A cloud-based method for scheduling and processing mediation tasks.

Abstract

This invention discloses a cloud-based method for scheduling and processing mediation tasks. The method includes: acquiring multi-source dispute data through a heterogeneous data receiving gateway; converting the dispute data into a graph structure mapping by calling a graph computation engine; establishing a global evidence conflict topology graph and calculating the conflict density value of global tasks; extracting interrupt signal vectors from terminal devices using the interactive interface acquisition probe of regular mediation nodes and generating multi-dimensional time-series operation features; retrieving historical baseline data from a situational awareness scheduling bus to generate a dynamic degradation index; combining the conflict density value with resource anti-congestion scheduling allocation and task distribution; performing partial derivative operations on regular mediation nodes to obtain the deterioration rate of the dynamic degradation index; stripping dense mutually exclusive edge regions when the deterioration rate exceeds the sudden increase boundary value; forwarding the stripped high-density topology subgraph to expert processing nodes for rights confirmation; and injecting execution receipts in reverse to complete the assembly and writing back of log status.

Description

Technical Field

[0001] This invention relates to the field of cloud computing and data processing technology, specifically to a method for scheduling and processing mediation tasks based on a cloud platform. Background Technology

[0002] With the rapid development of e-commerce, the number of online transaction disputes continues to rise. Online dispute resolution systems typically rely on cloud platforms for centralized collection and distribution of tasks. The system transfers data such as interactive texts between buyers and sellers, transaction orders, and supporting documents transmitted through the front-end gateway to the processing center, where it is distributed to various mediation nodes connected to the platform for data processing and rights confirmation decisions.

[0003] In existing online mediation scenarios, task scheduling primarily relies on static load balancing strategies. Cloud servers typically push newly received dispute tasks to online mediation nodes in a polling manner, based on the order in which tasks entered the queue or according to basic business category rules. During execution, the system directly issues unstructured natural language complaint text and attachments, relying on the computing resources of the mediation nodes for text loading and logical comparison. Ultimately, the system collects the nodes' judgment instructions to trigger fund allocation and order status transitions.

[0004] When faced with complex dispute data, existing scheduling and processing methods suffer from a technical deficiency: they cannot dynamically match system computing resources with the actual processing difficulty of unstructured dispute data. Because existing mediation systems often struggle to accurately assess the complexity of logical contradictions in evidence, they cannot accurately predict the computational resource consumption of dispute data on nodes before task distribution. When computationally intensive tasks containing complex evidence conflicts are round-robin assigned to regular mediation nodes, it easily leads to data processing overload on local nodes. Simultaneously, existing systems lack a real-time monitoring link that directly feeds back the load of human-computer interaction behavior at node ends, relying solely on delayed responses at the application layer for passive intervention, failing to trigger congestion prevention scheduling before node operators experience processing overload. Furthermore, when dealing with node congestion, existing systems cannot decouple complex evidence conflict verification tasks from regular processing flows, nor can they logically separate complex evidence cores that cause overload, resulting in unrelieved single-point computational pressure and hindering the overall data processing timeliness and system stability of mediation tasks. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a cloud-based method for scheduling and processing mediation tasks, which addresses the issues of resource allocation imbalance and scheduling response lag caused by computational overload in existing mediation systems when dealing with complex dispute data.

[0006] This invention provides a cloud-based method for scheduling and processing mediation tasks and data, comprising the following steps:

[0007] Step S100: Obtain multi-source dispute data from external input through the heterogeneous data receiving gateway, call the graph computing engine to convert the multi-source dispute data into a graph structure mapping, establish a global evidence conflict topology graph, and calculate the conflict density value of the global task.

[0008] Step S200: The interactive interface acquisition probe of the conventional mediation node calls the system standard input event API interface to extract the underlying input event vector of the terminal device at a preset sampling frequency. The underlying input event vector is processed locally to generate multi-dimensional time-series operation features, and is pushed unidirectionally to the core data scheduling server cluster through the asynchronous message communication interface.

[0009] Step S300: The situational awareness scheduling bus retrieves the historical baseline data of the corresponding regular mediation node from the time series database, compares and calculates the data using multi-dimensional time series operation features, generates the dynamic degradation index of the regular mediation node, and combines the conflict density value of each task in the task queue to perform resource anti-congestion scheduling and allocation, and distributes the task instructions to the regular mediation node with the matching status.

[0010] In step S400, the situational awareness scheduling bus performs time partial derivative calculation on the regular mediation node in the execution state to obtain the deterioration rate of the dynamic degradation index of the regular mediation node. When it is determined that the deterioration rate of the dynamic degradation index exceeds the sudden increase boundary value configured by the system, a state freeze command is sent to the regular mediation node and an architecture reorganization command is sent to the graph computing engine. The graph computing engine cuts off and logically strips the dense mutually exclusive edge region in the evidence conflict topology graph corresponding to the current task.

[0011] In step S500, the core data scheduling server cluster packages the high-density topology subgraph obtained by stripping into independent network data packets, forwards them across logically isolated network segments to the expert processing node, the expert processing node performs comprehensive verification and confirmation of rights for multi-dimensional related evidence, and injects the execution receipt with attached status modification mark in reverse, and the core data scheduling server cluster completes the final assembly and write-back operation of the mediation log status.

[0012] Furthermore, the method for calculating the conflict density value of the global task is as follows: count all nodes in the underlying evidence conflict topology graph and remove isolated nodes, and count the remaining nodes with connections as the total number of effective basic nodes; extract the total number of mutually exclusive edges, divide the total number of mutually exclusive edges by the total number of effective basic nodes to obtain the ratio, and multiply the ratio by the preset structural coefficient to obtain the first product; read the conflict weight attribute bound to each mutually exclusive edge and perform linear summation to calculate the weighted conflict severity of the current mediation task, multiply the topology conflict connectivity by the extracted structural coefficient, divide the weighted conflict severity by the total number of mutually exclusive edges for normalization, multiply by the extracted weight coefficient, and add the two products to calculate and output the final conflict density value.

[0013] Furthermore, the calculation method for the dynamic degradation index of the regular mediation node is as follows: retrieve the standard baseline distribution model formed by the corresponding regular mediation node in the historical stable operation period from the time series database, and extract the historical mean and historical standard deviation of the response delay increment, text input rate variance, and modification frequency; use the standard score calculation method to convert the actual collected feature values ​​into dimensionless relative offsets; multiply the relative offsets of the response delay increment, the relative offsets of the text input rate variance, and the relative offsets of the modification frequency by their respective bound feature penalty weight factors and sum them up to obtain the dynamic degradation index of the regular mediation node.

[0014] Furthermore, the dynamic deterioration rate is calculated as follows: during the operation of the regular mediation node, a high-frequency sampling timer is configured to extract the real-time dynamic degradation index, and the backward difference algorithm is used for calculation; the real-time dynamic degradation index at the current moment is subtracted from the real-time dynamic degradation index at the previous sampling moment to obtain the index difference value, and then the aforementioned index difference value is divided by the sampling time interval of the high-frequency sampling timer to calculate and obtain the dynamic deterioration rate of the regular mediation node.

[0015] Furthermore, the method for correcting and updating the feature penalty weight factor is as follows: extract the actual deviation direction of the multi-dimensional temporal operation features within the sliding time window before the state freeze command is triggered by the regular mediation node, calculate the first-order difference value at the state freeze critical moment, and identify the feature with the largest absolute value of the first-order difference value as the dominant feature dimension that causes the dynamic degradation exponential to surge; divide the dynamic deterioration rate recorded at the state freeze critical moment by the surge boundary value configured by the system and take the absolute value, multiply the aforementioned absolute value by the weight learning rate parameter configured by the system to obtain the correction compensation amount; add the feature penalty weight factor before the update to the correction compensation amount to complete the update of the feature penalty weight factor corresponding to the dominant feature dimension.

