Intelligent management and control implementation method for call center work order balancing deployment

Abstract

This invention discloses an intelligent management and control method for balanced allocation of call center work orders, specifically relating to the field of call center work order allocation. It addresses the problem of call sessions remaining connected and resulting in invalid work order allocation when external resources are limited. By obtaining resource response status feedback from external third-party systems, an available attenuation slope and a call queue influx slope are formed around the target business resource. The intersection collision time is further fitted and combined with a preset buffer waiting time to output a backpressure circuit breaker signal. Then, based on the backpressure circuit breaker signal, call sessions with intent related to the target business resource are intercepted and placed in a suspended state. Subsequently, virtual macro work orders are clustered to generate virtual macro work orders. A batch locking request is initiated to the external third-party system using the global identifier of the virtual macro work order to obtain a combined credential pool. Finally, resource credential distribution and underlying signaling redirection are completed based on value weight allocation rules.

Description

Technical Field

[0001] This invention relates to the field of call center work order allocation, and more specifically, to an intelligent management and control method for balanced allocation of call center work orders. Background Technology

[0002] In service scenarios such as air travel and ticketing, which heavily rely on real-time resources, call centers are responsible for the centralized processing of customer requests, queuing, and work order allocation. Existing automatic call distribution systems and work order scheduling methods in call centers typically use communication-side information such as agent availability, queue waiting time, and customer level as the primary allocation criteria. After a call session enters the queue, the agent then initiates an external resource query and locking operation. This approach can complete call allocation in general scenarios, but its scheduling basis remains primarily at the communication layer, failing to incorporate the real-time availability of target business resources into the front-end queuing control chain.

[0003] When external third-party systems experience rate limiting, response timeouts, or a rapid decrease in target service resources, the call queue continues to push towards human operators. This results in a large number of call sessions entering the agent processing stage before resources are available. At this point, agent terminals repeatedly initiate resource query and lock requests for the same target service resource, easily leading to high-frequency concurrent access and repeated contention, causing invalid connections, queue backlogs, and disordered subsequent reallocation in the work order processing chain. Current technology lacks a mechanism to link the resource response status reported by external third-party systems with the call queue changes, making it difficult to identify risks in advance and reorganize call session allocation paths during resource-scarce phases.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent management and control method for balanced allocation of call center work orders. This method obtains resource response status feedback from an external third-party system, forms an available attenuation slope and a call queue influx slope around the target business resource, further fits the intersection collision time, and outputs a backpressure circuit breaker signal based on a preset buffer waiting time. Then, based on the backpressure circuit breaker signal, it intercepts call sessions with intent related to the target business resource and places them in a suspended state. Subsequently, it clusters and generates virtual macro work orders, initiates a batch locking request to the external third-party system using the global identifier of the virtual macro work order, obtains a combined credential pool, and completes resource credential distribution and underlying signaling redirection based on value weight allocation rules, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In response to the resource response status reported by the external third-party system, assess the available decay slope of the target business resources and the influx slope of the call queue.

[0008] Based on the available attenuation slope and the inflow slope, the intersection collision time is fitted, and when the intersection collision time is within the preset buffer waiting time, the back pressure fuse signal is triggered.

[0009] In response to the back pressure fuse signal, intercept call sessions with the intent to associate with the target service resources and put them in a suspended state, and cluster multiple call sessions in the suspended state to generate virtual macro work orders;

[0010] Using the global identifier of the virtual macro work order, a batch locking request is initiated to the external third-party system to obtain the combined credential pool, and the resource credentials are distributed to the agent terminal based on the value weight allocation rule. For call sessions that have not been allocated resource credentials, the underlying signaling is redirected.

[0011] Furthermore, when evaluating the available attenuation slope and inflow slope, the resource response status corresponding to the same target service resource is grouped into a target service resource availability sequence according to the collection time. The entry queue events and manual waiting events associated with the target service resource are bucketed according to the same sampling window and counted as a call queue inflow count sequence. The two are then aligned with the same time axis and used as synchronous input objects for slope evaluation.

[0012] Furthermore, when constructing the target business resource availability sequence, if the external third-party system returns the lock increment but not the remaining resource quantity directly, the current remaining resource quantity is calculated by taking the most recent valid remaining resource quantity as the benchmark and combining the lock increment and release increment of the current sampling point; when constructing the call queue influx counting sequence, call sessions that have been withdrawn and have not re-entered the call queue are removed from the statistical results.

[0013] Furthermore, when fitting the intersection collision time, the last value of the target service resource availability sequence is first extracted as the current remaining resource quantity, and the number of call sessions that are still in the call queue at the current evaluation endpoint and whose service request category belongs to the target service resource is counted as the current associated queue quantity. Then, the intersection collision time is determined based on the current remaining resource quantity, the current associated queue quantity, the available decay slope, and the influx slope.

[0014] Furthermore, the preset buffer waiting time is determined based on the historical valid connection waiting time sequence corresponding to the target service resource, and the back pressure circuit breaker signal is bound to the target service resource identifier and the associated intent matching key for output; the associated intent matching key is formed by establishing a correspondence between the service request category and the target service resource identifier, and serves as the matching basis for subsequent screening of call sessions with an intent associated with the target service resource.

[0015] Furthermore, in response to the back pressure fuse signal blocking a call session, the call session that meets the associated intent matching key is removed from the manual connection assignment path and written to the suspended state, while the communication link of the relevant call session is reserved; when the same call session re-enters the call queue during the period when the back pressure fuse signal is in effect, the relevant call session is updated in place, and the original target service resource identifier and suspended state are retained.

[0016] Furthermore, when generating virtual macro work orders through clustering, call sessions in the suspended state are first grouped according to the target service resource identifier. Then, within the same target service resource, a time adjacency sequence is formed in ascending order of the suspended write time. Single-chain hierarchical clustering is performed using the difference in the suspended write time of adjacent call sessions as a distance metric. Different clusters are formed at positions where the adjacent difference exceeds the clustering time threshold. Clusters containing only a single call session are retained as the basis for work order generation.

