suppression, and instability alarm control operations based on the time series stability analysis results.

[0009] Further, the specific steps for collecting user social interaction data from multiple platforms are as follows: Collect user social interaction data: obtain user ID, interaction object ID, session ID, event type, original event ID, business serial number, source platform ID, source node ID, event occurrence time, event writing time, and event cloud upload time through the source platform link; obtain text length, text digest code, number of attachments, interaction direction marker, and event status code through the interaction content link; obtain cache refresh batch number, breakpoint retransmission batch number, retry count, node cache dwell time, and source log sequence number through the link operation record; obtain the complete number of fields by counting the total number of non-empty core fields in a single record; and obtain the complete number of fields by monitoring the task queue. The test link obtains the task number, task enqueue time, task dequeue time, task status flag, task trigger source flag, task queue depth, and task queue capacity. The profile update record link obtains the profile update number, profile update time, master record number, and profile update trigger source flag. The statistical update record link obtains the statistical update number, statistical update time, master record number, and statistical update trigger source flag. The anomaly detection record link obtains the anomaly detection number, anomaly detection time, anomaly judgment result, master record number, and anomaly detection trigger source flag. The trace write record link obtains the trace write number, trace write time, master record number, and trace write trigger source flag.

[0010] Further, the specific steps for preprocessing and storing user social interaction data to construct an event timeline database are as follows: Illegal and empty records in the user social interaction data are removed; time units, field names, status code definitions, and text digest code lengths are standardized; numerical fields are mapped to a unified interval using Z-score standardization and range normalization; the time base of time fields is corrected; the mapping number field is associated; and the definitions of text digest codes, event status codes, and interaction direction markers are consolidated. Within the same user ID, interaction object ID, session ID, and event type range, candidate event pairs are established based on the maximum time interval t; the corresponding offset duration is obtained based on the difference between the event occurrence, writing, and cloud upload times; and the number of consistent core fields is obtained through item-by-item comparison. The algorithm obtains the consistent length of the digest prefix by counting the consecutive consistent bits of the digest code prefix, obtains the payload difference by summing the difference between the text length, the number of attachments, the event status code, and the interaction direction marker, and obtains the batch association marker based on the same cache hit. The algorithm counts the number of revoked duplicate statistics, profile updates, and anomaly detections to obtain the number of duplicate calculations and reductions, counts the number of anomaly judgment changes to obtain the number of anomaly judgment flips, counts the number of tasks that re-enter the statistical update and profile update links after timeline reconstruction to obtain the number of downstream recalculation tasks, and counts the number of master records that enter the statistical update, profile update, anomaly detection, and trace write links to obtain the number of master records released. The algorithm stores the preprocessed user social interaction data and constructs the event timeline database.

[0011] Furthermore, the specific steps for performing source merging analysis using temporal offset features, field matching features, and link association features are as follows: Obtain the event occurrence offset duration, event write offset duration, event cloud upload offset duration, number of consistent core fields, length of consistent summary prefix, number of load differences, and batch association markers; subtract the square of the event occurrence offset duration from the event write offset duration, and subtract the square of the event occurrence offset duration from the event cloud upload offset duration, sum them, and take the square root to obtain the temporal discrete term; sum the number of consistent core fields, length of consistent summary prefix, and batch association markers, divide by the number of load differences plus one, and obtain the field fitting ratio; divide the temporal discrete term by the event occurrence offset duration plus one to obtain the asynchronous projection term; divide the field fitting ratio by the sum of the asynchronous projection term and one to obtain the source merging value.

[0012] Furthermore, the specific steps for performing master record selection and mirror record attachment, independent event retention, and merge result parameter statistics based on the merge analysis results are as follows: By comparing the source merging value and the merging threshold in real time, when the source merging value is less than the merging threshold, the independent timeline write path of the current candidate event pair is retained, and it is verified whether the event offset duration is greater than the independent separation duration, whether the summary prefix consistency length is less than the summary fitting threshold, whether the batch association marker is zero, and whether the load difference number is greater than the difference threshold; when the above conditions are met, the two records are treated as independent events; when the above conditions are not met, the current candidate event pair is written to the observation queue, and a review is performed in the next observation period. If the source merging value is still less than the merging threshold after the review, it is marked as independent. Event logging; when the same-source merge value is greater than or equal to the merge threshold, the current candidate event pair is written into the same-source merge cluster, and the master record is selected in the order of the earliest event occurrence time and the largest number of complete fields; when the number of complete fields is the same, the record with the latest event cloud upload time is selected as the master record; the remaining records are marked as mirror records and attached to the master record, and the duplicate anomaly count, duplicate profile update task, and duplicate statistics task triggered by the mirror record are canceled; within the same master record number association range, the number of mirror records and the number of master records are counted according to the mirror record attachment result, and the number of erroneously segmented candidates and the number of continuous anomaly fragments are counted according to the breakage of candidate event fragments under the event occurrence time sequence, and sent to the event chain integration modeling module together with the same-source merge value.

[0013] Furthermore, the specific steps for integrating the merge result parameters with the task queue operation data to perform event chain integration mapping analysis are as follows: The event chain integration mapping process is executed using the same-source merge value, number of mirror records, number of master records, number of candidate missegments, number of consecutive abnormal segments, calculation task queue depth, and calculation task queue capacity as basic inputs; based on the event type, interaction direction marker, source platform number combination, and batch association marker status, valid event records within the current observation window that have completed same-source merge verification, have the same event type, the same interaction direction marker, the same source platform number combination, and consistent batch association marker status are extracted from the event timeline database, generating misaligned merge records. Features, mirror-attached features, fragment fragmentation features, and task queuing features are used. Tensor concatenation is performed on the misalignment merging features, mirror-attached features, fragment fragmentation features, and task queuing features to construct an event chain integration state matrix. The number of fragments merged, the number of repeated calculations reduced, the number of anomaly judgment flips, and the number of downstream recalculation tasks corresponding to the observation window that meets the integration threshold condition are used as training labels. With the constraints of increasing the number of fragments merged, increasing the number of repeated calculations reduced, decreasing the number of anomaly judgment flips, and decreasing the number of downstream recalculation tasks, a multilayer perceptron is used to perform nonlinear mapping on the event chain integration state matrix to construct an event chain integration model and output the chain segment integration value.

[0014] Further, the specific steps for performing fragment merging, task deduplication and folding, and mirror record marking operations based on the integrated mapping analysis results, and outputting the chain segment integration results are as follows: By comparing the chain segment integration value and the integration threshold in real time, when the chain segment integration value is less than the integration threshold, the primary record priority processing path is maintained, the continuous abnormal fragment extension path triggered by the mirror record is frozen, and only the primary record is allowed to participate in the abnormal fragment extension; when the chain segment integration value is greater than or equal to the integration threshold, the candidate fragments of missegmentation are reordered according to the event occurrence time within the same user number and the same session number range, and candidate fragments with consistent event type, consistent interaction object number, consistent summary prefix length greater than or equal to the splicing summary threshold, and event occurrence interval less than or equal to the splicing duration threshold are merged into reconstructed abnormal fragments; only one primary computing task is retained in the computing task queue; a primary record mapping relationship is established for the mirror record, and the separate triggering of statistical updates, profile updates, and anomaly detection by the mirror record is stopped.