[0016] The present invention, by adopting the above technical solution, can bring the following beneficial effects:

[0017] 1. This invention acquires multi-source dispute data through a heterogeneous data receiving gateway and calls a graph computing engine to perform dependency parsing to extract disputed entities and constraint claims. Based on the semantic similarity comparison of text vectors, it identifies logical contradictions and constructs mutually exclusive edges. Then, it uses the influence of funds to quantify and assign weights to the mutually exclusive edges to establish an underlying evidence conflict topology graph. This realizes the transformation of dispute facts at the natural language level into measurable and calculable graph network topology features, providing a numerical evaluation benchmark for the system's task concurrency control and resource anti-congestion scheduling allocation.

[0018] 2. This invention intercepts hardware interrupt signals at the operating system kernel level of a conventional mediation node to statistically generate response latency increments, text input rate variance, and modification frequency. It then compares the collected multi-dimensional time-series operation features with historical benchmark distribution models in a time-series database to generate a dynamic degradation index that reflects the terminal's execution capabilities in real time. This establishes a monitoring link that directly feeds back human-computer interaction behavior characteristics, thereby enabling real-time quantification and tracking of the operation and processing load of distributed processing nodes and triggering anti-congestion scheduling and allocation before resource congestion occurs.

[0019] 3. This invention obtains the deterioration rate of node operating status by performing differential approximate derivative calculation on the dynamic degradation index. When the deterioration rate exceeds the sudden increase boundary value, density clustering analysis is used to lock the high-density conflict region in the underlying evidence conflict topology graph and perform logical stripping. The stripped high-complexity topology subgraph is routed as an independent data micro-task to the expert processing node. This achieves structural decoupling between computationally intensive evidence conflict verification and routine mediation business processes, avoids system congestion of a single node caused by complex dispute evidence, and ensures the timeliness of overall task data processing. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the cloud-based mediation task scheduling and data processing system of the present invention.

[0021] Figure 2 This is the overall flowchart of the mediation task scheduling and data processing method of the present invention;

[0022] Figure 3 This is a flowchart of the construction and density quantization calculation of the underlying evidence conflict topology graph of this invention;

[0023] Figure 4 This is a flowchart of the node anti-congestion scheduling and cooling compensation allocation mechanism of the present invention;

[0024] Figure 5 This is a flowchart of the closed-loop process for high-density conflict kernel logic stripping and cross-domain state assembly in this invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1

[0027] See attached document Figure 1-2 This invention provides a cloud platform-based method for scheduling and processing mediation tasks and data, comprising the following steps:

[0028] In step S100, the core data scheduling server cluster 110 acquires externally input multi-source dispute data through the heterogeneous data receiving gateway 111. The core data scheduling server cluster 110 calls the graph operation engine 112 to convert the multi-source dispute data into a graph structure mapping, establish a global evidence conflict topology graph, and calculate the conflict density value of the global task. The formula for calculating the conflict density value by the graph operation engine 112 is as follows:

[0029]

[0030] In the formula, This represents the conflict density value. This represents the total number of mutually exclusive edges in the evidence conflict topology graph. This represents the total number of valid basic nodes obtained through filtering. For the first The conflict weight of a mutually exclusive edge; These are dimensionless, pre-defined structural coefficients. These are dimensionless preset weighting coefficients;

[0031] Step S200: The interactive interface acquisition probe of the conventional mediation node calls the system standard input event API interface to extract the underlying input event vector of the terminal device at a preset sampling frequency. The underlying input event vector is processed locally to generate multi-dimensional time-series operation features, and is pushed unidirectionally to the core data scheduling server cluster 110 through the asynchronous message communication interface.

[0032] In step S300, the situational awareness scheduling bus 113 retrieves the historical baseline data of the corresponding conventional mediation node 200 from the time series database 123, performs comparison calculation using multi-dimensional time series operation features, generates a dynamic degradation index for the conventional mediation node 200, and combines the conflict density value of each task in the task queue to perform resource anti-congestion scheduling allocation, distributing task instructions to the conventional mediation node 200 with matching status.

[0033] In step S400, the situational awareness scheduling bus performs time partial derivative calculations on the regular mediation node in the execution state to obtain the deterioration rate of the dynamic degradation index of the regular mediation node. When it is determined that the deterioration rate of the dynamic degradation index exceeds the burst boundary value configured by the system, a state freeze command is sent to the regular mediation node, and an architecture reorganization command is sent to the graph computing engine. The graph computing engine cuts off and logically strips the dense mutually exclusive edge regions in the evidence conflict topology graph corresponding to the current task.

[0034] In step S500, the core data scheduling server cluster packages the high-density topology subgraph obtained by stripping into independent network data packets, forwards them across logically isolated network segments to the expert processing node, the expert processing node performs comprehensive verification and confirmation of rights for multi-dimensional related evidence, and injects the execution receipt with attached status modification mark in reverse, and the core data scheduling server cluster completes the final assembly and write-back operation of the mediation log status.

[0035] Furthermore, the implementation of the above method relies on the distributed cloud platform 100 architecture to build a mediation task scheduling and data processing system 10. The cloud platform 100 includes a core data scheduling server cluster 110, which integrates a heterogeneous data receiving gateway 111, a graph computing engine 112, and a situational awareness scheduling bus 113.

[0036] The core data scheduling server cluster 110 has a persistent storage matrix 120 for communication connection. In order to support the isolated flow of multi-dimensional data states, the persistent storage matrix 120 specifically includes an object storage server 121 for storing unstructured credential files, a distributed graph database 122 for carrying evidence conflict topology graphs, and a time series database 123 configured to record node operation feature parameters.

[0037] The access layer of the mediation task scheduling and data processing system 10 is deployed with multiple Class I terminal devices, defined as regular mediation nodes 200. The regular mediation nodes 200 are distributed in an isolated physical network environment and access the core data scheduling server cluster 110 through a secure socket layer virtual private network. The regular mediation nodes 200 are configured with an interactive interface acquisition probe 210. The interactive interface acquisition probe 210 calls the standard input event API interface (such as the Raw Input mechanism) provided by the operating system to capture the underlying input event vector generated by peripheral device interaction. This mechanism can effectively bypass the data forgery of the application layer virtual sandbox environment and the lag and dead state of the software application layer, and directly obtain the underlying hardware response latency that reflects the actual interactive load of the operator.

[0038] The access layer of the mediation task scheduling and data processing system 10 also deploys multiple second-type terminal devices, defined as expert processing nodes 300. The system permission level of expert processing nodes 300 is higher than that of regular mediation nodes 200 in the logical domain. Expert processing nodes 300 are independent server clusters with specific domain automated graph rights confirmation models deployed internally. Expert processing nodes 300 reside in the intranet background transfer pool of cloud platform 100 and are specially configured to receive data micro-tasks generated by graph computing engine 112.

[0039] Example 2

[0040] See attached document Figure 3-5 In step S100, after the cloud platform 100 receives the interactive text, transaction order data, and voucher attachments uploaded by the buyer and seller, the core data scheduling server cluster 110 calls the internal graph processing engine 112 to perform graph mapping of the unstructured data. The specific steps are as follows:

[0041] In step S101, the heterogeneous data receiving gateway 111 performs format conversion on the received multi-source heterogeneous dispute data, and the core data scheduling server cluster 110 summarizes and merges the converted interactive text, transaction order data and image extraction text into a one-dimensional text sequence.