[0017] Furthermore, when generating virtual macro work orders, call sessions within the same cluster are sorted in ascending order of their pending write times to form a call session sequence. A global identifier is generated based on the target service resource identifier, the back pressure circuit breaker signal generation time, and the beginning and end positions of the call session sequence. After any call session exits the communication link, the relevant call session is first removed from the corresponding cluster, and then a global identifier is regenerated for the remaining call sessions.

[0018] Furthermore, when initiating a batch lock request to an external third-party system, the global identifier, the target business resource identifier, the number of lock requests determined based on the call session sequence length, and the request initiation time are written into the same request batch. The lock results returned by the external third-party system are deduplicated by the unique credential identifier and screened for validity by the target business resource identifier to form a combined credential pool.

[0019] Furthermore, when the number of resource credentials in the combined credential pool is less than the number of call sessions in the call session sequence, a value weight vector is constructed based on the performance priority, suspension duration, and original order, and the call sessions bound to the resource credentials are determined by lexicographical order. For the remaining call sessions that have not been allocated resource credentials, the manual connection routing information is replaced, the underlying signaling is redirected, and the call sessions are transferred to the asynchronous compensation stream.

[0020] The technical effects and advantages of the intelligent management and control method for balanced dispatching of call center work orders in this invention are as follows:

[0021] This invention introduces the resource response status of an external third-party system into the call center work order allocation chain, so that the advancement of the call queue no longer depends solely on the queuing information on the communication side, but also simultaneously considers the availability trend of the target business resources and the influx trend of the call queue. Based on the pre-judgment chain formed by the availability decay slope, the influx slope, and the intersection collision time, risk identification can be completed before the target business resources become tight, and then a decision can be made on whether to trigger the back pressure circuit breaker signal, so that subsequent allocation actions are based on a unified resource status judgment.

[0022] After a call session enters a suspended state, this invention does not proceed with each call session independently. Instead, it clusters multiple suspended call sessions around the target business resource to generate a virtual macro work order, and uses the global identifier of the virtual macro work order to initiate a batch locking request to an external third-party system. This processing method reorganizes the originally scattered resource contention actions into batch processing objects targeting the same business resource. The correspondence between call sessions, virtual macro work orders, and combined credential pools is clear, providing a stable foundation for subsequent resource write-back, session routing, and status updates.

[0023] After the combined credential pool is formed, this invention further distributes resource credentials to agent terminals through value weight allocation rules, and performs underlying signaling redirection on call sessions that have not been allocated resource credentials. This allows call sessions to be directed to either a manual processing chain or an asynchronous compensation flow based on whether they have redeemable resources. As a result, the call center's connection targets are consistent with the external resource redemption results, and the interface relationship between agent terminal processing targets, the destination of unredeemed sessions, and subsequent compensation paths is clearer. Attached Figure Description

[0024] Figure 1 This is an overall flowchart of the intelligent management and control method for balanced dispatching of work orders in call centers according to the present invention;

[0025] Figure 2 This is a schematic diagram illustrating the evaluation of attenuation slope and inflow slope available in this invention;

[0026] Figure 3 This is a schematic diagram of the intersection collision time fitting and back pressure fuse signal generation of the present invention;

[0027] Figure 4 This is a schematic diagram illustrating the call session suspension, clustering, and virtual macro work order generation of the present invention;

[0028] Figure 5 This is a schematic diagram of the combined credential pool allocation and underlying signaling redirection of the present invention. Detailed Implementation

[0029] 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.

[0030] Please see Figures 1-5 This invention provides an intelligent management and control method for balanced dispatching of work orders in call centers, including:

[0031] S1: In response to the resource response status feedback from the external third-party system, assess the available decay slope of the target business resources and the influx slope of the call queue.

[0032] S2: Based on the available attenuation slope and the inflow slope, fit the intersection collision time, and trigger the back pressure fuse signal when the intersection collision time is within the preset buffer waiting time.

[0033] S3: In response to the back pressure fuse signal, intercept call sessions with the intent to associate with the target service resources and put them in a suspended state, and cluster multiple call sessions in the suspended state to generate virtual macro work orders;

[0034] S4: Using the global identifier of the virtual macro work order, initiate a batch locking request to the external third-party system, obtain the combined credential pool, and distribute the resource credentials to the agent terminal based on the value weight allocation rule, and perform underlying signaling redirection for call sessions that have not been allocated resource credentials.

[0035] Specifically, this invention uses the resource response status fed back by an external third-party system and the session change records in the call queue as joint inputs. First, it constructs a temporal change relationship between the resource side and the queue side around the target service resource to obtain intermediate results that can be used to judge the resource capacity and the queue influx trend. Then, it fits the intersection collision time with the intermediate results and outputs a back pressure circuit breaker signal when the target service resource enters the coverage of the preset buffer waiting time. This causes the call session with the intention associated with the target service resource to no longer proceed directly along the original manual connection and dispatch path, but instead enters a suspended state and generates a virtual macro work order by clustering according to the target service resource and the time proximity relationship. On this basis, it organizes external batch locking requests with the global identifier of the virtual macro work order and uses a combined credential pool to perform secondary diversion of the suspended call sessions. Call sessions that obtain resource credentials enter the agent terminal processing chain, while call sessions that do not obtain resource credentials are redirected to the underlying signaling and transferred to the asynchronous compensation stream.

[0036] In the intelligent management and control implementation method for balanced allocation of call center work orders, before the call session enters the back pressure circuit breaker and suspension aggregation processing, there are already two types of input foundations that can be directly invoked. One type is the resource response status continuously fed back by the external third-party system, and the other type is the arrival event flow and queuing event flow that have been formed in the call queue on the call center side. However, discrete status receipts and original queuing records alone are not enough to support the fitting of the intersection collision time in the subsequent step S2. Therefore, step S1 continues to expand on the basis of this input, and performs window normalization, slope solution and result output on the remaining available change trend of the target business resources and the influx change trend of the call queue, forming the available decay slope and influx slope that are directly referenced in the subsequent step S2.