[0015] Furthermore, the specific steps for time-series stability analysis based on the chain segment integration results and task execution status are as follows: Obtain the chain segment integration value, the number of main record releases, the number of repeated calculation reductions, the number of anomaly judgment flips, and the number of downstream recalculated tasks; Square the number of main record releases and the number of repeated calculation reductions, sum them, and take the square root to obtain the main chain recovery term; Square the number of anomaly judgment flips and the number of downstream recalculated tasks, sum them, and take the square root to obtain the residual disturbance term; Divide the residual disturbance term by the chain segment integration value plus one to obtain the disturbance reduction term; Divide the main chain recovery term by the sum of the disturbance reduction term and one to obtain the timeline stability value.

[0016] Furthermore, the specific steps for performing timeline recovery, maintenance suppression, and instability alarm control operations based on the timeline stability analysis results are as follows: By comparing the timeline stability value and the stability threshold in real time, when the timeline stability value is greater than or equal to the stability threshold, the mirror record status field in the metadata table within the master record's associated range is changed from the maintained state to the archived state, and the index key of the mirror record is pointed to the master record number; master records with anomaly detection flip count less than or equal to the flip control threshold and downstream recalculation task number less than or equal to the recalculation control threshold are allowed to enter the statistical link, and master record-driven profile updates, anomaly detection, and trace writes are restored; when the timeline stability value is less than the stability threshold, the master record priority processing path is maintained, allowing only one master record to be processed at a time. Triggering one profile update and one anomaly detection within each observation window freezes downstream recalculation tasks triggered by mirror records. In the anomaly determination result table, a lock flag is written to the anomaly determination result field within the same master record's associated range, and the number of new recalculation tasks within each observation window is limited to m. In the retention duration configuration table, a short retention duration field is added to the retention duration fields for mirror record mapping relationships, fragment merging results, and merged source record comparison relationships. Candidate fragments that have not yet been verified continue to be retained in the order of event occurrence time, event write time, and source log number. When the timeline stability value of n consecutive observation windows is still less than the stability threshold, an event timeline instability alarm record is generated, retaining only the anomaly detection, trace query, and basic profile update tasks driven by the master record.

[0017] Furthermore, a second aspect of the present invention provides a cloud-based user social interaction data management system, employing a cloud-based user social interaction data management method, comprising: a multi-platform event acquisition and preprocessing module, used to acquire user social interaction data from multiple platforms, preprocess the user social interaction data, store it, and construct an event timeline database; a same-source misalignment merging identification module, used to perform same-source merging analysis through time series offset features, field matching features, and link association features, and perform master record selection and mirror record attachment, independent event retention, and merging result parameter statistics operations based on the merging analysis results; an event chain integration modeling module, used to perform event chain integration mapping analysis by integrating merging result parameters and task queue running data, and perform segment merging, task deduplication and folding, and mirror record marking operations based on the integration mapping analysis results, and output chain segment integration results; and a timeline stability verification module, used to perform time series stability analysis based on chain segment integration results and task execution status, and perform timeline recovery, maintenance suppression, and instability alarm control operations based on the time series stability analysis results. Attached Figure Description

[0018] Figure 1 This is a flowchart of a cloud computing-based user social interaction data management method according to the present invention;

[0019] Figure 2This is a structural diagram of a cloud-based user social interaction data management system according to the present invention;

[0020] Figure 3 This invention provides a spatial distribution surface diagram for determining the misaligned projection homogeneous merging based on temporal discreteness and field alignment.

[0021] Figure 4 This is a steady-state quantization time series change trend diagram of the event timeline of this invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solution, the present invention will be described in detail below with reference to embodiments. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.

[0023] Please see Figures 1-4 This invention provides a technical solution: a cloud-based user social interaction data management method, comprising: S1, collecting user social interaction data from multiple platforms, preprocessing the user social interaction data, storing it, and constructing an event timeline database; S2, performing source merging analysis using time-series offset features, field matching features, and link association features, and performing master record selection, mirror record attachment, independent event retention, and merging result parameter statistics operations based on the merging analysis results; S3, performing event chain integration mapping analysis by combining merging result parameters with task queue running data, performing segment merging, task deduplication and folding, and mirror record marking operations based on the integration mapping analysis results, and outputting the chain segment integration results; S4, performing time-series stability analysis based on the chain segment integration results and task execution status, and performing timeline recovery, maintenance suppression, and instability alarm control operations based on the time-series stability analysis results.

[0024] Specifically, the steps for collecting user social interaction data from multiple platforms are as follows: Collecting user social interaction data: Obtaining user ID, interaction object ID, session ID, event type, original event ID, business serial number, source platform ID, source node ID, event occurrence time, event writing time, and event cloud upload time through the source platform link. The source platform link refers to the access channel used to connect to the business front-end of each social platform and receive raw interactive event data in the form of a message queue or interface callback. Its input end is the event reporting interface of each social platform, and its output end is the standardized raw event record stream. Obtaining the text length and text digest code through the interaction content link. The number of attachments, interaction direction markers, and event status codes are included. The interactive content link refers to the processing channel used to parse and extract the text, attachments, and status information of interactive events. Its input is the event record preprocessed by the source platform link, and its output is the completed interactive content field set. The cache refresh batch number, breakpoint retransmission batch number, retry count, node cache dwell time, and source log sequence number are obtained through the link operation record. The link operation record refers to the log data stream that tracks and records the operating status, caching behavior, and retry mechanism generated by each data acquisition and processing link during transmission. Its input is the runtime monitoring probe of each link, and its output is... Link-level operational metrics and anomaly markers; the complete number of fields is obtained by counting the total number of non-empty core fields in a single record; the task queue monitoring link is used to obtain the computation task number, task enqueue time, task dequeue time, task status marker, task trigger source marker, computation task queue depth, and computation task queue capacity. The task queue monitoring link is a monitoring channel used to listen to and collect real-time status and task flow information of downstream computation task queues. Its input is the task queue status query interface or heartbeat reporting data, and its output is full lifecycle time-series data of the task dimension; the profile update record link is used to obtain the profile update number, profile update time, and master record. The record number and profile update trigger source marker are used to obtain the profile update record link, which is an audit channel used to record the change trajectory and trigger reason of the user profile model during each data update. Its input is the callback log of the profile update service, and its output is a structured record of the profile update operation. The statistical update record link is used to obtain the statistical update number, statistical update time, master record number and statistical update trigger source marker. The statistical update record link is a tracking channel used to record the operation details and trigger source of multidimensional statistical indicators during each recalculation or incremental update. Its input is the task execution log of the statistical calculation engine, and its output is a structured record of the statistical update operation.The anomaly detection record link obtains the anomaly detection number, anomaly detection time, anomaly judgment result, master record number, and anomaly detection trigger source marker. This anomaly detection record link is a traceable channel used to record the input features, judgment conclusion, and trigger context when the anomaly detection algorithm judges each master record. Its input is the execution log of the anomaly detection service, and its output is a structured record of the anomaly detection result. The trace write record link obtains the trace write number, trace write time, master record number, and trace write trigger source marker. This trace write record link is an audit channel used to record the backtracking association information and database entry trajectory of each master record written to the event timeline database after preprocessing. Its input is the application layer logging of the database write operation, and its output is a structured record of the write trace.

[0025] In this implementation plan, through multi-link collaborative collection, the main identifier, time sequence parameters, content characteristics, link operation status, task queue indicators, and update and detection records of each link in the downstream consumption link of user social interaction events are uniformly obtained. Field association and caliber alignment are completed under the dimension of main record number to form a structured raw data set covering the entire life cycle of the event.