[0042] In step S102, the graph processing engine 112 performs entity extraction and relation mapping on the one-dimensional text sequence. The graph processing engine 112 has a pre-trained sequence labeling model deployed inside. The system inputs the one-dimensional text sequence into the sequence labeling model and uses word boundary marking to assign entity category labels to each character in the one-dimensional text sequence. Based on this, it identifies the disputed entities in the dispute and the constraint claims made by the buyer and seller against the disputed entities. The disputed entities include specific parts of the goods, logistics flow nodes, and transaction amount details. The constraint claims include descriptions of the goods' status, logistics time node determination results, and claims for amount reduction. The graph processing engine 112 calls the dependency parsing algorithm to extract the subject-verb-object grammatical dependencies between the disputed entities and the constraint claims in the text statement. It pairs the disputed entities with direct grammatical dependencies with the constraint claims and outputs entity-claim data pairs.

[0043] In step S103, the graph operation engine 112 constructs a basic topology based on entity-claim data pairs. The graph operation engine 112 instantiates the identified disputed entities and constraint claims into a set of nodes in the topology space. Based on the pairing relationship of entity-claim data pairs, the graph operation engine 112 establishes directed association edges between the corresponding disputed entity nodes and constraint claim nodes, generating a set of regular edges. The set of nodes and the set of regular edges together form the basic evidence topology graph.

[0044] In step S104, the graph operation engine 112 performs logical verification on the basic evidence topology graph and generates mutually exclusive edges. The graph operation engine 112 traverses the disputed entity nodes in the basic evidence topology graph and retrieves different constraint claim nodes connected to the same disputed entity node. The graph operation engine 112 extracts the text vector features of different constraint claim nodes. Specifically, it calls a pre-trained word embedding model to map the natural language characters of different constraint claim nodes into fixed-dimensional high-dimensional floating-point vectors, and then calculates the semantic cosine similarity between the high-dimensional floating-point vectors of different constraint claim nodes. The graph operation engine 112 has a built-in contradictory semantic feature dictionary. When different constraint claim nodes hit the contradictory semantic feature dictionary and the calculated semantic cosine similarity is lower than the logical consistency judgment threshold set by the system, the system determines that there is an irreconcilable logical conflict for the corresponding disputed entity node. The graph operation engine 112 generates mutually exclusive edges between the corresponding different constraint claim nodes and establishes a set of mutually exclusive edges.

[0045] In step S105, the graph processing engine 112 calculates the conflict weight of each mutually exclusive edge in the set of mutually exclusive edges. The system quantifies and assigns weights based on the financial influence of the disputed entities corresponding to the mutually exclusive edges. The calculation model for the conflict weight of a single mutually exclusive edge is as follows:

[0046]

[0047] In the formula, For the first The conflict weight of a mutually exclusive edge; For the first The local amount of the disputed entity directly corresponding to each mutually exclusive edge; This is the cumulative total transaction amount for the current mediation task. To match the semantic opposition level coefficients output by the contradictory semantic feature dictionary to the graph processing engine 112, the system transforms the discrete levels into a normalized mapping function. Continuous floating-point values ​​within a range; and These are all system adjustment parameters pre-configured by the graph processing engine 112, and their value ranges are all limited by the system. Within the range;

[0048] The graph processing engine 112 merges the set of mutually exclusive edges and the conflict weight parameters carried by the set of mutually exclusive edges into the basic evidence topology graph, completes the construction of the underlying evidence conflict topology graph of the current mediation task, and transmits the underlying evidence conflict topology graph to the distributed graph database 122 for data disk storage.

[0049] Furthermore, in step S100, after the underlying evidence conflict topology graph is stored on disk, the mediation task scheduling and data processing system 10 needs to convert the graph structure into a metric for resource scheduling. The core data scheduling server cluster 110 calls its internal graph computation engine 112 to perform the quantitative calculation process of the graph conflict density. The specific steps are as follows:

[0050] In step S106, the graph processing engine 112 retrieves the underlying evidence conflict topology graph corresponding to the current mediation task from the distributed graph database 122. The graph processing engine 112 executes the depth-first traversal algorithm to count the node attributes in the underlying evidence conflict topology graph. The graph processing engine 112 traverses all nodes in the underlying evidence conflict topology graph, identifies and removes isolated nodes that have no connection edges with any other node, and counts the remaining nodes with connection relationships as the total number of valid basic nodes. The graph processing engine 112 identifies connected subgraphs with mutual exclusion markers and extracts the total number of mutual exclusion edges.

[0051] Step S107: The graph processing engine 112 calculates the topological conflict connectivity based on the total number of mutually exclusive edges and the total number of valid basic nodes. The topological conflict connectivity characterizes the density of conflict distribution in the physical structure of the underlying evidence network. The formula for calculating the topological conflict connectivity is as follows:

[0052]

[0053] In the formula, The degree of topological conflict connectivity; This represents the total number of mutually exclusive edges in the evidence conflict topology graph. This represents the total number of valid basic nodes obtained through filtering.

[0054] In step S108, the graph processing engine 112 reads the conflict weight attribute bound to each mutual exclusion edge. The graph processing engine 112 calls the accumulator component 1121 to linearly sum the conflict weights of all mutual exclusion edges and divide by the total number of mutual exclusion edges to perform mean averaging. The weighted average conflict severity of the current mediation task is calculated. The calculation model for the weighted average conflict severity is as follows:

[0055]

[0056] In the formula, Weighted conflict severity; For the first The conflict weight of a mutually exclusive edge; This represents the total number of mutually exclusive edges in the evidence conflict topology graph.

[0057] Step S109: The graph processing engine 112 extracts the preset structural coefficients and weight coefficients;

[0058] In step S110, the graph computation engine 112 substitutes the topological conflict connectivity and weighted conflict severity into the aforementioned publicly disclosed graph conflict density calculation formula, multiplies the topological conflict connectivity with the extracted structural coefficients, multiplies the weighted conflict severity with the extracted weight coefficients, and adds the two products to calculate and output the final graph conflict density value. The graph computation engine 112 encapsulates the graph conflict density value into the attribute message of the current mediation task, and the core data scheduling server cluster 110 sends the attribute message carrying the graph conflict density value to the situational awareness scheduling bus 113 for subsequent use when performing resource anti-congestion scheduling allocation.

[0059] Furthermore, in step S200, before the mediation task scheduling and data processing system 10 performs resource scheduling, the mediation task scheduling and data processing system 10 uses the interactive interface running locally on the conventional mediation node 200 to collect probes 210 to extract multi-dimensional time-series operation features. The specific steps are as follows:

[0060] In step S201, the interactive interface acquisition probe 210 initializes the sliding time window in the operating system kernel of the conventional mediation node 200. The interactive interface acquisition probe 210 sets the length parameter and data sampling frequency of the sliding time window to provide the time boundary for dynamic acquisition. In a specific implementation, the sliding time window slides in equal steps with the system physical clock.