[0037] S101. Resource response status is merged and target business resource timing is constructed.

[0038] Once the resource response status reported by the external third-party system arrives, the feedback record corresponding to the same target business resource will be written into the resource status sequence according to the collection time. The target business resource is defined as a business resource entity that can be actually occupied by the call session and controlled by the external third-party system. The resource status sequence is defined as a set of resource feedback records arranged in time sequence on a continuous time axis, which will be used to generate the availability change trajectory.

[0039] In specific processing, the target business resource identifier is used as the primary key to merge the remaining resource quantity, successful locking quantity, failed locking quantity, rate limiting response flag, and timeout response flag returned by the external third-party system. When multiple feedback records exist within the same sampling window, the record with the latest timestamp and without message truncation is retained as the status value of that sampling point. The rate limiting response flag or timeout response flag is mapped to the resource non-expansion constraint, which is only used to limit the subsequent interpretation range of availability recovery and does not change the name and boundary of the target business resource. After merging, the remaining resource quantity of each sampling point is extracted in chronological order within a fixed sliding time window to form the target business resource availability sequence, which serves as the unique input sequence for the availability decay slope.

[0040] In a call center scenario, if an external third-party system returns a lock increment instead of the remaining resource quantity between adjacent sampling points, the lock increment is first reverse-calculated based on the most recent valid remaining resource quantity to obtain the remaining resource quantity at the current sampling point. The result of the reverse calculation is still written into the same target business resource availability sequence, so that subsequent calculations always revolve around the unified target business resource availability, avoiding object drift during the call in step S2.

[0041] In one feasible implementation, the fixed sliding time window is taken as the median of the historical call connection waiting time sequence of the call center. The sampling step size is one-tenth of the fixed sliding time window and not less than 1 second. When the number of effective sampling points participating in the fitting is less than 5, an insufficient sampling mark is output and the calculation is delayed until the next sliding time window. During the construction of the target business resource availability sequence, if the external third-party system in the first... Each sampling point only returns the locked increment. and release increment Then take the first Remaining resources for each valid sampling point Based on, according to

[0042]

[0043] Update # Remaining resources at each sampling point ; if the release increment is not returned, take When performing three-point median filtering on the call queue influx counting sequence, only the second to the second-to-last sampling points are replaced in situ, with the replacement relationship being as follows:

[0044]

[0045] The first and last sampling points retain their original values.

[0046] S102. Call queue event stream sorting and inflow counting sequence construction.

[0047] After the available quantity sequence of the target business resource is established, the call center synchronously extracts the call sessions associated with the target business resource. The association here is based on the business request category that has been determined when the call session enters the queue. Call sessions that request the same flight cancellation or change, the same cabin class rebooking, or the same hotel inventory relocking, pointing to the same target business resource, are included in the same call queue statistics to avoid merging irrelevant sessions into the subsequent slope calculation.

[0048] In practice, events entering the queue, events transferring to human assistance, and events withdrawing from the queue are extracted from the call queue. Only events entering the queue and events transferring to human assistance that re-enter the queue within the statistical window are counted as inflow events. Sessions that have been withdrawn and have not re-entered the queue are removed. Subsequently, the event stream is bucketed using the same sliding time window as S101, and the number of newly entering call sessions corresponding to each sampling point is counted to obtain the call queue inflow count sequence. To ensure comparability, the sampling start point, sampling step size, and sampling end point of the call queue inflow count sequence are consistent with the target service resource availability sequence, and a target service resource identifier is added to each sampling point so that one call queue inflow count sequence corresponds to only one target service resource.

[0049] When a short-term jitter occurs within the statistical window, i.e., when the instantaneous arrival count of a single sampling point is abnormally amplified due to batch playback by the switch, the slope is not directly calculated using the value of that single point. Instead, a three-point median filter is used to replace the in-situ call queue influx counting sequence. The replaced sequence still maintains the original number of sampling points, only correcting the interference of sudden spikes on trend judgment, so that the influx slope obtained later can truly reflect the continuous influx trend of the call queue.

[0050] S103. Dual slope calculation, boundary determination and result output.

[0051] After the target service resource availability sequence and the call queue influx count sequence are aligned within the same window, first-order linear trend fitting is performed on the two sequences to obtain the available decay slope of the target service resources and the influx slope of the call queue. In order to make the calculation results directly applicable to the collision time fitting in step S2, the same fitting method is used to process the two types of sequences, that is, the least squares linear fitting is performed within their respective sliding time windows with the sampling time as the independent variable and the corresponding sampling value as the dependent variable.

[0052] For the sequence of available resources for the target business, let the first... The sampling time for each sampling point is The corresponding remaining resource quantity is Then the attenuation slope can be used. The following formula can be used to obtain:

[0053]

[0054] in, This is the average value of the sampling times within the current sliding time window. This represents the average number of resources remaining within the current sliding time window. This represents the number of sampling points used in the fitting process. Used to represent the rate of change of the availability of target business resources per unit of time, when The value of 'time' indicates that the target business resources are in a decaying state; the larger the absolute value, the faster the available resources decrease. At that time, the result is retained and output as a non-decayed state label, which is used by step S2 to perform boundary discrimination before collision time fitting.