[0026] Specifically, the steps for preprocessing and storing user social interaction data to construct an event timeline database are as follows: Illegal and empty records are removed from the user social interaction data; time units, field names, status code definitions, and text summary code lengths are standardized; numerical fields are mapped to a unified interval through Z-score standardization and range normalization. Z-score standardization involves subtracting the mean of all samples in the field from the original value of each numerical field and then dividing by the standard deviation, so that the transformed data follows a standard normal distribution with a mean of 0 and a standard deviation of 1. Range normalization involves adjusting the Z-score data... The standardized scores are then linearly scaled to the same specified range based on the maximum and minimum values ​​of each field to eliminate dimensional differences and unify the value ranges of different fields. The time base of the time field is corrected, the association mapping number field is adjusted, and the definitions of text summary codes, event status codes, and interaction direction markers are narrowed down. Within the same user ID, interaction object ID, session ID, and event type range, candidate event pairs are established based on the maximum time interval t. The maximum time interval t refers to a pre-set threshold for determining whether two interaction events within the same session can constitute an association analysis pair; its value can be determined based on social media... The distribution of business experience or historical data in interactive scenarios is determined. For example, in instant messaging scenarios, t can be set between 30 and 300 seconds. The specific value is configured by the system administrator based on the actual interaction frequency and event density. If the t value is too small, it will lead to incomplete coverage of candidate event pairs and omission of effective associations. If the t value is too large, it will introduce too many noisy event pairs and increase the computational overhead of subsequent comparisons. The corresponding offset duration is obtained based on the difference between the event occurrence, writing, and cloud upload times. The number of consistent core fields is obtained through item-by-item comparison. The length of the consistent digest prefix is ​​obtained by the consecutive consistent bits of the digest code prefix. The length of the consistent digest prefix is ​​accumulated by the length of the main text, the number of attachments, and the event duration. The difference between the status code and the interaction direction marker is used to obtain the number of payload differences. Batch association markers are obtained based on the same cache hit and flushing. The number of revoked duplicate statistics, profile updates and anomaly detections is counted to obtain the number of repeated calculations and compressions. The number of anomaly judgment results changes is counted to obtain the number of anomaly judgment reversals. The number of tasks that re-enter the statistical update and profile update links after timeline reconstruction is counted to obtain the number of downstream recalculation tasks. The number of master records that enter the statistical update, profile update, anomaly detection and trace writing links is counted to obtain the number of master records released. The preprocessed user social interaction data is stored and an event timeline database is constructed.

[0027] This implementation plan effectively eliminates noise, dimensional differences, and format inconsistencies in the original data by performing cleaning and standardization operations on user social interaction data, including removing illegal and empty records, unifying time units and field definitions, standardizing numerical fields using Z-scores and range normalization, correcting time bases for time fields, and mapping numbered fields. Furthermore, within the same user ID, interaction object ID, session ID, and event type range, candidate event pairs are constructed based on a preset maximum time interval t, and multi-dimensional correlation features such as offset duration, number of consistent core fields, consistent summary prefix length, and number of payload differences are extracted. Combined with cache refresh batch correlation markers, this achieves a refined representation of the implicit temporal relationships and content similarities between events. Simultaneously, by statistically analyzing the number of repeated calculations, anomaly detection flips, downstream recalculation tasks, and master record releases, the performance optimization and stability improvement effects of data preprocessing on downstream statistics, profile updates, and anomaly detection links are quantitatively evaluated. Finally, a high-quality event timeline database with a well-structured, unified definition, and clear correlations is generated, providing a consistent and reliable data foundation for subsequent social interaction behavior analysis, user profile construction, and anomaly pattern recognition.

[0028] Specifically, the steps for performing source merging analysis using time-series offset features, field matching features, and link association features are as follows: Obtain the event occurrence offset duration, event write offset duration, event cloud upload offset duration, core field consistency count, digest prefix consistency length, load difference count, and batch association marker; Square the difference between the event write offset duration and the event occurrence offset duration, and square the difference between the event cloud upload offset duration and the event occurrence offset duration. Summate the two squares and take the square root to obtain the time-series discrete term. This time-series discrete term is used to quantify the comprehensive dispersion of latency deviations in the two stages from event occurrence to write and from event occurrence to cloud upload; directly add the core field consistency count, digest prefix consistency length, and batch association marker to obtain the total field consistency count. Add one to the load difference count as the denominator. This addition operation aims to avoid... To prevent division by zero when the number of load differences is zero, a numerical smoothing penalty term is introduced to mitigate the excessive amplification effect of the field fit ratio in the case of zero differences. The field fit ratio is obtained by calculating the quotient of the total number of consistent fields divided by the denominator. This field fit ratio measures the comprehensive fit level of candidate event pairs in terms of core field matching, summary semantic coherence, and batch association consistency. The asynchronous projection term is obtained by dividing the temporal discrete term by the event occurrence offset duration plus one. The addition of one in the event occurrence offset duration is also used to prevent division by zero and control the numerical stability of the asynchronous projection term when the offset duration approaches zero. Finally, the field fit ratio is divided by the sum of the asynchronous projection term and one. The resulting quotient is the homogeneity merging value. This homogeneity merging value comprehensively reflects the homogeneity similarity of candidate event pairs in multiple dimensions such as temporal deviation, field matching, summary semantics, and batch association. Figure 3The figure shows the spatial distribution surface of the misaligned projection homogeneous merging determination based on temporal discreteness and field alignment in this embodiment. The X-axis represents the event occurrence offset duration (seconds), the Y-axis represents the temporal discreteness term (seconds), and the Z-axis represents the homogeneous merging value. The surface trend intuitively shows that when the event occurrence offset duration is small and the temporal discreteness term is large, the homogeneous merging value is low, and the corresponding candidate event pairs tend to be judged as independent events. When the event occurrence offset duration increases and the temporal discreteness term decreases, the homogeneous merging value increases significantly, and the corresponding candidate event pairs are more easily identified as homogeneous misaligned projection records and merged. This surface figure clearly defines the merging threshold surface, aiding in understanding the spatial response characteristics of the homogeneous merging determination mechanism.

[0029] The specific formula for the homogeneous merge value is as follows:

[0030] In the formula, This represents the homogeneous merging value, reflecting the degree to which candidate events form misaligned projection records in the write link, even though they belong to the same real event. This indicates the number of consistent core fields, reflecting the number of consistent fields among user ID, interaction object ID, session ID, event type, business serial number, text length, number of attachments, and event status code; This indicates the length of the consistent summary prefix, reflecting the number of consecutively consistent bits in the summary code prefixes of the two records. This indicates the batch association flag. It is set to 1 when there is a batch number association with the same cache refresh batch number or the same breakpoint retransmission batch number association, and 0 when there is no association. This indicates the number of payload differences, reflecting the cumulative result of differences in text length, number of attachments, event status codes, and interaction direction markers; It represents the event write offset duration, reflecting the difference between the candidate events and the write time. This indicates the cloud migration offset time of an event, reflecting the difference between the candidate event and the cloud migration time. This indicates the time offset of the event occurrence, reflecting the difference between the time of occurrence of the candidate event.