[0061] In step S202, the interactive interface acquisition probe 210 acquires the response delay increment of the regular mediation node 200. When the cloud platform 100 sends a new task message to the regular mediation node 200, it writes the task sending timestamp into the message protocol header. The interactive interface acquisition probe 210 listens for the first interrupt signal generated by the peripheral hardware of the regular mediation node 200 and records the interaction timestamp of the first interrupt signal. The interactive interface acquisition probe 210 retrieves the historical baseline response time of the regular mediation node 200 recorded in the time series database 123 through a network request. The interactive interface acquisition probe 210 calculates the response delay increment based on the interaction timestamp, the task sending timestamp, and the historical baseline response time. The calculation formula is as follows:

[0062]

[0063] In the formula, This is the increment of response delay calculated at the current moment; The interaction timestamps recorded by probe 210 are collected for the interactive interface; The task distribution timestamp carried in the message protocol header; The historical baseline response time of the conventional mediation node 200 is acquired by the acquisition probe 210 for the interactive interface, and the calculated response delay increment is then used to... When the value is less than zero, it is forcibly corrected to zero to eliminate the interference of positive fluctuations caused by a single response being faster than the baseline on subsequent state assessments;

[0064] In step S203, the interactive interface acquisition probe 210 acquires the text input rate variance. Within the sliding time window, the interactive interface acquisition probe 210 records the character input event signal triggered by the regular mediation node 200. The interactive interface acquisition probe 210 divides the sliding time window into multiple discrete time sub-intervals, counts the frequency of character input events in each time sub-interval, and calculates the discrete variance of the frequency of character input events in multiple time sub-intervals.

[0065] Step S204: The interactive interface acquisition probe 210 acquires the modification frequency and establishes a key mapping filter table. The key mapping filter table presets feature key values ​​related to content deletion and status rollback. In specific implementation, the feature key values ​​specifically include the underlying hardware scan codes corresponding to the backspace key, delete key, and undo combination key. The interactive interface acquisition probe 210 filters and counts the cumulative number of times the feature key values ​​are triggered within the sliding time window, and defines the cumulative number of times as the modification frequency.

[0066] In step S205, the interactive interface acquisition probe 210 aligns and packages the response delay increment, text input rate variance, and modification frequency according to the end timestamp of the sliding time window. The interactive interface acquisition probe 210 encapsulates the aligned feature data into a multi-dimensional time-series operation feature vector. The interactive interface acquisition probe 210 pushes the multi-dimensional time-series operation feature vector unidirectionally to the time series database 123 of the core data scheduling server cluster 110 for persistent storage through an asynchronous message communication interface, so that it can be called by the situational awareness scheduling bus 113.

[0067] Further, in step S300, after the situational awareness scheduling bus 113 acquires the multi-dimensional timing operation feature vector of the regular mediation node 200, it performs a baseline comparison between the multi-dimensional timing operation feature vector and the historical operating state, as well as the calculation of the dynamic degradation index. The specific steps are as follows:

[0068] In step S301, the situational awareness scheduling bus 113 extracts the multi-dimensional time-series operation feature vector of the regular mediation node 200 from the time series database 123. The multi-dimensional time-series operation feature vector includes the response delay increment, text input rate variance, and modification frequency within the current sliding time window. At the same time, the situational awareness scheduling bus 113 retrieves the standard baseline distribution model formed by the corresponding regular mediation node 200 in the historical stable operation period from the time series database 123. The standard baseline distribution model records the historical mean and historical standard deviation of the three features contained in the multi-dimensional time-series operation feature vector.

[0069] In step S302, the situational awareness scheduling bus 113 performs dimensional unification processing on the multi-dimensional time-series operation feature vector. Since the physical units of the response delay increment, text input rate variance, and modification frequency are inconsistent, the situational awareness scheduling bus 113 uses a standard fraction calculation method to convert the actual collected feature values ​​into dimensionless relative offsets. The specific calculation formula for the relative offset is as follows:

[0070]

[0071] In the formula, For the current moment The relative offset of the feature item; For the first The actual collected values ​​of the feature; In the standard baseline distribution model, the first Historical mean of the feature; In the standard baseline distribution model, the first Historical standard deviation of the feature; This is a preset, minimal normal quantity to prevent the denominator from being zero;

[0072] In step S303, the situational awareness scheduling bus 113, based on the system's preset feature penalty weighting factors, performs a weighted summation of the relative offsets of the calculated features, and outputs the dynamic degradation index of the regular mediation node 200. Specifically, the system configures independent penalty weighting factors for the response delay increment, text input rate variance, and modification frequency. To eliminate the impact of physical dimension differences on the architecture's computing power, the situational awareness scheduling bus 113 uses relative offsets to calculate the dynamic degradation index. The calculation formula is as follows:

[0073]

[0074] In the formula, The dynamic downgrade index for the 200th regular mediation node; The relative offset of the response delay increment; This represents the relative offset of the variance in text input rate; To modify the relative offset of the frequency; , , These are the feature penalty weighting factors bound to the corresponding relative offsets. To ensure the rigor of the degradation assessment boundary, the dynamic degradation index is calculated by weighted summation. When the value is less than zero, the situational awareness scheduling bus 113 truncates it and assigns it a value of zero, indicating that the regular mediation node 200 is currently in an excellent healthy operating state without any degradation in computing power;

[0075] Step S304: The situational awareness scheduling bus 113 performs dynamic evolution of the standard baseline distribution model; the system smoothly corrects itself as the proficiency of the regular mediation node 200 changes; the situational awareness scheduling bus 113 receives transaction loop receipts from the external e-commerce transaction core system; when the transaction loop receipt indicates that the mediation status corresponding to the current task is successful, and the situational awareness scheduling bus 113 determines that the graph conflict density value of the current task is lower than the preset safety threshold, the situational awareness scheduling bus 113 extracts the operational feature values ​​within the current task cycle to update the historical average value; the baseline smoothing update formula is as follows:

[0076]

[0077] In the formula, For the updated number Historical mean of the feature; For the previous version Historical mean of the feature; For the first The actual collected values ​​of the feature; The forgetting decay rate parameter configured for the system, and its value range is [value range missing]. ;

[0078] In step S305, the situational awareness scheduling bus 113 writes the calculated dynamic degradation index into the current running status register of the regular mediation node 200. The situational awareness scheduling bus 113 synchronizes the updated standard baseline distribution model to the time series database 123 of the persistent storage matrix 120, and uses the updated standard baseline distribution model to overwrite the old baseline data records for matching and comparison in the next sliding time window.

[0079] Further, in step S300, after obtaining the dynamic degradation index of the regular mediation node 200, the situational awareness scheduling bus 113 constructs a security boundary for resource scheduling using the degradation warning threshold and performs initial scheduling determination. The specific steps are as follows:

[0080] Step S306: The situational awareness scheduling bus 113 initializes the degradation warning threshold and the preset difficulty baseline of the mediation task for the global nodes. The situational awareness scheduling bus 113 extracts the historical dataset of dynamic degradation index of all regular mediation nodes 200 of the cloud platform 100 based on the sliding window records of the past 24 hours from the time series database 123. The situational awareness scheduling bus 113 calculates the global mean and global standard deviation of the historical dataset of dynamic degradation index, and multiplies the global standard deviation by a preset sensitivity coefficient. Then, it is added to the global mean, and the arithmetic sum is set as the degraded warning threshold for the current scheduling cycle. At the same time, the situational awareness scheduling bus 113 extracts the map conflict density dataset of historical mediation tasks within the same sliding window from the time series database 123. The situational awareness scheduling bus 113 calculates the global mean and global standard deviation of the historical map conflict density, and multiplies the global standard deviation by the preset sensitivity coefficient. Then add it to the global mean, and set its arithmetic sum as the preset difficulty baseline;

[0081] Step S307: When a newly entered mediation task is in the flow queue of the cloud platform 100, the situational awareness scheduling bus 113 parses the task attribute message, extracts the graph conflict density value corresponding to the mediation task, and sends a polling request to the status register to obtain the real-time dynamic degradation index of the candidate regular mediation node 200 that is currently in an idle state or whose number of queued tasks is lower than the maximum buffer queue length configured by the system.