[0055] For the call queue influx counting sequence, let the first... The number of incoming call sessions at each sampling point is The inflow slope was obtained under the same sampling time sequence. :

[0056]

[0057] in, This represents the average number of incoming call sessions within the current sliding time window. Used to characterize the rate of change of the call queue per unit time, when This indicates that the influx of call queue pressure is continuously increasing. After the calculation is complete, , The corresponding target service resource identifier, sliding time window boundary, and non-attenuation state marker are encapsulated together as the slope evaluation result. If the number of valid sampling points participating in the fitting is less than the preset minimum number of sampling points, an insufficient sampling marker is output, and the slope evaluation result corresponding to the target service resource is not sent to step S2 temporarily, but is recalculated after the next sampling window is completed. For target service resources that meet the minimum number of sampling points, the slope evaluation result is immediately written into the intersection collision time fitting input queue of the subsequent step S2 as the basis for the preliminary calculation of triggering the back pressure fuse signal.

[0058] After processing in step S1, the resource response status and call queue event stream fed back by the external third-party system are uniformly organized into a slope evaluation result that can be directly calculated and invoked. This result includes at least the target service resource availability sequence, the call queue influx count sequence, the availability decay slope, the influx slope, and the associated non-decay status marker and insufficient sampling marker. All of the above objects are uniquely mapped around the same target service resource. The subsequent step S2 uses this to perform intersection collision time fitting and buffer waiting time comparison, thereby continuing the backpressure fuse triggering signal determination chain under the condition of resource scarcity and continuous influx of queues.

[0059] After step S1 is completed, a sequence of available target service resources, a call queue influx count sequence, available attenuation slope, influx slope, non-attenuation status markers, and insufficient sampling markers have been formed around the same target service resource. The call center side also maintains the distribution of call sessions still in the queue at the corresponding time. However, these results are still trend representations and cannot be directly applied to call dispatch control. Therefore, step S2 continues based on this input, merging the slope evaluation result with the current queue occupancy status into a calculable collision fitting input, calculating the intersection collision time, and outputting a determination result based on the preset buffer waiting time to determine whether a back pressure fuse signal is triggered. The preset buffer waiting time is determined by the 90th quantile of the historical effective connection waiting time sequence arranged in ascending order. When the number of historical samples is not less than 100, the 90th quantile value is directly used. When the number of historical samples is less than 100, the arithmetic mean of the average connection waiting time and the maximum connection waiting time in the most recent billing cycle is used as the initial value.

[0060] S201. Collision Fitting Input Assembly and Effective Object Screening.

[0061] After outputting the slope evaluation results in step S1, the available attenuation slope, inflow slope, available target service resource sequence, and call queue inflow count sequence are read item by item using the target service resource identifier as an index. Target service resources marked as undersampled are directly kept in the pending sampling state and do not enter the subsequent fitting chain of this step. After this processing, all target service resources entering collision fitting have a sampling basis in the same window, and no further time axis misalignment will occur.

[0062] After the valid targets are identified, the last value of the available target service resources sequence is extracted from the current assessment endpoint as the remaining resource quantity. Simultaneously, the number of call sessions still in the call queue at the current assessment endpoint, whose service request category belongs to the target service resource, is counted to form the current associated queue quantity. The current associated queue quantity represents the number of call sessions under the same target service resource that have not yet been connected, withdrawn, and still occupy a position in the call queue. This quantity directly participates in the subsequent intersection collision time calculation. Current Associated Queue Quantity The number of relevant call sessions that are still in the call queue at the current evaluation endpoint is taken. Call sessions that have been connected, withdrawn, or terminated are excluded during the statistics. If the same call session enters the queue multiple times within the statistics window, only the latest record that is still in the queue is counted as 1.

[0063] Subsequently, the target service resource identifier, the current remaining resource quantity, the current associated queue quantity, the available decay slope, the influx slope, and the current evaluation endpoint are assembled into a collision fitting input set. If a non-decay status marker exists, the process does not exit immediately but is instead sent to the next sub-step for boundary judgment along with the current associated queue quantity. This is because if the current associated queue quantity has reached or exceeded the current remaining resource quantity even if the target service resource continues to decay, an immediate collision conclusion must still be given and the circuit breaker decision must be initiated.

[0064] S202. Intersection and Collision Time Fitting and Boundary Normalization.

[0065] After the collision fitting input group is established, the difference between the current remaining resource quantity and the current associated queue quantity is calculated to obtain the resource queue difference. The resource queue difference is used to characterize the remaining space that the target service resources can still handle at the current evaluation endpoint. When the difference is no greater than 0, it means that the number of related call sessions in the current queue has covered or exceeded the upper limit that the target service resources can handle. At this time, there is no need to push forward further. The intersection collision time is directly determined to be 0, and the result is sent to the next sub-step to participate in the back pressure fuse signal determination.

[0066] When the resource queue deficit is greater than 0, a resource queue deficit prediction function is constructed using the current evaluation endpoint as the starting point. This function does not regenerate a new resource sequence or queue sequence; instead, it directly references the available decay slope and inflow slope output in step S1, calculating the deficit convergence speed according to the same time dimension. The remaining space of the target service resource changes downward with the available decay slope, while the occupancy pressure of the call queue on this remaining space changes upward with the inflow slope. Therefore, the deficit convergence speed is determined by… Confirmed. Intersection and collision time. The following formula can be used to obtain:

[0067]

[0068] in, This represents the current remaining amount of resources, derived from the last value of the target business resource availability sequence; The current associated queue count is derived from the call queue statistics of the current evaluation endpoint; The available attenuation slope output from step S1; The inflow slope output in step S1; The intersection collision time represents the time point at which the resource queue difference first converges to 0, starting from the current evaluation endpoint. When the time is 0, it indicates that the difference has not contracted in the direction of the collision under the current trend. The time of the intersection collision is recorded as infinite and output as the unapproached state.

[0069] To ensure that subsequent judgments have a unique reference, after the calculation is completed, the intersection collision time is re-bound to the target business resource identifier to generate a collision fitting result. The collision fitting result only retains the quantities called in the current step and subsequent steps, namely the target business resource identifier, the intersection collision time, and the non-approximation state, and no longer carries static information unrelated to the judgment of this step.

[0070] S203. Preset buffer waiting time comparison and back pressure fuse signal output.