[0031] In this implementation scheme, a temporal discrete term is calculated by comprehensively considering the temporal deviation between the event write offset time and the event cloud offset time relative to the event occurrence offset time. The field fit ratio is obtained by combining the sum of the core field consistency number, the summary prefix consistency length and the batch association marker, and the smoothed ratio of the number of differences in the same payload plus one. Then, the homogeneous merging value is generated by the ratio of the field fit ratio to the asynchronous projection term after the temporal discrete term is normalized by adding one to the event occurrence offset time. This enables multi-dimensional quantitative evaluation of candidate event pairs in terms of temporal consistency, field matching degree, summary semantic overlap and batch association strength. At the same time, the addition of one to the denominator avoids division by zero and introduces a smoothing penalty, providing a continuous and comparable similarity measurement basis for subsequent homogeneous merging and independent event determination.

[0032] Specifically, the steps for performing master record selection and mirror record attachment, independent event retention, and merge result parameter statistics based on the merge analysis results are as follows: Real-time comparison of the same-source merge value and merge threshold. The same-source merge value refers to the same-source similarity metric score calculated based on multi-dimensional features such as user ID, interaction object ID, session ID, event type, digest code prefix, and proximity of event occurrence times for candidate event pairs. The merge threshold is a pre-set similarity threshold used to determine whether two event records belong to the same original interaction behavior. When the same-source merge value is less than the merge threshold, the independent timeline write path of the current candidate event pair is retained, and the event occurrence offset duration is verified. The criteria for determining whether two events are causally related include: whether the independent separation time is greater than the independent separation time, whether the summary prefix consistency length is less than the summary fit threshold, whether the batch association flag is zero, and whether the number of load differences is greater than the difference threshold. The independent separation time refers to the minimum time interval threshold required to determine if two events have no causal relationship in time sequence. The summary fit threshold refers to the minimum consecutive consistent bit length requirement for determining if the text summary code prefixes of two records have substantial semantic overlap. The batch association flag is a Boolean flag reflecting whether two records in a candidate event pair hit the same cache refresh batch number. The difference threshold refers to determining if two records have significant content bias in load dimensions such as text length, number of attachments, event status code, and interaction direction flag. The criteria for determining the difference threshold are as follows: When the above conditions are met, the two records are treated as independent events; when the above conditions are not met, the current candidate event pair is written into the observation queue and reviewed in the next observation period. If the same-source merge value is still less than the merge threshold after review, it is marked as an independent event record; when the same-source merge value is greater than or equal to the merge threshold, the current candidate event pair is written into the same-source merge cluster, and the master record is selected in the order of the earliest event occurrence time and the largest number of complete fields; when the number of complete fields is the same, the record with the latest event cloud upload time is selected as the master record. This master record selection strategy prioritizes the record version that was captured first in time sequence, has the most complete field information, and is the most stable in the cloud, effectively avoiding the impact of link jitter. To address the issue of missing fields or disordered timing in multi-version data caused by dynamic, breakpoint re-transmission, or cache refresh, which repeatedly triggers downstream statistical updates, profile updates, and anomaly detection links, the system reduces the number of master records allowed to pass through and decreases the number of repeated calculations. Remaining records are marked as mirror records and attached to the master records, and the repeated anomaly counts, repeated profile update tasks, and repeated statistical tasks triggered by mirror records are cancelled. Within the same master record number association range, the number of mirror records and master records is counted based on the mirror record attachment results. The number of incorrectly segmented candidate fragments and the number of consecutive abnormal fragments are counted based on the breakage of candidate event fragments in the order of event occurrence, and these are sent to the event chain integration modeling module along with the same-source merge value.

[0033] In this implementation scheme, by comparing the same-origin merging value and the merging threshold in real time, the system performs homology discrimination and adaptive merging decisions on candidate event pairs. This converges multiple duplicate records generated by the same original interaction behavior into a single master record and its mirror attachment structure. Based on the priority of the earliest event occurrence time, the largest number of complete fields, and the latest cloud upload time, the record version with the most complete information, the most reliable time sequence, and the most stable landing is selected as the master record. The duplicate tasks and counts triggered by the mirror record in the downstream statistical update, profile update, and anomaly detection links are canceled. This reduces the number of master records released from the source, reduces the number of duplicate calculations and anomaly judgment flips, and simultaneously counts the number of mirror records, the number of erroneously segmented candidates, and the number of continuous anomaly segments, providing a high-quality event time sequence foundation after deduplication for event chain integration modeling.

[0034] Specifically, the steps for integrating the merge result parameters with the task queue operation data to perform event chain integration mapping analysis are as follows: The event chain integration mapping process is executed using the same-source merge value, number of mirror records, number of master records, number of candidate missegments, number of consecutive abnormal segments, calculation task queue depth, and calculation task queue capacity as basic inputs; based on the event type, interaction direction marker, source platform number combination, and batch association marker status, valid event records within the current observation window that have completed same-source merge verification, have the same event type, the same interaction direction marker, the same source platform number combination, and consistent batch association marker status are extracted from the event timeline database. A misaligned merge feature is generated, which reflects the proportion of merged ambiguity events caused by the same-source merge threshold boundary judgment. Based on the degree of deviation, a mirror attachment feature is generated, reflecting the distribution of the number of mirror records associated with a single master record and the depth of the attachment hierarchy. A fragmentation feature is also generated, reflecting the number and interval distribution of instances where the same session is mistakenly divided into multiple independent fragments due to sequence breaks at the time of event occurrence. A task queuing feature is generated, reflecting the ratio of the computation task queue depth to the computation task queue capacity within the current observation window and its rate of change. Tensor concatenation is performed on the misalignment merging feature, mirror attachment feature, fragmentation feature, and task queuing feature to construct an event chain integration state matrix. The number of fragments merged, the number of repeated calculations reduced, the number of anomaly detection flips, and the number of downstream recalculation tasks corresponding to the observation window that meets the integration threshold are used as training labels, where fragments... The number of merged segments refers to the number of event segments successfully merged under the integration threshold condition. The number of duplicate calculation reductions refers to the number of duplicate statistics and duplicate profile updates that were revoked due to the merge operation. The number of anomaly judgment reversals refers to the number of times the anomaly judgment result for the same event changed from abnormal to normal before and after merging. The number of downstream recalculation tasks refers to the number of tasks that re-enter the statistical update and profile update chain after timeline reconstruction. Using an increase in the number of merged segments, an increase in the number of duplicate calculation reductions, a decrease in the number of anomaly judgment reversals, and a decrease in the number of downstream recalculation tasks as multi-objective constraints, joint optimization is achieved by quantifying these constraints as penalty terms in the loss function. Specifically, a negative logarithmic term for the number of merged segments and the number of duplicate calculation reductions is set in the loss function to encourage merging. To address the issue of downstream perturbations caused by oversensitivity in merging, a square term is set between the number of anomaly detection flips and the number of downstream recalculation tasks to penalize such perturbations. An L2 regularization term is also introduced to control model complexity. A three-layer multilayer perceptron is used to perform nonlinear mapping on the event chain integration state matrix. The hidden layer dimensions are 128 and 64, respectively. The ReLU activation function is used, and the output layer dimension is 1 with Sigmoid activation to constrain the chain segment integration value to the interval [0,1]. Within the sliding observation window, the model is periodically retrained based on an online incremental learning strategy. Training samples are sampled in batches, and the input features are Z-score standardized. The standardized parameters are updated by the historical data mean and standard deviation of the previous complete observation window. For missing fields, the historical mean of the feature is used to fill them.Construct an event chain integration model and output the chain segment integration value.