[0082] In step S308, the situational awareness scheduling bus 113 compares the real-time dynamic degradation index of the candidate regular mediation node 200 with the degradation warning threshold. When the real-time dynamic degradation index is strictly less than the degradation warning threshold, the situational awareness scheduling bus 113 determines that the candidate regular mediation node 200 is within the safe operation boundary. The situational awareness scheduling bus 113 grants the processing permission for high-complexity tasks and allows the distribution of mediation tasks with a map conflict density value greater than the preset difficulty baseline to the candidate regular mediation node 200.

[0083] In step S309, when the real-time dynamic degradation index is greater than or equal to the degradation warning threshold, the situational awareness scheduling bus 113 determines that the candidate regular mediation node 200 has crossed the safe operation boundary. The situational awareness scheduling bus 113 intercepts the routing distribution of mediation tasks with a graph conflict density value greater than the preset difficulty baseline to the candidate regular mediation node 200 at the system bus level. The situational awareness scheduling bus 113 generates a state change signaling and sends the state change signaling to the core data scheduling server cluster 110, triggering the subsequent cooling compensation allocation execution steps.

[0084] Furthermore, when the situational awareness scheduling bus 113 determines that the candidate regular mediation node 200 has crossed the safe operation boundary, the mediation task scheduling and data processing system 10 activates the cooling compensation allocation mechanism to restore the concurrent operation capability of the processing node through physical load reduction. The specific implementation steps are as follows:

[0085] In step S310, the situational awareness scheduling bus 113 receives an internally triggered state change signaling, modifies the identifier parameters in the node registry 1131 maintained internally by the situational awareness scheduling bus 113, updates the status identifier of the candidate regular mediation node 200 that crosses the safe operation boundary from the regular allocation state to the cooling compensation state, and locks the task distribution channel, refusing to assign mediation tasks with a map conflict density value greater than the aforementioned preset difficulty baseline to the candidate regular mediation node 200 in the cooling compensation state.

[0086] Step S311: The situational awareness scheduling bus 113 activates the low-complexity task retrieval subroutine. The situational awareness scheduling bus 113 traverses the global flow pool of the cloud platform 100 and, based on the data structure of the underlying evidence conflict topology graph, retrieves mediation tasks whose graph conflict density value is strictly lower than the preset cooling standard value. The preset cooling standard value is calculated as the product of the global mean of the graph conflict density of historical mediation tasks minus the global standard deviation and the preset leniency coefficient m. The tasks selected by the situational awareness scheduling bus 113 based on the preset cooling standard value are standardized simple tasks with no mutually exclusive edges or very few core dispute nodes.

[0087] In step S312, the situational awareness scheduling bus 113 constructs a set of cooling tasks from the retrieved and matched standardized simple tasks. The situational awareness scheduling bus 113 extracts the queuing timestamp of each task in the cooling task set. The queuing timestamp is a time mark uniformly added by the mediation task scheduling and data processing system 10 when the mediation task initially enters the global flow pool. The situational awareness scheduling bus 113 distributes the standardized simple tasks to the candidate regular mediation nodes 200 in the cooling compensation state in chronological order.

[0088] In step S313, during the process of candidate conventional mediation node 200 processing standardized simple tasks, the interactive interface acquisition probe 210 continuously collects multi-dimensional time-series operation features and transmits them back to the time series database 123. The situational awareness scheduling bus 113 calculates and updates the dynamic degradation index of candidate conventional mediation node 200 in the cooling compensation state in real time according to the dynamic degradation index calculation formula disclosed in the aforementioned embodiment.

[0089] In step S314, the situational awareness scheduling bus 113 compares the updated dynamic degradation index with the safety recovery threshold in real time. In order to prevent the node state from switching frequently and at high frequency near the critical boundary value, the situational awareness scheduling bus 113 subtracts the constant anti-oscillation hysteresis amount configured by the system from the degradation warning threshold, and then sets the safety recovery threshold.

[0090] Step S315: When the situational awareness scheduling bus 113 detects that the dynamic degradation index of the candidate regular mediation node 200 has dropped to a level strictly below the safety recovery threshold, it determines that the processing capability of the candidate regular mediation node 200 has been restored. The situational awareness scheduling bus 113 modifies the identifier parameter in the node registry 1131 to restore the status identifier bit of the candidate regular mediation node 200 to the regular allocation state. The situational awareness scheduling bus 113 releases the locking and interception restriction of the task distribution channel and reopens the processing permission of high-complexity tasks to the candidate regular mediation node 200.

[0091] Furthermore, in step S400, when the regular mediation node 200 is handling a highly complex mediation task with a graph conflict density value greater than a preset difficulty baseline, the mediation task scheduling and data processing system 10 initiates a high-frequency monitoring program for the dynamic degradation index to identify state changes caused by the regular mediation node 200 processing complex evidence kernels. The specific implementation steps are as follows:

[0092] In step S401, the situational awareness scheduling bus 113 configures a high-frequency sampling timer 1132 during the operation of the regular mediation node 200. The sampling time interval of the high-frequency sampling timer 1132 is strictly less than the regular sliding time window step of the dynamic degradation index. The situational awareness scheduling bus 113 extracts the real-time dynamic degradation index at multiple consecutive discrete time points through the high-frequency sampling timer 1132 and stores the extracted real-time dynamic degradation index into the memory buffer queue 1133 inside the situational awareness scheduling bus 113.

[0093] In step S402, the situational awareness scheduling bus 113 calculates the time partial derivative of the dynamic degradation index based on the real-time dynamic degradation index in the memory buffer queue 1133 to obtain the dynamic deterioration rate of the conventional mediation node 200. In a specific implementation, since the underlying physical devices of the computer cannot directly handle the limit differentiation of continuous functions, the situational awareness scheduling bus 113 uses a backward difference algorithm to calculate the dynamic degradation index at discrete time points. The formula for calculating the dynamic deterioration rate is as follows:

[0094]

[0095] In the formula, The dynamic deterioration rate calculated at the current moment; For the regular mediation node 200 Real-time dynamic degradation index at any given moment; This represents the dynamic degradation index of the regular mediation node 200 at the previous sampling time. This is the sampling time interval of the high-frequency sampling timer 1132;

[0096] In step S403, the situational awareness scheduling bus 113 extracts deterioration rate samples from the time series database 123 when the regular mediation node 200 handles similar high-complexity tasks in the past. Specifically, the situational awareness scheduling bus 113 calculates the difference between the map conflict density value of the historical mediation task and the map conflict density value of the current mediation task. When the absolute value of the difference is less than the preset density tolerance, the situational awareness scheduling bus 113 determines the corresponding historical mediation task as a similar high-complexity task in the past. The situational awareness scheduling bus 113 extracts the upper limit of the normal distribution information interval of the deterioration rate sample and sets the upper limit of the normal distribution information interval of the deterioration rate sample as the sudden increase boundary value of the node's running state collapse. ;

[0097] In step S404, the situational awareness scheduling bus 113 will use the dynamically calculated deterioration rate and the sudden increase boundary value to... Numerical comparisons were performed, and the dynamic deterioration rate was found to be strictly greater than the sudden increase boundary value. When the situational awareness scheduling bus 113 determines that the regular mediation node 200 is trapped in a local high-conflict data resource bottleneck in the current mediation task, the situational awareness scheduling bus 113 sends a status freeze command to the interactive interface acquisition probe 210 of the regular mediation node 200.