[0071] After obtaining the intersection collision time, the process proceeds to buffer boundary comparison. In this embodiment, the preset buffer waiting time is defined as the waiting boundary formed by taking the 90th percentile of the historical connection waiting time sequence in ascending order of the business scenario corresponding to the target business resource. Its dimension is consistent with the intersection collision time, and it is pre-written into the judgment rule set according to the target business resource. Step S2 reads the preset buffer waiting time and compares it item by item with the intersection collision time in the collision fitting result.

[0072] When the intersection collision time is 0, or the intersection collision time is greater than 0 but not greater than the preset buffer waiting time, it is determined that the target service resource will enter the collision interval before the call session is normally connected, and a backpressure circuit breaker signal is generated. When the intersection collision time is infinite, or the intersection collision time is greater than the preset buffer waiting time, the current target service resource remains in an untriggered state. The backpressure circuit breaker signal is directly bound to the target service resource identifier and written with an associated intent matching key of the same caliber as the target service resource. The associated intent matching key represents the basis for subsequent screening of call sessions with intents associated with the target service resource. Its source is consistent with the service request category caliber used when constructing the call queue influx counting sequence in step S1, so that step S3 can perform targeted interception of call sessions entering the call queue, rather than uniformly processing all call sessions. The associated intent matching key is generated by concatenating the service request category and the target service resource identifier and written into the call session queuing record. Subsequent steps S2 and S3 both use the associated intent matching key as the matching basis for screening call sessions with intents associated with the target service resource.

[0073] After the back pressure fuse is generated, the signal generation time and the corresponding intersection collision time are also output synchronously as the basis for the response in step S3. If the intersection collision time of the same target service resource is recalculated within a subsequent evaluation period and falls outside the preset buffer waiting time, step S2 will re-output a non-trigger result for the target service resource for subsequent call dispatch control link update and status determination; if it is still within the preset buffer waiting time, the back pressure fuse signal of the same target service resource will continue to be output, so that subsequent suspension and aggregation processing always revolve around the same target service resource in a closed loop.

[0074] After processing in step S2, the trend evaluation result formed in step S1 is further converged into a collision fitting result and a backpressure fuse signal that can directly drive allocation control. This includes at least the intersection collision time, non-approaching state, associated intent matching key, and backpressure fuse signal uniquely bound to the target service resource. These results are directly provided to subsequent steps to intercept call sessions with intents associated with the target service resource and place them in a suspended state, enabling subsequent virtual macro work order generation and batch locking processing to be based on the objects for which collision prediction has been completed.

[0075] Step S1 has already formed the available attenuation slope and inflow slope around the target service resource. Step S2 further outputs the intersection collision time, the associated intent matching key, and the back pressure circuit breaker signal uniquely bound to the target service resource. The call queue simultaneously retains related call sessions that have been in the queue and those that are continuously entering. However, simply completing the collision prediction is not enough to cut off subsequent invalid connection paths. Therefore, step S3 continues to expand on this result, performing targeted interception, hanging state writing, and clustering and merging on call sessions with intents associated with the target service resource, and forming virtual macro work orders that can be directly called in subsequent batch locking.

[0076] S301. Perform call session interception and suspension status writing based on the back pressure fuse signal.

[0077] After outputting the backpressure fuse signal in step S2, the associated intent matching key and signal generation time bound to the signal are read first according to the target service resource identifier. Then, two types of candidate objects are extracted from the call queue: one type is the call session that is still in the queue at the time of signal generation, and the other type is the call session that newly entered the call queue during the signal's effective period. Only after extraction is completed does the association judgment begin to avoid mistakenly merging ended or connected call sessions into the suspended chain.

[0078] In specific processing, the service request category determined when each call session enters the queue is read and matched one by one with the associated intent matching key. Call sessions that match are determined to have an intent associated with the target service resource, are immediately intercepted from the original manual dispatch path, and written to a suspended state. Simultaneously, the call session is removed from the ACD (Agent Dispatch Candidate) set, its status is switched from pending dispatch to suspended in the queue control table, and the communication link between the call session and the switching side is retained, thus preventing the ordinary manual dispatch process from taking over the call session. The suspended state indicates that the call session still maintains a communication link but is not yet in a controlled waiting state for manual agent assignment. This state is simultaneously bound to the target service resource identifier, the backpressure circuit breaker signal generation time, and the suspension writing time, serving as the time boundary basis for subsequent clustering and work order construction.

[0079] When the same call session re-enters the call queue multiple times during the period when the backpressure fuse is active, a new suspended object is not generated repeatedly. Instead, the original call session identifier is used to perform an in-situ update on the call session, overwriting the existing record with the latest suspension write time, while keeping the target service resource identifier and associated intent matching key unchanged. After this processing, a set of suspended call sessions is obtained. Each call session in the set of suspended call sessions has been removed from the manual connection and dispatch path and has a clear mapping established with a unique target service resource.

[0080] S302. Clustering and merging of suspended call sessions oriented towards target business resources.

[0081] After the set of suspended call sessions is formed, the target service resource identifier is used as the first-level grouping condition to separate and process suspended call sessions corresponding to different target service resources. After the first-level grouping is completed, within the same target service resource, the time proximity relationship is organized according to the effective interval corresponding to the generation time of the back pressure circuit breaker signal, thereby avoiding the incorrect aggregation of call sessions across resources and across trigger cycles into the same batch of objects.

[0082] Within the same target service resource group, a temporal adjacency sequence is constructed for all suspended call sessions. The temporal adjacency sequence represents a sequence of call sessions arranged in ascending order of their suspended write times. The time interval between two adjacent items in the sequence is used to determine whether they belong to the same cluster. Specifically, a single-chain hierarchical clustering rule is used, with the difference in suspended write times between adjacent call sessions as the distance metric. When the adjacent difference is not greater than the clustering time threshold, the two call sessions are merged into the same cluster; when the adjacent difference is greater than the clustering time threshold, they are split at that position to form a new cluster. The clustering time threshold is updated on a rolling daily basis, and each update recalculates the upper quartiles using the historical effective connection waiting time sequence of the previous 7 days.