[0035] In this implementation scheme, based on the same-source merging value, the number of mirror records, the number of master records, the number of candidate missegments, the number of consecutive abnormal segments, and the depth and capacity of the computation task queue, misalignment merging features, mirror attachment features, segment fragmentation features, and task queuing features are extracted within an observation window where the event type, interaction direction marker, source platform number combination, and batch association marker status are consistent. These features are then tensor-concatenated to form an event chain integration state matrix. Furthermore, a loss function is constructed with constraints such as increasing the number of segment mergings, increasing the number of repeated calculations, decreasing the number of anomaly judgment flips, and decreasing the number of downstream recalculation tasks. A multilayer perceptron is used to perform nonlinear mapping on the state matrix and perform online incremental training. Finally, the chain segment integration value is output, realizing the quantitative evaluation and adaptive optimization decision of the event chain merging and integration effect.

[0036] Specifically, the steps for performing segment merging, task deduplication and folding, and mirror record marking operations based on the integrated mapping analysis results, and outputting the chain segment integration results, are as follows: The chain segment integration value and integration threshold are compared in real time. The chain segment integration value is a comprehensive score output by the event chain integration model to quantify the merging and integration effect of event records within the current observation window. The integration threshold is a pre-set critical value used to determine whether to initiate the reconstruction and merging operation of erroneously segmented data. When the chain segment integration value is less than the integration threshold, it indicates that the current event chain merging state has met the integration requirements. The system maintains the primary record priority processing path, freezes the continuous abnormal segment extension path triggered by the mirror record, and only allows the primary record to participate in abnormal segment extension, thereby avoiding erroneous interference of redundant data from the mirror record to the abnormal segment boundaries. When the chain segment integration value is greater than or equal to the integration threshold, it indicates that the current event chain... There is room for further integration. Within the same user ID and session ID range, candidate segments that are incorrectly segmented can be reordered according to the time of event occurrence. Candidate segments that meet the following criteria are merged into reconstructed abnormal segments: consistent event type, consistent interaction object ID, consistent summary prefix length greater than or equal to the concatenation summary threshold, and event occurrence interval less than or equal to the concatenation duration threshold. The concatenation summary threshold refers to the minimum consecutive consistent number of bits required to determine whether two segments have semantic continuity in the text summary code prefix. The concatenation duration threshold refers to the maximum allowable time interval for determining whether two segments can be connected in time sequence. After merging, only one main computing task is retained in the computing task queue. A main record mapping relationship is established for the mirror record, and the separate triggering of statistical updates, profile updates, and anomaly detection for the mirror record is stopped. This reduces the downstream redundant computing overhead while ensuring the integrity of the abnormal segments.

[0037] In this implementation plan, the event chain merging effect is controlled by comparing the chain segment integration value with the integration threshold in real time: when the chain segment integration value is lower than the integration threshold, the main record is maintained as the priority processing path and the interference of the mirror record on the expansion of abnormal segments is frozen; when the chain segment integration value reaches or exceeds the integration threshold, the candidate segments of missegmentation are rearranged according to the time of event occurrence within the same user ID and session ID range. The segments are merged into reconstructed abnormal segments based on the conditions of consistent event type, consistent interaction object ID, consistent summary prefix length not lower than the splicing summary threshold, and event occurrence interval not exceeding the splicing duration threshold. Only one main calculation task is enqueued, and the mirror record is attached to the main record to prevent it from triggering statistical updates, profile updates, and anomaly detection on its own. This suppresses redundant calculations while ensuring the continuity and integrity of abnormal segments, and further reduces the number of downstream recalculation tasks and the number of repeated calculations.

[0038] Specifically, the steps for time-series stability analysis based on chain segment integration results and task execution status are as follows: Obtain the chain segment integration value, the number of main record releases, the number of repeated calculation reductions, the number of anomaly judgment flips, and the number of downstream recalculation tasks. Before participating in subsequent calculations, each of these quantities undergoes range normalization, mapping it to the interval [0,1] to eliminate dimensional differences between different counting indicators, ensuring that the number of main record releases, the number of repeated calculation reductions, the number of anomaly judgment flips, and the number of downstream recalculation tasks are numerically comparable and dimensionless enough for composite calculations. The normalized number of main record releases is incremented by one and squared; the normalized number of repeated calculation reductions is incremented by one and squared; the two squared values ​​are summed and the square root is taken to obtain the main chain recovery term. This main chain recovery term quantifies the combined recovery strength of the main record throughput efficiency and the repeated calculation suppression effect. A larger value indicates a more effective recovery and reduction of the core business chain. Significantly, the number of anomaly detection flips after normalization is squared, and the number of downstream recalculation tasks after normalization is squared. The square root of the sum of the two squared values ​​is then used to obtain the residual disturbance term. The residual disturbance term is used to measure the degree of residual fluctuation in anomaly detection stability and downstream link computational load caused by the merging operation and timeline reconstruction. The smaller the value, the lower the interference caused by the system to anomaly detection and downstream links. The residual disturbance term is divided by the chain segment integration value plus one to obtain the disturbance reduction term. The addition of one in the chain segment integration value is intended to prevent division by zero and smooth the excessive amplification effect when the chain segment integration value is small. Finally, the main chain recovery term is divided by the sum of the disturbance reduction term and one. The quotient value is the timeline stability value. The timeline stability value comprehensively reflects the balance between the core link recovery strength and the residual disturbance suppression effect of the system after event chain integration processing. The higher the value, the more stable and controllable the overall timeline operation is.

[0039] The specific formula for the timeline stationary value is as follows:

[0040] ;

[0041] In the formula, This represents the stable value of the timeline, reflecting the degree to which the main record timeline has recovered stability after mirror record erasure, fragment merging, and duplicate calculation reduction. This indicates the number of master records released, reflecting the number of master records entering the downstream normal processing chain within the current observation window; This indicates the number of times duplicate calculations were reduced, reflecting the number of duplicate statistics, duplicate profile updates, and duplicate anomaly detections that were undone after the main record folded task. This indicates the number of times the anomaly determination has been flipped, reflecting the number of times the anomaly determination result has been flipped within the same master record's associated range; This indicates the number of downstream recalculation tasks, reflecting the number of statistical tasks and profile update tasks that are retried due to timeline reconstruction. This represents the chain segment integration value, reflecting the integration result output by the event chain integration model.

[0042] Table 1 shows the timeline steady state quantification evaluation parameters in this embodiment. The first test sample had a chain integration value of 2.0, 10 master records were released, 9 repeated calculations were reduced, 10 anomaly detection flips were made, and 9 downstream recalculation tasks were performed. The calculated timeline stability value was 2.71. The second test sample had a chain integration value of 2.5, 12 master records were released, 11 repeated calculations were reduced, 9 anomaly detection flips were made, and 8 downstream recalculation tasks were performed. The calculated timeline stability value was 3.98. The third test sample had a chain integration value of 4.0, 15 master records were released, 14 repeated calculations were reduced, 12 anomaly detection flips were made, and 11 downstream recalculation tasks were performed. The calculated timeline stability value was 5.15. The fourth test sample had a chain integration value of 3.5, 14 master records were released, 13 repeated calculations were reduced, 11 anomaly detection flips were made, and 10 downstream recalculation tasks were performed. The calculated timeline stability value was 4.77.