[0098] In step S405, after the interactive interface acquisition probe 210 receives the status freeze command, it suspends the input interrupt response of the peripheral hardware of the conventional mediation node 200, locks the local I / O read and write permissions of the corresponding main task process, and freezes the current interactive flow state of the main task in the conventional mediation node 200. The situational awareness scheduling bus 113 synchronously sends an architecture reorganization command to the graph computing engine 112, triggering the subsequent low-level evidence conflict topology graph segmentation and stripping steps.

[0099] Furthermore, in step S400, after receiving the architecture reorganization instruction sent by the situational awareness scheduling bus 113, the graph processing engine 112 performs dynamic structural segmentation on the underlying evidence conflict topology graph to separate the high-density data blocks that cause the state collapse of the regular mediation node 200. The specific implementation steps are as follows:

[0100] Step S406: The graph processing engine 112 retrieves the underlying evidence conflict topology graph corresponding to the current freezing task. The graph processing engine 112 performs connected subgraph density clustering analysis on the underlying evidence conflict topology graph. Specifically, the graph processing engine 112 expands the search with the mutual exclusion edges in the underlying evidence conflict topology graph as the core, extracts the number of local mutual exclusion edges, the number of local nodes, and the local conflict weights bound to the local mutual exclusion edges in each local connected region, and calculates the local conflict density value of each local connected region. The formula for calculating the local conflict density value is as follows:

[0101]

[0102] In the formula, The calculated local conflict density value; This represents the number of locally mutually exclusive edges within a locally connected region. This represents the number of local nodes within the locally connected region. For the first local connected region Local conflict weights of mutually exclusive edges; These are the aforementioned dimensionless preset structural coefficients; As the aforementioned dimensionless preset weight coefficient, the graph processing engine 112 extracts the graph conflict density value of the current frozen task as a whole, and sets the graph conflict density value of the current frozen task as a preset segmentation threshold. The graph processing engine 112 locks the local connected region with the highest local conflict density value that is greater than the preset segmentation threshold.

[0103] In step S407, the graph processing engine 112 performs logical stripping on the locked local connected regions. The graph processing engine 112 cuts off the regular connection edges between the locked local connected regions and the rest of the topology graph in the distributed graph database 122, and extracts the locked local connected regions independently as high-density topology subgraphs. After the stripping operation is completed, the remaining non-core low-density topology structures in the underlying evidence conflict topology graph constitute the residual graph.

[0104] In step S408, the situational awareness scheduling bus 113 sends a status unfreezing command to the interactive interface acquisition probe 210. The interactive interface acquisition probe 210 unlocks the local I / O read / write permission lock of the main task process corresponding to the regular mediation node 200. The regular mediation node 200 loads the residual graph with reduced difficulty and continues to execute the subsequent mediation interaction process. After the high-density topology subgraph is logically removed, the actual computing resource consumption faced by the regular mediation node 200 decreases accordingly.

[0105] Further, in step S500, the core data scheduling server cluster 110 packages and forwards the high-density topology subgraph to the expert processing node 300. After the expert processing node 300 completes the data micro-task for the high-density topology subgraph, it mediates the task scheduling and data processing system 10 to perform cross-domain fusion of the micro-task processing results and the main task residual graph processing progress. The specific implementation steps are as follows:

[0106] In step S501, the core data scheduling server cluster 110 encapsulates the high-density topology subgraph into independent data microtasks. The core data scheduling server cluster 110 reads the text feature vectors and original voucher indexes bound to each node and edge in the high-density topology subgraph, and serializes the text feature vectors and original voucher indexes into a standard data exchange format. In specific implementation, the standard data exchange format is JSON format. The core data scheduling server cluster 110 assigns a globally unique task identifier to the data microtask and writes the context routing label of the main task into the protocol header of the data microtask.

[0107] In step S502, the core data scheduling server cluster 110 asynchronously unloads the data micro-tasks to the background transfer pool 102 through the intranet routing bus 101 of the cloud platform 100. Based on the permission matching logic, the background transfer pool 102 distributes the data micro-tasks to the expert processing nodes 300 that are in an idle state. The expert processing nodes 300 perform multi-dimensional cross-evidence analysis on the dense mutually exclusive edge network inside the high-density topology subgraph without reading or loading the context data of the residual graph.

[0108] In step S503, the expert processing node 300 generates a cross-domain judgment instruction for the mutual exclusion edges inside the high-density topology subgraph locally. The cross-domain judgment instruction carries a globally unique task identifier and a confirmation result payload. The confirmation result payload includes confirmation labels for the disputed entity attributes and mutual exclusion edge elimination markers. The expert processing node 300 asynchronously transmits the cross-domain judgment instruction back to the core data scheduling server cluster 110 through the intranet routing bus 101 of the cloud platform 100.

[0109] In step S504, the heterogeneous data receiving gateway 111 of the core data scheduling server cluster 110 receives the cross-domain analysis instruction, the core data scheduling server cluster 110 parses the cross-domain analysis instruction, extracts the globally unique task identifier, and the core data scheduling server cluster 110 queries the original underlying evidence conflict topology map mapped in the distributed graph database 122 based on the globally unique task identifier, and at the same time queries the processing progress log of the regular mediation node 200 bound to the globally unique task identifier.

[0110] In step S505, the core data scheduling server cluster 110 retrieves the regular processing instructions generated by the regular mediation node 200 for the residual graph based on the queried processing progress log, and the core data scheduling server cluster 110 extracts the low-conflict node status update parameters from the regular processing instructions.

[0111] In step S506, the graph processing engine 112 performs logical concatenation and topology reconstruction of the data states. The graph processing engine 112 defines the weighting result payload in the cross-domain judgment instruction as a micro-task state set, and defines the low-conflict node state update parameters in the regular processing instruction as a residual graph state set. The graph processing engine 112 overwrites the attributes of the original topology structure based on the micro-task state set and the residual graph state set, and reconstructs a complete globally consistent evidence topology graph. The logical merging formula of the globally consistent evidence topology graph is as follows:

[0112]

[0113] In the formula, This is the reconstructed global consistency evidence topology graph; This refers to the high-density topological subgraph stripped in the aforementioned steps; For expert processing node 300, the micro-task status set is processed. This is a function for overwriting node attributes and eliminating mutually exclusive edges in high-density topological subgraphs; its output is the updated conflict-free local graph structure. The residual image retained in the aforementioned steps; The residual graph state set fed back by the regular mediation node 200; This is a function for updating node attributes of the residual graph, and its output is the updated residual graph structure. This refers to the merging operation between the node set and edge set of the two updated graph structures mentioned above;

[0114] In step S507, the graph processing engine 112 extracts the underlying unique hash code from the microtask state set and the residual graph state set, and the graph processing engine 112 performs the overwriting and merging of graph node attributes based on the underlying unique hash code.

[0115] In step S508, the graph processing engine 112 performs connectivity verification on the assembled global consistency evidence topology graph. The graph processing engine 112 traverses all nodes of the global consistency evidence topology graph and checks whether there are dangling edges or unresolved mutual exclusion logic conflicts within the global consistency evidence topology graph. When the graph processing engine 112 detects dangling edges or unresolved mutual exclusion logic conflicts within the global consistency evidence topology graph, the graph processing engine 112 determines that the connectivity verification has failed. The graph processing engine 112 triggers an exception rollback mechanism, cancels the attribute overwrite operation of the microtask state set and the residual graph state set, and routes the original underlying evidence conflict topology graph to the manual review queue. When the graph processing engine 112 detects that there are no dangling edges and no unresolved mutual exclusion logic conflicts within the global consistency evidence topology graph, the graph processing engine 112 determines that the verification has passed. The graph processing engine 112 persists the global consistency evidence topology graph to the distributed graph database 122, completing the data closure of the heterogeneous task.

