A full-process digital customer service system and method for small and medium-sized financial institutions

Abstract

The present invention discloses a full-process digitalization system and method for customer service of small and medium-sized financial institutions, specifically relating to the technical field of full-process digitalization of customer service. Through multi-modal precise understanding of customer requests and task process initialization, a deep semantic parsing model is introduced to extract the customer's main intention and secondary intention, and two key calculation indicators, "historical process tracing anomaly coefficient" and "abnormal closed-loop failure coefficient", are introduced to achieve dual quantitative evaluation of the execution stability of the customer service system and the accuracy of closed-loop judgment. The above two coefficients are organically integrated into the "pseudo-closed-loop deception response error index", and a unified quantitative model is established to determine whether there is a high-risk pseudo-closed-loop phenomenon in the customer service system. Based on the comparison of the pseudo-closed-loop deception response error index and the set threshold, real-time intelligent marking of the pseudo-closed-loop process is achieved, high-risk processes are accurately identified, and an early warning or re-judgment mechanism is dynamically triggered.

Description

Technical Field

[0001] The present invention relates to the technical field of full-process digitization of customer service, and more specifically, to a full-process digitization system and method for customer service of small and medium-sized financial institutions. Background Art

[0002] With the rapid development of FinTech, small and medium-sized financial institutions are increasingly prioritizing the digital upgrade of customer service processes to improve service response speed, reduce labor costs, and enhance customer satisfaction. Currently, customer service systems rely heavily on semantic recognition, business process modeling, and multi-channel data integration technologies to automate the entire process, from customer intent identification to task execution. However, under complex operating conditions such as asynchronous multi-channel interactions, ambiguous customer expressions, and delayed business process status awareness, customer service systems are prone to a combination of "process traceability anomalies" and "closed-loop misjudgment." This can lead to the system falsely indicating "completed" when the customer's request has not actually been closed. This issue, categorized as a "pseudo-closed-loop deceptive response error," manifests itself in: improper synchronization of process context and a failure of the task status marking mechanism. This leads to the system mistakenly marking an interrupted or unfinished service process as "completed" and returning positive feedback to the customer. This error not only severely disrupts the customer experience, leading to duplicate requests and disrupting service logic, but can also contaminate training data and mislead subsequent business strategy decisions, leading to customer complaints, increased compliance pressure, and increased operational costs. Therefore, there is an urgent need to build a full-process digitalization method for customer service of small and medium-sized financial institutions with the ability to model cross-channel process consistency, identify pseudo-closed-loop states, and perform context-aware compensation, so as to improve the service stability and response credibility of the system in complex scenarios. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a full-process digital system and method for customer service of small and medium-sized financial institutions to solve the problems raised in the above-mentioned background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A method for digitizing the entire customer service process of small and medium-sized financial institutions, comprising the following steps:

[0006] Step S1: Intervene in customer requests through a multimodal approach, identify the user's main intention and secondary intention using a deep semantic parsing model, and initialize the process nodes;

[0007] Step S2: calling the multi-channel context synchronization unit to integrate the customer's historical requests and processing trajectories on different channels, constructing the customer's personal process map, and obtaining a cross-channel task execution association map;

[0008] Step S3: Match the business process template required to execute the current customer intention according to the process map, record the status of each task node, and form a full-process log chain;

[0009] Step S4: parse the events in the entire process log chain, synchronously obtain the historical process tracing anomaly information, calculate the historical process tracing anomaly coefficient, and evaluate the degree of anomaly in the system's historical process tracing;

[0010] Step S5: Check whether the logical end of the process has reached the expected state, and simultaneously obtain abnormal closed-loop failure information to calculate the abnormal closed-loop failure coefficient to evaluate the misjudgment degree of the system closed-loop judgment;

[0011] Step S6: Build a pseudo-closed-loop fraud response error model based on the historical process tracing anomaly coefficient and the abnormal closed-loop failure coefficient, output a pseudo-closed-loop fraud response error index, and evaluate the current pseudo-closed-loop fraud response error level of the customer service full-process digital system;

[0012] Step S7: compare the pseudo closed-loop deception response error index with a preset pseudo closed-loop deception response error index threshold, and mark the pseudo closed-loop risk process.