[0043] Table 1. Quantitative Evaluation Parameters for Timeline Steady State

[0044]

[0045] like Figure 4 As shown, this is a steady-state quantitative time-series change trend chart of the event timeline provided in this application embodiment. The observation window number is the continuous monitoring interval for the merging of social interaction data from multiple platforms, the integration of event chains, and the verification of time-series stability. The four sets of detection samples correspond to four typical operating states. Combined with the four sets of data in Table 1 and Figure 4As can be seen, the performance of various characteristic parameters of sample 1 is weak, with a timeline stability value of only 2.71, which is below the stability threshold. This corresponds to an unstable and uncontrolled state in the initial stage of the system, where the homogeneous merging analysis is insufficient and the event chain integration is fragmented. The chain segment integration value and task optimization effect of sample 2 gradually become apparent, with the timeline stability value rising to 3.98, which is above the stability threshold. This corresponds to a critical state of initial integration of the event chain and gradual reduction of system disturbances. The synergy of various parameters of sample 3 is significantly improved, with a timeline stability value of 5.15, which is in the high range. This corresponds to a stable operating state with complete integration of the event chain and effective improvement in the number of repeated calculations. The parameter configuration of sample 4 tends to be optimal, with a timeline stability value of 4.77, which remains high and stable. This corresponds to a highly stable and optimal operating state with accurate homogeneous merging analysis and efficient task folding control. Table 1 provides a direct and quantitative comparison of the differences in chain segment integration values, master record release counts, repeated calculation reduction counts, anomaly judgment reversal counts, and downstream recalculation task counts under four operating states. This accurately reveals the strong correlation between the event chain integration effect and the timeline stability value, fully verifying that the multi-feature fusion homogeneous merging analysis, event chain integration mapping, and time series stability control mechanism of this application can effectively reduce system disturbances and avoid repeated calculations and anomaly judgment reversals. Homogeneous mirror record merging optimization and task folding control synergistically improve the operational stability of the event timeline. This provides intuitive data support and scientific basis for homogeneous misalignment identification of multi-platform social interaction data, event chain integration optimization, and timeline instability prevention, effectively ensuring the accuracy and reliability of the entire process of cloud-based user social interaction data management.

[0046] In this implementation plan, by performing dimensionless processing and compound operations on the normalized master record release number, repeated calculation reduction number, anomaly judgment flip number, and downstream recalculation task number, a main chain recovery term reflecting the core link recovery strength and a residual disturbance term measuring the residual fluctuation of the merging operation are constructed respectively. Then, the disturbance reduction term is generated by combining the chain segment integration value and the timeline stability value is calculated. This achieves a comprehensive quantitative evaluation of the system's multi-dimensional balance effect after event chain integration processing in terms of business throughput efficiency, repeated calculation suppression, anomaly judgment stability, and downstream link load disturbance. The higher the value, the more stable and controllable the overall timeline operation is.

[0047] Specifically, the steps for performing timeline recovery, maintenance suppression, and instability alarm control operations based on the time series stability analysis results are as follows: The timeline stability value and stability threshold are compared in real time. The timeline stability value is a quantitative indicator used to evaluate the overall operational stability of the event timeline, calculated by combining the main chain recovery term and residual disturbance term. The stability threshold is a pre-set critical value used to determine whether the timeline has entered a stable convergence state. When the timeline stability value is greater than or equal to the stability threshold, it indicates that the current event chain integration effect has met the requirements for stable system operation. In the metadata table within the main record's association range, the mirror record status field is changed from "maintained" to "unstable". The archived state is maintained, and the index key of the mirror record is pointed to the master record number, thereby solidifying the master-slave merging relationship at the metadata level and blocking the independent triggering path of the mirror record to the downstream link. Master records with anomaly detection flip counts less than or equal to the flip control threshold and the number of downstream recalculation tasks less than or equal to the recalculation control threshold are allowed to enter the statistical link, and master record-driven profile updates, anomaly detection, and trace writes are restored. The flip control threshold refers to the critical value for determining whether the number of anomaly detection result changes of a single master record during the merging process is within an acceptable range. The recalculation control threshold refers to determining whether the number of downstream recalculation tasks triggered by a single master record is within the system's acceptable range. The critical value within the allowable load range; when the stable value of the timeline is less than the stable threshold, it indicates that the timeline has not yet converged to a stable state. The system maintains the primary record priority processing path, allowing only one profile update and one anomaly detection for each primary record within one observation window. Downstream recalculation tasks triggered by mirror records are frozen to prevent disturbance propagation. In the anomaly judgment result table, a lock flag is written to the anomaly judgment result field within the same primary record association range to prevent the anomaly judgment conclusion from being incorrectly referenced by downstream links during unstable periods. The number of new recalculation tasks in each observation window is limited to m, where m is a positive integer dynamically determined based on the computing task queue capacity and the current queue depth. The upper limit value is set; the retention duration field for mirror record mapping relationship, fragment merging result, and merged source record comparison relationship is added to the retention duration configuration table. Candidate fragments that have not yet been verified continue to be retained in the order of event occurrence time, event writing time, and source log number to ensure that merged source data is traceable and verifiable during unstable periods. When the stable value of the timeline of n consecutive observation windows is still less than the stability threshold, where n is the preset threshold for the number of consecutive observation windows, an event timeline instability alarm record is generated. Only the anomaly detection, traceability query, and basic profile update tasks driven by the main record are retained, and data consumption of non-core links is suspended to protect the overall stability of the system.

[0048] In this implementation plan, the stability of the event timeline is controlled in a tiered manner by comparing the timeline stability value with the stability threshold in real time: when the timeline stability value meets the standard, the mirror record status is archived and a master record index is established for association. For master records that meet the flip and recalculation control conditions, the statistical link is allowed to proceed and the profile update, anomaly detection, and trace writing are restored; when the timeline stability value does not meet the standard, the master record priority path is maintained and the single window trigger frequency is strictly limited. Downstream tasks of mirror records are frozen and anomaly judgment results are locked. At the same time, the retention time of merged related records is extended to ensure data traceability; if the standard is not met for multiple consecutive windows, an instability alarm is generated and the system is downgraded to retain only the core anomaly detection and basic query links. This dynamically balances the benefits of data integration and the risks of system operation while ensuring the stability of the timeline.