[0116] Furthermore, after completing the persistent writing of the globally consistent evidence topology graph, the mediation task scheduling and data processing system 10 outputs business results based on the reconstructed data structure and performs adaptive iteration of situational awareness monitoring parameters. The specific implementation steps are as follows:

[0117] In step S509, the core data scheduling server cluster 110 reads the globally consistent evidence topology graph stored in the distributed graph database 122, the core data scheduling server cluster 110 calls the preset business rule decision engine 114, the core data scheduling server cluster 110 inputs the attributes of the leaf nodes after eliminating mutual exclusion edges in the globally consistent evidence topology graph into the business rule decision engine 114, and the business rule decision engine 114 outputs the fund transfer allocation ratio and responsibility determination code for the current mediation task. The core data scheduling server cluster 110 assembles the fund transfer allocation ratio and responsibility determination code into a structured mediation resolution message. The structured mediation resolution message is constructed in XML format and encapsulates the settlement routing addressing field of the corresponding third-party payment gateway 500.

[0118] In step S510, the situational awareness scheduling bus 113 retrieves the execution split log of the current mediation task. The situational awareness scheduling bus 113 extracts the actual deviation direction of the multi-dimensional timing operation features of the regular mediation node 200 within the sliding time window before triggering the state freeze command. The situational awareness scheduling bus 113 calculates the first-order difference value of each multi-dimensional timing operation feature at the state freeze critical moment. The situational awareness scheduling bus 113 identifies the feature with the largest absolute value of the first-order difference value as the dominant feature dimension that causes the sudden increase of the dynamic degradation index.

[0119] In step S511, the situational awareness scheduling bus 113 calculates the correction compensation amount of the feature penalty weight factor corresponding to the dominant feature dimension. The situational awareness scheduling bus 113 increases the feature penalty weight factor bound to the dominant feature dimension to improve the monitoring sensitivity of the conventional mediation node 200 to the same fatigue features in subsequent processing. The correction formula for the feature penalty weight factor is as follows:

[0120]

[0121] In the formula, For the updated number Feature penalty weight factor; For the previous version Feature penalty weight factor; The weight learning rate parameter configured for a dimensionless system; For situational awareness scheduling bus 113, the critical moment of the triggered state is frozen. Recorded dynamic deterioration rate; The aforementioned downgrade warning threshold; As the dominant feature indicator function, when the first When the feature is the dominant feature dimension ,otherwise , The aforementioned sudden increase boundary value;

[0122] In step S512, the situational awareness scheduling bus 113 performs normalization verification on all updated feature penalty weight factors. The situational awareness scheduling bus 113 divides each feature penalty weight factor by the algebraic sum of all feature penalty weight factors to ensure that the sum of the updated feature penalty weight factors is always equal to the constant 1. The situational awareness scheduling bus 113 writes the normalized feature penalty weight factors into the monitoring configuration item of the corresponding conventional mediation node 200 to complete the closed-loop evolution of situational awareness parameters.

[0123] In step S513, the core data scheduling server cluster 110 sends the structured mediation resolution message to the external e-commerce transaction core system 400. The external e-commerce transaction core system 400 performs fund settlement and order status update based on the structured mediation resolution message.

Claims

1. A method for scheduling and processing mediation tasks based on a cloud platform, characterized in that, Includes the following steps: Step S100: Obtain multi-source dispute data from external input through the heterogeneous data receiving gateway, call the graph computing engine to convert the multi-source dispute data into a graph structure mapping, establish a global evidence conflict topology graph, and calculate the conflict density value of the global task. Step S200: The interactive interface acquisition probe of the conventional mediation node calls the system standard input event API interface to extract the underlying input event vector of the terminal device at a preset sampling frequency. The underlying input event vector is processed locally to generate multi-dimensional time-series operation features, and is pushed unidirectionally to the core data scheduling server cluster through the asynchronous message communication interface. Step S300: The situational awareness scheduling bus retrieves the historical baseline data corresponding to the conventional mediation node from the time series database, performs comparison calculation using the multi-dimensional time series operation features, generates the dynamic degradation index of the conventional mediation node, combines the conflict density value of each task in the task queue, performs resource anti-congestion scheduling allocation, and distributes task instructions to the conventional mediation node with matching status. In step S400, the situational awareness scheduling bus performs time partial derivative calculation on the regular mediation node in the execution state to obtain the deterioration rate of the dynamic degradation index of the regular mediation node. When it is determined that the deterioration rate of the dynamic degradation index exceeds the sudden increase boundary value configured by the system, a state freeze command is sent to the regular mediation node and an architecture reorganization command is sent to the graph computing engine. The graph computing engine cuts off and logically strips the dense mutually exclusive edge region in the evidence conflict topology graph corresponding to the current task. In step S500, the core data scheduling server cluster packages the high-density topology subgraph obtained by stripping into independent network data packets, forwards them across logically isolated network segments to the expert processing node, the expert processing node performs comprehensive verification and confirmation of rights for multi-dimensional related evidence, and injects the execution receipt with attached status modification mark in reverse, and the core data scheduling server cluster completes the final assembly and write-back operation of the mediation log status.

2. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 1, characterized in that, In step S100, establishing a global evidence conflict topology map specifically includes: Using a pre-trained sequence labeling model, the word boundary tagging method is used to assign entity category labels to each character in a one-dimensional text sequence, identify the disputed entities in the dispute and the constraint claims made by the buyers and sellers against the disputed entities; the dependency parsing algorithm is called to extract the subject-verb-object grammatical dependencies between the disputed entities and the constraint claims in the text sentences, and the disputed entities with direct grammatical dependencies are paired with the constraint claims to output entity-claim data pairs. The identified disputed entities and constraint claims are instantiated as a set of nodes in the topological space, and directed association edges are established between the corresponding disputed entity nodes and constraint claim nodes to generate a set of regular edges, forming a basic evidence topological graph. Extract text vector features of nodes with different constraints and calculate semantic cosine similarity. When a contradictory semantic feature dictionary is hit and the semantic cosine similarity is lower than the logical consistency judgment threshold set by the system, it is determined that there is an irreconcilable logical conflict for the corresponding disputed entity node. Mutually exclusive edges are generated between the corresponding nodes with different constraints, and a set of mutually exclusive edges is established. The financial impact of mutually exclusive edges on the disputed entities is quantified and weighted. The conflict weight of each mutually exclusive edge in the set of mutually exclusive edges is calculated. The set of mutually exclusive edges and the conflict weight parameters carried by the set of mutually exclusive edges are merged into the basic evidence topology graph to complete the construction of the underlying evidence conflict topology graph.

3. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 1, characterized in that, In step S100, calculating the conflict density value of the global task specifically includes: Identify and remove isolated nodes, count the remaining nodes with connections as the total number of valid basic nodes, and extract the total number of mutually exclusive edges. Read the conflict weight attribute bound to each mutual exclusion edge, sum the conflict weights of all mutual exclusion edges linearly, and average them by dividing by the total number of mutual exclusion edges to calculate the weighted conflict severity of the current mediation task. The topological conflict connectivity is multiplied by the extracted structural coefficients. The weighted conflict severity is then normalized by dividing by the total number of mutually exclusive edges and multiplied by the extracted weight coefficients. These two products are then summed to calculate the final graph conflict density value. The formula for calculating the graph conflict density value is as follows: In the formula, This represents the conflict density value. This represents the total number of mutually exclusive edges in the evidence conflict topology graph. This represents the total number of valid basic nodes obtained through filtering. For the first The conflict weight of a mutually exclusive edge; These are dimensionless, pre-defined structural coefficients. These are dimensionless preset weighting coefficients.

4. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 1, characterized in that, In step S200, generating multi-dimensional time-series operation features specifically includes: Initialize the sliding time window in the operating system kernel of the regular mediation node; Listen to the first input event signal generated by the peripheral hardware of the regular mediation node and record the interaction timestamp. Calculate the response delay increment based on the interaction timestamp, the task issuance timestamp and the historical baseline response time. Record character input event signals within a sliding time window, count the frequency of character input events in each time sub-interval and calculate the discrete variance, which is then output as the text input rate variance. Establish a key mapping filter table to filter and count the cumulative number of underlying hardware scan codes corresponding to the backspace key, delete key, and undo key combination key triggered within the sliding time window, and define it as the modification frequency; The response delay increment, text input rate variance, and modification frequency are aligned and packaged according to the end timestamp of the sliding time window, and encapsulated into a multidimensional temporal operation feature vector.

5. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 4, characterized in that, In step S300, the dynamic degradation index of the conventional mediation node is generated, specifically including: Retrieve the standard baseline distribution model formed by the corresponding regular mediation node in the historical stable operation period from the time series database, and extract the historical mean and historical standard deviation of response delay increment, text input rate variance, and modification frequency. The standard fraction calculation method is used to convert the actual collected feature values ​​into dimensionless relative offsets; The dynamic degradation index is calculated using a relative offset, and the calculation formula is as follows: In the formula, For regular mediation nodes The dynamic degradation index at any given moment; The relative offset of the response delay increment; This represents the relative offset of the variance in text input rate; To modify the relative offset of the frequency; , , These are the feature penalty weighting factors bound to the corresponding relative offsets. To ensure the rigor of the degradation assessment boundary, the dynamic degradation index is calculated by weighted summation. When the value is less than zero, the situational awareness scheduling bus 113 truncates it and assigns a value of zero, indicating that the regular mediation node 200 is currently in an excellent healthy operating state without any degradation in computing power.

6. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 1, characterized in that, In step S300, resource congestion prevention scheduling and allocation are performed, specifically including: Calculate the global mean and global standard deviation of the historical dataset of dynamic degradation index based on the sliding window records of the past 24 hours. Multiply the global standard deviation by a preset sensitivity coefficient and add the result to the global mean. Set the arithmetic sum as the degradation warning threshold for the current scheduling cycle. At the same time, calculate the global mean and global standard deviation of the graph conflict density of historical mediation tasks. Multiply the global standard deviation by the preset sensitivity coefficient and add the result to the global mean. Set the result to the preset difficulty baseline. The real-time dynamic degradation index of the candidate regular mediation node is compared with the degradation warning threshold. When the real-time dynamic degradation index is greater than or equal to the degradation warning threshold, it is determined that the safe operation boundary has been crossed. At the system bus level, the routing of mediation tasks with a graph conflict density value greater than the preset difficulty baseline is intercepted. Modify the identifier parameters in the node registry to update the status identifier of candidate regular mediation nodes that have crossed the safe operation boundary to the cooling compensation state; Mediation tasks whose search map conflict density values ​​are strictly lower than a preset cooling standard value are constructed into a cooling task set. Standardized simple tasks are distributed to candidate regular mediation nodes in the cooling compensation state in chronological order. During the process of the candidate regular mediation nodes processing the standardized simple tasks, their dynamic degradation index is updated in real time and compared with the safety recovery threshold. When the dynamic degradation index drops to strictly less than the safety recovery threshold, the status flag of the candidate regular mediation node is restored to the regular allocation state.

7. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 1, characterized in that, In step S400, obtaining the deterioration rate of the dynamic degradation index of the regular mediation node and sending an architecture reorganization command to the graph computing engine specifically includes: In the operation process of the regular mediation node, a high-frequency sampling timer is configured to extract the real-time dynamic degradation index, and the backward difference algorithm is used to calculate and obtain the dynamic deterioration rate. After classifying historical mediation tasks as similar high-complexity tasks in the past, the upper limit of the normal distribution confidence interval of the deterioration rate sample is set as the burst boundary value. When the dynamic deterioration rate is strictly greater than the sudden increase boundary value, a state freeze command is sent to the acquisition probe of the interaction interface of the regular mediation node. Suspend the peripheral hardware input event response of the process corresponding to the current main task and lock the local I / O read and write permissions of the process, freeze the interactive flow state of the current main task on the local regular mediation node, and send the architecture reorganization command synchronously.

8. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 1, characterized in that, The process of severing and logically stripping the densely mutually exclusive edge region in the evidence conflict topology graph corresponding to the current task specifically includes: The search is expanded with the mutual exclusion edges in the underlying evidence conflict topology as the core. The number of local mutual exclusion edges, the number of local nodes, and the local conflict weights in each local connected region are extracted, and the local conflict density value of each local connected region is calculated. The local connected region with the highest local conflict density value that is greater than the preset segmentation threshold is locked and logically stripped. The regular connection edges between the locked local connected region and the rest of the topology graph are cut off. The locked local connected region is extracted as a high-density topology subgraph, and the remaining non-core low-density topology structure constitutes the residual graph.

9. The method for scheduling and processing mediation tasks based on a cloud platform according to claim 1, characterized in that, Step S500 specifically includes: Read the text feature vectors and original voucher indexes of each node and edge in the high-density topology subgraph, serialize them into a standard data exchange format, encapsulate them into independent data micro-tasks, and assign a globally unique task identifier to asynchronously unload them to the background transfer pool. Parse cross-domain analysis commands, extract globally unique task identifiers, and query the original underlying evidence conflict topology and processing progress logs. Retrieve the regular processing instructions generated for the residual graph, define the weighting result payload in the cross-domain judgment instruction as the micro-task state set, and define the low-conflict node state update parameters in the regular processing instructions as the residual graph state set. Extract the underlying unique hash code from the microtask state set and the residual graph state set, and perform graph node attribute overwriting and merging based on the underlying unique hash code to reconstruct the complete global consistency evidence topology graph; Check if there are any dangling edges or unresolved mutual exclusion logic conflicts within the global consistency evidence topology graph. If the verification is successful, persist the global consistency evidence topology graph to the distributed graph database.

10. A method for scheduling and processing mediation tasks based on a cloud platform according to claim 9, characterized in that, After completing the final assembly and write-back of the mediation log status, the process also includes: The leaf node attributes after eliminating mutually exclusive edges in the globally consistent evidence topology graph are input into the business rule decision engine, which outputs the fund transfer allocation ratio and liability determination code, and assembles them into a structured mediation resolution message to be sent to the external e-commerce transaction core system. Extract the actual deviation direction of the multi-dimensional time-series operation features of the regular mediation node before triggering the state freeze command, calculate the first-order difference value at the critical moment of state freeze, and identify the feature with the largest absolute value of the first-order difference value as the dominant feature dimension that causes the sudden increase of the dynamic degradation index. Calculate the correction compensation amount of the feature penalty weight factor corresponding to the dominant feature dimension, increase the feature penalty weight factor bound to the dominant feature dimension, and after performing normalization verification on all updated feature penalty weight factors, overwrite the corresponding monitoring configuration item in the regular mediation node.

Images