[0083] After clustering, each cluster of call sessions simultaneously meets two conditions: they correspond to the same target service resource and are within the same time proximity range of continuous backpressure circuit breaker signal activation intervals. If a cluster contains only one call session, the cluster is retained and the process proceeds to the next sub-step, because the subsequent batch locking request requires a virtual macro ticket as the initiating object, and a virtual macro ticket can consist of one or more pending call sessions. Upon completion, the target service resource clustering results are output, consisting of multiple clusters, each of which can serve as the direct basis for generating virtual macro tickets.

[0084] S303. Virtual macro work order generation, global identifier writing and object call output.

[0085] After the target service resource clustering results are formed, virtual macro work orders are generated for each cluster. At this time, call sessions are no longer re-filtered. Instead, all call session identifiers, target service resource identifiers, and the corresponding back pressure circuit breaker signal generation times in the clusters are directly read to construct virtual macro work orders. A virtual macro work order is defined as a batch locking carrier formed by merging multiple suspended call sessions around the same target service resource. It subsequently serves as the sole business object for initiating batch locking requests to external third-party systems.

[0086] When generating a virtual macro work order, call session identifiers within the same cluster are first arranged in ascending order of their pending write times to form a call session sequence. Then, a global identifier is generated using the target service resource identifier, the backpressure circuit breaker signal generation time, and the first and last positions of the call session sequence. The global identifier represents a work order-level identifier that uniquely identifies the boundary of the virtual macro work order. Subsequent steps directly use this global identifier to initiate batch locking requests to external third-party systems and receive the returned combined credential pool using this global identifier. The global identifier is generated using a fixed concatenation format: "Target Service Resource Identifier - Backpressure Circuit Breaker Signal Generation Time - First Call Session Identifier - Last Call Session Identifier." When a call session within a cluster exits, causing changes to the first and last call session identifiers, a new global identifier is generated, overwriting the original one.

[0087] After writing the global identifier, the current processing status of all call sessions within the cluster is updated from suspended to macro work order aggregation, ensuring that each call session belongs to only one virtual macro work order and avoiding subsequent duplicate locking. At any given time, any call session is only allowed to be in one of the following states: suspended, macro work order aggregation, resource binding, or asynchronous compensation flow transfer. The state transition order is fixed as follows: from suspended to macro work order aggregation, and then, based on the resource credential allocation result, either entering the resource binding state or the asynchronous compensation flow transfer state.

[0088] After the virtual macro work order is generated, three types of subsequent call objects are output synchronously: the global identifier of the virtual macro work order, the target service resource identifier bound to the global identifier, and the call session sequence contained within the virtual macro work order. If a call session has disconnected from the communication link during the generation of the virtual macro work order, the call session is removed from the corresponding cluster before writing the call session sequence, and the remaining call sessions are reordered before the global identifier is written; if the cluster is empty after removal, the work order generation action for that cluster is canceled, and no virtual macro work order is output. After this processing, each virtual macro work order entering the subsequent steps corresponds to a set of valid call sessions that are still in a suspended state.

[0089] After processing in step S3, the backpressure fuse signal has been substantially transformed into an interception result at the call session level and an aggregation result at the work order level, forming at least a set of suspended call sessions, a clustering result of target service resources, a virtual macro work order, a global identifier, and a call session sequence bound to the virtual macro work order. All of the above objects establish a unique transmission relationship around the same target service resource. Subsequent steps can directly initiate batch locking requests to external third-party systems using the global identifier as the entry point, and use the call session sequence within the virtual macro work order as the corresponding basis for resource credential write-back and allocation.

[0090] Step S1 has already formed the available attenuation slope of the target service resources and the influx slope of the call queue. Step S2 further outputs the intersection collision time and back pressure fuse signal. Step S3, based on this determination, forms a virtual macro work order, a global identifier, and a call session sequence that are uniquely bound to the target service resources. However, the suspension aggregation only completes the concurrent intent folding and has not yet truly backfilled the available resources into the specific call session. Therefore, step S4 continues to unfold on the virtual macro work order generation result, and sequentially completes the batch lock request, the combined credential pool acquisition, the resource credential allocation and binding, the agent terminal distribution, and the underlying signaling redirection of the remaining call sessions.

[0091] S401. Batch lock request construction and combined credential pool generation.

[0092] After outputting the virtual macro work order in step S3, first read the global identifier, target service resource identifier, and call session sequence of the virtual macro work order, and determine the number of call sessions in the call session sequence. .in, This indicates the total number of call sessions participating in this resource redemption within the virtual macro work order. It is directly derived from the call session sequence length and will be used to compare and determine the number of returned resource vouchers.

[0093] After determining the quantity, the call center uses the global identifier as the batch primary key, and links the target business resource identifier and... A batch lock request is written together with the request and a one-time batch lock call request is initiated to an external third-party system. The request body includes at least a global identifier, a target business resource identifier, the number of locks requested, and the time the request was initiated. After the external third-party system returns the lock result, it first verifies whether the returned result belongs to the current virtual macro work order based on the global identifier. Then, it performs deduplication and validity screening on each successful lock record in the returned result. The deduplication is based on the unique credential identifier of the resource credential itself, and the validity screening is based on the target business resource identifier being consistent and being in a redeemable state. Through this processing, a combined credential pool is obtained. The combined credential pool represents the set of resource credentials returned by the same batch lock request and uniquely corresponding to the same virtual macro work order.

[0094] After the combined voucher pool is formed, the number of resource vouchers in it is counted and recorded as follows. If the external third-party system only returns partially valid lock results, then The number of resource certificates actually entering the combined certificate pool is determined by the actual number of such certificates. If the returned results contain invalid or duplicate certificates, these will be removed before entering the combined certificate pool and will not be counted. After processing, the output shows the pool of combined credentials and the number of resource credentials bound to the same global identifier. and the number of call sessions This serves as the direct input for subsequent allocation and binding.