[0049] Specifically, this embodiment provides a cloud-based user social interaction data management system, applied to a cloud-based user social interaction data management method, including: a multi-platform event collection and preprocessing module, used to collect user social interaction data from multiple platforms through source platform links, interaction content links, link operation records, task queue monitoring links, profile update record links, statistical update record links, anomaly detection record links, and traceability write record links; and sequentially performing illegal and empty record removal, time unit and field caliber unification, Z-score standardization and range normalization processing on numerical fields, time base correction for time fields, and association mapping for number fields on the user social interaction data. After preprocessing operations, candidate event pairs are constructed based on the same user ID, interaction object ID, session ID, and event type within the maximum time interval t. Correlation features such as offset duration, number of consistent core fields, consistent summary prefix length, and number of payload differences are extracted. Combined with statistical results of cache flushing batch association markers, repeated calculation reduction times, anomaly judgment flip times, downstream recalculation task counts, and master record release counts, a well-structured event timeline database is finally generated. A homogeneous misalignment merging identification module is used to calculate homogeneous merging values ​​based on candidate event pairs in the event timeline database. Homogeneous merging analysis is performed by comparing homogeneous merging values ​​with merging thresholds in real time. When homogeneous merging occurs... When the value is less than the merging threshold, the event offset duration, summary prefix consistency length, batch association markers, and payload differences are checked to determine independent events or added to the observation queue for review. When the homogeneous merging value is greater than or equal to the merging threshold, candidate event pairs are grouped into homogeneous merging clusters, and the master record is selected according to the priority of the earliest event occurrence time, the largest number of complete fields, and the latest event cloud upload time. The remaining records are marked as mirror records and attached to the master record. The duplicate anomaly count, duplicate profile update task, and duplicate statistics task triggered by the mirror record are simultaneously canceled. The number of mirror records, the number of master records, the number of erroneously segmented candidates, and the number of continuous anomaly segments are statistically output. The event chain integration modeling module is used to determine the homogeneous merging value, Based on the number of mirror records, the number of master records, the number of candidate missegments, the number of consecutive abnormal segments, the depth of the computation task queue, and the capacity of the computation task queue, the system extracts misalignment merging features, mirror attachment features, segment fragmentation features, and task queuing features within an observation window where the event type, interaction direction marker, source platform number combination, and batch association marker status are consistent. These features are then tensor-concatenated to form an event chain integration state matrix. A multilayer perceptron is used to perform nonlinear mapping on the state matrix. A loss function is constructed with constraints such as increasing the number of segment merging, increasing the number of repeated calculations, decreasing the number of anomaly judgment flips, and decreasing the number of downstream recalculation tasks. Online incremental training is then performed, and the chain segment integration value is output as the chain segment integration result.The timeline stability verification module calculates the timeline stability value based on the chain segment integration value, the number of master record releases, the number of repeated calculation reductions, the number of anomaly judgment flips, and the number of downstream recalculation tasks. It performs time-series stability analysis by comparing the timeline stability value with a stability threshold in real time. When the timeline stability value is greater than or equal to the stability threshold, it performs mirror record status archiving, master record index association, release of statistical links meeting control conditions, profile updates, anomaly detection, and trace write recovery operations. When the timeline stability value is less than the stability threshold, it maintains the master record priority path, limits the single-window trigger frequency, freezes downstream tasks of mirror records, locks anomaly judgment results, and extends the retention time of merged records. If multiple consecutive windows fail to meet the standard, it generates an event timeline instability alarm and downgrades to retaining only the core anomaly detection and basic query links.

[0050] In this implementation plan, a multi-platform event acquisition and preprocessing module cleans, standardizes, and extracts related features from user social interaction data from source platform links, interactive content links, link operation records, task queue monitoring links, profile update record links, statistical update record links, anomaly detection record links, and traceability write record links to construct an event timeline database. A same-source misalignment merging and identification module selects main records and attaches mirror records based on same-source merging values ​​and merging thresholds, and counts the number of mirror records, main records, candidate missegments, and consecutive abnormal segments. An event chain integration and modeling module utilizes same-source merging values ​​and mirror records... The event chain integration state matrix is ​​constructed based on the number of records, the number of master records, the number of candidate erroneous segments, the number of consecutive abnormal segments, and the depth and capacity of the computation task queue. The chain segment integration value is output through online training using a multilayer perceptron. The timeline stability verification module calculates the timeline stability value based on the chain segment integration value, the number of master records released, the number of repeated calculations reduced, the number of anomaly judgment flips, and the number of downstream recalculation tasks. It then performs timeline recovery, maintenance suppression, and instability alarm control operations with a stability threshold as the boundary. This ensures the stability of the event timeline while achieving deduplication and merging of cross-platform social interaction data, reconstruction of abnormal segments, and dynamic balance management of downstream computing resources.

[0051] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the present invention. These examples are merely for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or variations without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, variations, or combinations, or the direct application of the concept and technical solution of the present invention to other situations without modification, should all be considered within the scope of protection of the present invention.

Claims

1. A cloud computing-based method for managing user social interaction data, characterized in that, Includes the following steps: S1 collects user social interaction data from multiple platforms, preprocesses the user social interaction data, stores it, and builds an event timeline database; S2 performs source merging analysis based on time-series offset features, field matching features, and link association features. Based on the merging analysis results, it performs operations such as master record selection, mirror record attachment, independent event retention, and merging result parameter statistics. S3 integrates the merged result parameters with the task queue running data to perform event chain integration mapping analysis. Based on the integration mapping analysis results, it performs fragment merging, task deduplication and folding, and mirror record marking operations, and outputs the chain segment integration results. S4 performs time-series stability analysis based on the chain segment integration results and task execution status, and performs timeline recovery, maintenance suppression, and instability alarm control operations according to the time-series stability analysis results.

2. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for collecting user social interaction data from multiple platforms are as follows: Collect user social interaction data: Obtain user ID, interaction object ID, session ID, event type, original event ID, business serial number, source platform ID, source node ID, event occurrence time, event write time, and event cloud upload time through the source platform link; obtain text length, text digest code, number of attachments, interaction direction marker, and event status code through the interaction content link; obtain cache refresh batch number, breakpoint retransmission batch number, number of retries, node cache dwell time, and source log sequence number through the link operation record; obtain the complete number of fields by counting the total number of non-empty core fields in a single record; obtain the computation task ID, task enqueue time, task dequeue time, task status marker, task trigger source marker, computation task queue depth, and computation task queue capacity through the task queue monitoring link; The image update record link obtains the image update number, image update time, master record number, and image update trigger source marker; the statistical update record link obtains the statistical update number, statistical update time, master record number, and statistical update trigger source marker; the anomaly detection record link obtains the anomaly detection number, anomaly detection time, anomaly judgment result, master record number, and anomaly detection trigger source marker; and the trace write record link obtains the trace write number, trace write time, master record number, and trace write trigger source marker.

3. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for preprocessing and storing user social interaction data and constructing an event timeline database are as follows: Illegal and empty records in user social interaction data are removed. Time units, field names, status code definitions, and text digest code lengths are standardized. Numerical fields are mapped to a unified interval using Z-score standardization and range normalization. The time base of time fields is corrected, and the mapping of number fields is associated. The definitions of text digest codes, event status codes, and interaction direction markers are consolidated. Within the same user ID, interaction object ID, session ID, and event type range, candidate event pairs are established based on the maximum time interval t. The corresponding offset duration is obtained based on the difference between the event occurrence, writing, and cloud upload times. Each item is compared to obtain the number of consistent core fields, and the digest prefix is ​​obtained based on the number of consecutive consistent bits in the digest code prefix. To determine the length, the difference between the main text length, the number of attachments, the event status code, and the interaction direction marker is accumulated to obtain the payload difference quantity. Batch association markers are obtained based on the same cache hit and flushing. The number of revoked duplicate statistics, profile updates, and anomaly detections is counted to obtain the number of duplicate calculations and reductions. The number of anomaly judgment results changes is counted to obtain the number of anomaly judgment reversals. The number of tasks that re-enter the statistical update and profile update links after timeline reconstruction is counted to obtain the number of downstream recalculation tasks. The number of master records that enter the statistical update, profile update, anomaly detection, and trace write links is counted to obtain the number of master records released. The preprocessed user social interaction data is stored and an event timeline database is constructed.

4. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for performing source merging analysis using temporal offset features, field matching features, and link association features are as follows: Get the event occurrence offset duration, event write offset duration, event cloud upload offset duration, number of consistent core fields, length of consistent summary prefix, number of payload differences, and batch association markers; Subtract the square of the event occurrence offset from the event write offset duration, and subtract the square of the event occurrence offset duration from the event cloud upload offset duration. Summing these values ​​and taking the square root yields the time series discrete term. The field fit ratio is obtained by summing the core field consistency count, the summary prefix consistency length, and the batch association marker, and then dividing by the number of load differences plus one. To avoid division by zero due to zero load differences and to implement a smoothing penalty, the time-series discrete term is divided by the event occurrence offset duration plus one to obtain the asynchronous projection term. Divide the field fit ratio by the sum of the asynchronous projection item and one to obtain the same-origin merge value.

5. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for performing master record selection and mirror record attachment, independent event retention, and merge result parameter statistics based on the merge analysis results are as follows: By comparing the source merging value and the merging threshold in real time, when the source merging value is less than the merging threshold, the independent timeline write path of the current candidate event pair is retained, and the following checks are performed: whether the event offset duration is greater than the independent separation duration, whether the summary prefix consistency length is less than the summary fit threshold, whether the batch association flag is zero, and whether the number of load differences is greater than the difference threshold. When the above conditions are met, the two records are treated as independent events. When the above conditions are not met, the current candidate event pair is written to the observation queue, and a review is performed in the next observation period. If the source merging value is still less than the merging threshold after the review, it is marked as an independent event record. When the same source merge value is greater than or equal to the merge threshold, the current candidate event pair is written into the same source merge cluster, and the master record is selected in the order of the earliest occurrence time of the event and the largest number of complete fields; when the number of complete fields is the same, the record with the latest event cloud time is selected as the master record; the remaining records are marked as mirror records and attached to the master record, and the duplicate anomaly count, duplicate profile update task and duplicate statistics task triggered by the mirror record are canceled; Within the same master record number association range, the number of mirror records and master records are counted based on the mirror record attachment results. The number of erroneously segmented candidates and the number of consecutive abnormal segments are counted based on the breakage of candidate event segments in the order of event occurrence. These are then sent to the event chain integration modeling module along with the same source merging value.

6. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for integrating and mapping the comprehensive merging result parameters with the task queue running data into an event chain are as follows: The event chain integration mapping process is executed based on the same-source merge value, the number of mirror records, the number of main records, the number of candidates for incorrect segmentation, the number of consecutive abnormal segments, the depth of the computation task queue, and the capacity of the computation task queue. Based on the event type, interaction direction marker, source platform number combination, and batch association marker status, valid event records within the current observation window that have completed source merging verification, share the same event type, interaction direction marker, source platform number combination, and batch association marker status are extracted from the event timeline database. These records generate misalignment merging features, mirror attachment features, fragment splitting features, and task queuing features. Tensor concatenation is performed on these features to construct an event chain integration state matrix. The number of merged fragments, the number of repeated calculations reduced, the number of anomaly judgment flips, and the number of downstream recalculation tasks corresponding to the observation window that meet the integration threshold are used as training labels. Constraints are set to increase the number of merged fragments, increase the number of repeated calculations reduced, decrease the number of anomaly judgment flips, and decrease the number of downstream recalculation tasks. A multilayer perceptron is used to perform nonlinear mapping on the event chain integration state matrix to construct an event chain integration model and output the chain segment integration value.

7. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for performing fragment merging, task deduplication and folding, and mirror record marking operations based on the integrated mapping analysis results, and outputting the chain segment integration results are as follows: By comparing the chain segment integration value and the integration threshold in real time, when the chain segment integration value is less than the integration threshold, the primary record is maintained as the priority processing path, the continuous abnormal segment extension path triggered by the mirror record is frozen, and only the primary record is allowed to participate in the abnormal segment extension. When the segment integration value is greater than or equal to the integration threshold, the erroneously segmented candidate segments are reordered according to the time of event occurrence within the same user ID and the same session ID range. Candidate segments with the same event type, the same interaction object ID, the same summary prefix length greater than or equal to the splicing summary threshold, and the event occurrence interval less than or equal to the splicing duration threshold are merged into reconstructed abnormal segments; only one main computing task is retained and entered into the computing task queue. Establish a master record mapping relationship for mirror records, and stop mirror records from triggering statistical updates, profile updates, and anomaly detection independently.

8. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for performing time-series stability analysis based on chain segment integration results and task execution status are as follows: Obtain the chain segment integration value, the number of main record releases, the number of repeated calculations and reductions, the number of anomaly judgment flips, and the number of downstream recalculation tasks; add one square to the number of main record releases and one square to the number of repeated calculations and reductions, sum them up and take the square root to obtain the main chain recovery item; perform a square operation on the number of anomaly judgment flips and the number of downstream recalculation tasks, sum them up and take the square root to obtain the residual disturbance item. Divide the residual perturbation term by the chain segment integration value plus one to obtain the perturbation reduction term; divide the main chain recovery term by the sum of the perturbation reduction term and one to obtain the timeline stability value.

9. The user social interaction data management method based on cloud computing according to claim 1, characterized in that: The specific steps for performing timeline recovery, suppression maintenance, and instability alarm control operations based on the time series stability analysis results are as follows: By comparing the stable value and the stable threshold of the timeline in real time, when the stable value of the timeline is greater than or equal to the stable threshold, the status field of the mirror record is changed from the "hold" status to the "archived" status in the metadata table within the scope of the master record association, and the index key of the mirror record is pointed to the master record number; master records with anomaly judgment flip count less than or equal to the flip control threshold and downstream recalculation task count less than or equal to the recalculation control threshold are allowed to enter the statistical link, and the profile update, anomaly detection and trace writing driven by the master record are restored; When the timeline stability value is less than the stability threshold, maintain the primary record priority processing path, allow each primary record to trigger one profile update and one anomaly detection within one observation window, freeze the downstream recalculation tasks triggered by the mirror record, write a lock flag to the anomaly judgment result field within the same primary record association range in the anomaly judgment result table, and limit the number of new recalculation tasks in each observation window to m. In the retention duration configuration table, add a short field to the retention duration field for mirror record mapping relationship, fragment merging result, and merged source record comparison relationship. Continue to retain candidate fragments that have not yet been verified in the order of event occurrence time, event write time, and source log sequence number. When the stable value of the timeline for n consecutive observation windows is still less than the stability threshold, an event timeline instability alarm record is generated, and only the anomaly detection, trace query and basic profile update tasks driven by the main record are retained.

10. A cloud-based user social interaction data management system, applied to the cloud-based user social interaction data management method described in any one of claims 1-9, characterized in that, include: The multi-platform event collection and preprocessing module is used to collect user social interaction data from multiple platforms, preprocess the user social interaction data, store it, and build an event timeline database. The same-source misalignment merging identification module is used to perform same-source merging analysis based on time-series offset features, field matching features, and link association features. Based on the merging analysis results, it performs operations such as master record selection, mirror record attachment, independent event retention, and merging result parameter statistics. The event chain integration modeling module is used to perform event chain integration mapping analysis by combining the merged result parameters with the task queue running data. Based on the integration mapping analysis results, it performs segment merging, task deduplication and folding, and mirror record marking operations, and outputs the chain segment integration results. The timeline stability verification module is used to perform timeline stability analysis based on the chain segment integration results and task execution status, and to perform timeline recovery, maintenance suppression, and instability alarm control operations according to the timeline stability analysis results.

Images