[0095] S402. Quantitative Relationship Judgment and Value Weight Allocation Sequence Construction.

[0096] After the combined voucher pool is generated, the number of resource vouchers is first... With the number of call sessions They are compared under the same virtual macro work order scope, and subsequent execution paths are determined accordingly. If... If so, it is determined that all call sessions corresponding to the current virtual macro work order can obtain resource credentials, and the entire call session sequence enters the waiting-to-bind sequence; if Then, the preset value weight allocation rule is activated, and a value is selected from the call session sequence. One call session enters the waiting-to-bind sequence, and the remaining call sessions enter the remaining call session sequence; if If the sequence to be bound is empty, all call sessions will be directly merged into the remaining call session sequence.

[0097] When entering the value weight allocation rule, instead of reconstructing new call session objects, the system reads the service request category, suspension write time, and original order of each call session in the call session sequence to construct a value weight vector. The value weight vector represents a ternary priority structure used for sorting. Its first dimension is the fulfillment priority level, provided by a pre-stored resource fulfillment priority table indexed by the service request category. Each service request category corresponds to a unique integer level, with higher values ​​indicating higher priority. The second dimension is the suspension duration, calculated as the difference between the current allocation time and the suspension write time. The third dimension is the original order, directly taken from the position in the call session sequence. Subsequently, a lexicographically stable sorting algorithm is used to sort all call sessions according to fulfillment priority level from high to low, suspension duration from long to short, and original order from front to back. The sorted first... One call session is written to the pending binding sequence, and the remaining call sessions are written to the remaining call session sequence.

[0098] This process concretizes the value weight allocation rule into an executable sorting and selection process, eliminating the need for manual intervention. The number of call sessions in the pending binding sequence is always equal to the number of actually redeemable resource vouchers. The number of call sessions in the remaining call session sequence is Both continue to be bound to the same global identifier and target business resource identifier to maintain consistent object boundaries for subsequent bindings and redirections.

[0099] In one embodiment, the resource fulfillment priority table is pre-built and stored in the configuration library by the call center based on the configured service level agreement and existing business rules. During construction, a category set is first established around all business request categories corresponding to the same target business resource. Then, the categories are sorted according to their processing order in the resource fulfillment process. The processing order is based primarily on the fulfillment order predetermined in the service level agreement, and the manual handling order already in effect in the business rules is used as the correction basis at the same level. After sorting, unique integer levels are written to each business request category in descending order of the sorting results, forming a one-to-one correspondence between business request categories and fulfillment priority levels. This correspondence is then associated with the target business resource identifier and written into the resource fulfillment priority table. Subsequently, when generating the value weight vector, the fulfillment priority level in the resource fulfillment priority table is directly retrieved using the business request category corresponding to the call session and the target business resource identifier. If a new business request category is added to the configuration library, the fulfillment priority level under the corresponding target business resource is inserted and rearranged according to the same sorting rules.

[0100] S403. Resource credential binding and distribution to agent terminals.

[0101] Once the binding sequence is determined, all resource credentials in the combined credential pool are read, and a resource credential sequence is formed according to the order in which the resource credentials entered the combined credential pool. No secondary filtering of resource credentials is performed here; instead, the resource credential sequence and the binding sequence are directly matched one-to-one according to their positions to generate resource binding results. The resource binding result represents the deterministic binding relationship between the call session and the resource credentials. Each resource binding result contains at least a global identifier, a target service resource identifier, a call session identifier, and a unique resource credential identifier.

[0102] when At that time, all call sessions in the call session sequence participate in the binding; when At that time, only the first one in the binding sequence Each call session participates in the binding process. After binding is complete, the processing status of the corresponding call session is updated from macro work order collection status to resource binding status, and the resource binding result corresponding to the call session is written to the agent terminal connection queue. The agent terminal connection queue represents the set of call sessions that have redeemable resource credentials. The agent terminal will subsequently only extract call sessions from this set and restore the manual connection path, so that all call sessions entering the manual processing link carry clear resource credentials.

[0103] When writing to the agent terminal's call continuation queue, the batch writing order is maintained according to the global identifier in the resource binding result, and the target business resource identifier is simultaneously attached. This allows the agent terminal to directly call the corresponding resource credentials to complete subsequent business implementation after connection. After processing, the output is a resource binding result set that can be directly sent to the agent terminal, and the call continuation entry switch for the call session with the obtained resource credentials is completed.

[0104] S404. Underlying signaling redirection and asynchronous compensation stream transfer for remaining call sessions.

[0105] After the resource binding result is written, if the remaining call session sequence is not empty, then underlying signaling redirection is performed on each call session. At this time, the calling object is not the resource binding result set, but the remaining call session sequence explicitly retained in step S402. Each call session in the sequence is associated with the same global identifier, but has not obtained resource credentials, and therefore does not enter the agent terminal call continuation queue.

[0106] In specific processing, the call session identifier is used as an index to read the current manual call routing information occupied by the call session on the switching side, and this routing information is replaced with the asynchronous compensation flow access route to complete the underlying signaling redirection. The asynchronous compensation flow is defined as a non-real-time processing channel that takes over call sessions that have not been allocated resource credentials, and is used to receive subsequent callbacks, compensation notifications, or delayed fulfillment processing. After the route replacement is completed, the processing status of the corresponding call session is updated from the macro work order collection status to the asynchronous compensation flow transfer status, and the global identifier, target service resource identifier, and call session identifier are written into the asynchronous compensation flow registration result, so that subsequent compensation processing can be traced back according to the original virtual macro work order batch.

[0107] when When, all call sessions in the call session sequence undergo underlying signaling redirection according to the above path; when At that time, only the remaining call session sequence Each call session undergoes underlying signaling redirection. Since the redirection action directly affects the underlying signaling path, call sessions that have not obtained resource credentials exit the manual connection contention link after this step and no longer occupy the agent terminal's connection entry point. After this processing, the asynchronous compensation flow registration result is output as the final handling result for call sessions without allocated resource credentials in this step.

[0108] After processing in step S4, the virtual macro work order has been specifically transformed into a batch lockable object that can be responded to by an external third-party system and a resource redemption result that can be executed by the call session, forming at least a combined credential pool and a number of resource credentials. Number of call sessions The system records the pending binding sequence, the resource binding result set, and the asynchronous compensation flow registration result. At this point, call sessions that have obtained resource credentials are sent to the agent terminal, while call sessions that have not obtained resource credentials are redirected via underlying signaling and transferred to the asynchronous compensation flow. This completes the closed loop of the intelligent management and control chain for balanced dispatching of call center work orders.

[0109] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent management and control of balanced dispatching of work orders in call centers, characterized in that, Including the following steps: In response to the resource response status reported by the external third-party system, assess the available decay slope of the target business resources and the influx slope of the call queue. Based on the available attenuation slope and the inflow slope, the intersection collision time is fitted, and when the intersection collision time is within the preset buffer waiting time, the back pressure fuse signal is triggered. In response to the back pressure fuse signal, intercept call sessions with the intent to associate with the target service resources and put them in a suspended state, and cluster multiple call sessions in the suspended state to generate virtual macro work orders; Using the global identifier of the virtual macro work order, a batch locking request is initiated to the external third-party system to obtain the combined credential pool, and the resource credentials are distributed to the agent terminal based on the value weight allocation rule. For call sessions that have not been allocated resource credentials, the underlying signaling is redirected.

2. The intelligent management and control method for balanced dispatching of call center work orders according to claim 1, characterized in that, When evaluating the available attenuation slope and inflow slope, the resource response status corresponding to the same target service resource is grouped into a target service resource availability sequence according to the collection time. The entry queue events and manual waiting events associated with the target service resource are bucketed according to the same sampling window and counted as a call queue inflow count sequence. The two are then aligned with the same time axis and used as synchronous input objects for slope evaluation.

3. The intelligent management and control method for balanced dispatching of call center work orders according to claim 2, characterized in that, When constructing the target business resource availability sequence, if the external third-party system returns the lock increment but not the remaining resource quantity directly, the current remaining resource quantity is calculated by taking the most recent valid remaining resource quantity as the benchmark and combining the lock increment and release increment of the current sampling point; when constructing the call queue influx counting sequence, call sessions that have been withdrawn and have not re-entered the call queue are removed from the statistical results.

4. The intelligent management and control method for balanced dispatching of call center work orders according to claim 3, characterized in that, When fitting the intersection collision time, first extract the last value of the target service resource availability sequence as the current remaining resource quantity, and count the number of call sessions that are still in the call queue at the current evaluation endpoint and whose service request category belongs to the target service resource as the current associated queue quantity. Then, determine the intersection collision time based on the current remaining resource quantity, the current associated queue quantity, the available decay slope, and the influx slope.

5. The intelligent management and control method for balanced dispatching of call center work orders according to claim 4, characterized in that, The preset buffer waiting time is determined based on the historical effective connection waiting time sequence corresponding to the target service resource, and the back pressure fuse signal is bound to the target service resource identifier and associated intent matching key for output; The associated intent matching key is formed by establishing a correspondence between the service request category and the target service resource identifier, and serves as the basis for subsequent screening of call sessions with intents associated with the target service resource.

6. The intelligent management and control method for balanced dispatching of call center work orders according to claim 5, characterized in that, When a call session is blocked by a back pressure fuse signal, the call session that meets the associated intent matching key is removed from the manual connection assignment path and written to the suspended state, while the communication link of the relevant call session is reserved. When the same call session re-enters the call queue during the period when the back pressure fuse signal is in effect, the relevant call session is updated in place, and the original target service resource identifier and suspended state are retained.

7. The intelligent management and control method for balanced dispatching of call center work orders according to claim 6, characterized in that, When generating virtual macro work orders through clustering, call sessions in the suspended state are first grouped according to the target service resource identifier. Then, within the same target service resource, a time adjacency sequence is formed in ascending order of the suspended write time. Single-chain hierarchical clustering is performed using the difference in the suspended write time of adjacent call sessions as the distance metric. Different clusters are formed at positions where the adjacent difference exceeds the clustering time threshold. Clusters containing only a single call session are retained as the basis for work order generation.

8. The intelligent management and control method for balanced dispatching of call center work orders according to claim 7, characterized in that, When generating a virtual macro work order, call sessions within the same cluster are sorted in ascending order of their pending write time to form a call session sequence. A global identifier is generated based on the target service resource identifier, the back pressure fuse generation time, and the first and last positions of the call session sequence. After any call session exits the communication link, the relevant call session is first removed from the corresponding cluster, and then a global identifier is regenerated for the remaining call sessions.

9. The intelligent management and control method for balanced dispatching of call center work orders according to claim 8, characterized in that, When initiating a batch lock request to an external third-party system, the global identifier, the target business resource identifier, the number of lock requests determined based on the call session sequence length, and the request initiation time are written into the same request batch. The lock results returned by the external third-party system are deduplicated by the unique credential identifier and screened for validity by the target business resource identifier to form a combined credential pool.

10. The intelligent management and control method for balanced dispatching of call center work orders according to claim 9, characterized in that, When the number of resource credentials in the combined credential pool is less than the number of call sessions in the call session sequence, a value weight vector is constructed based on the performance priority, suspension duration and original order, and the call sessions bound to the resource credentials are determined by lexicographical order. For the remaining call sessions that have not been allocated resource credentials, the manual access routing information is replaced, the underlying signaling is redirected, and the call sessions are transferred to the asynchronous compensation stream.

Images