[0013] In a preferred embodiment, in step S3, the business process template required to be executed by the current customer intention is matched according to the process map, specifically as follows: the user's intention is represented by a vector and its intention similarity is calculated with the starting intention vector in the business process template : ,in is the user intention vector, is the template intent vector; the intent similarity is compared with the preset intent similarity threshold. If the intent similarity is greater than the intent similarity threshold, the currently matched business process template is regarded as a candidate template; the process nodes in the candidate template are matched with the process map, and the process path deviation rate is calculated: deviation rate = number of non-matching nodes / total number of template nodes; the template with the lowest deviation rate is selected as the current task execution template.

[0014] In a preferred embodiment, the logic for obtaining the anomaly coefficient of historical process tracing is as follows:

[0015] Get historical process map collection Reconstruction process map set with system tracing is the total number of atlas sets; calculate the historical process atlas Reconstructing process flow chart with system tracing Flowchart of edit distance: ,in is the flow graph edit distance, Historical process map Convert to system traceability and reconstruction process map The number of edit operations, Historical process map The number of nodes, Reconstructing process maps for system traceability The number of nodes; calculate the execution path offset distance: ,in To perform path offset distance, Historical process map The length of the execution path in Reconstructing process maps for system traceability The length of the execution path in is the longest common subsequence length, ,in Historical process map The u-th execution operation in the execution path, Reconstructing process maps for system traceability The vth execution operation of the execution path;

[0016] Calculate the node frequency contribution of the historical process graph: ,in is the node frequency contribution, is the number of visits to the mth node in the execution path of the historical process map, and M is the number of all different nodes in the execution path; the node frequency contribution of the system retrospective reconstruction process map is calculated based on the same method. , calculate the contribution difference: ,in is the difference in contribution;

[0017] Calculate the anomaly coefficient of historical process tracing: ,in To trace the abnormal coefficient of historical process, are the preset proportional coefficients of flowchart edit distance, execution path offset distance, and contribution difference, respectively, and Both are greater than 0.

[0018] In a preferred embodiment, the logic for obtaining the abnormal closed-loop failure coefficient is as follows:

[0019] For each process Perform closed-loop state consistency judgment and obtain the closed-loop failure judgment value: ,in The closed loop fails to reach the judgment value. For the process The actual terminal state reached, For the process The expected closed-loop state set of When the value is equal to 1, obtain the process in the historical full process log chain The number of times the type has been successfully compensated for closed loops and process Total number of all abnormal closed-loop records of type , calculation process The probability of successfully closing the loop : , is a very small constant to prevent division by zero; calculate the abnormal closed loop failure coefficient: ,in is the abnormal closed-loop failure coefficient, is the total number of processes.

[0020] In a preferred embodiment, a pseudo-closed-loop deception response error model is constructed based on the historical process tracing abnormal coefficient and abnormal closed-loop failure coefficient, and a pseudo-closed-loop deception response error index is output. The formula based on the pseudo-closed-loop deception response error model is as follows: , where is the pseudo closed-loop deception response error index, To trace the abnormal coefficient of historical process, is the abnormal closed-loop failure coefficient, They represent the preset proportional coefficients of the historical process tracing anomaly coefficient and the abnormal closed loop failure coefficient, respectively, and Both are greater than 0.

[0021] In a preferred embodiment, the pseudo closed-loop fraud response error index is compared with a preset pseudo closed-loop fraud response error index threshold to mark the pseudo closed-loop risk process, as follows:

[0022] If the false closed-loop deception response error index is greater than the false closed-loop deception response error index threshold, the current process is marked as a high false closed-loop risk process;

[0023] If the pseudo closed-loop deception response error index is less than or equal to the pseudo closed-loop deception response error index threshold, there is no need to mark the current process.

[0024] In a preferred embodiment, a full-process digital customer service system for small and medium-sized financial institutions includes a customer intent recognition module, a business process modeling and execution module, a full-process log chain building module, a historical process tracing and analysis module, a closed-loop status determination module, a pseudo-closed-loop fraud response identification module, and a pseudo-closed-loop risk marking module;

[0025] The customer intent recognition module intervenes in customer requests in a multimodal manner, uses a deep semantic parsing model to identify the user's main intent and secondary intent, and initializes the process nodes;

[0026] The business process modeling and execution module calls the multi-channel context synchronization unit to integrate the customer's historical requests and processing trajectories across different channels, build a customer's personal process map, and obtain a cross-channel task execution association map;

[0027] The full-process log chain building module matches the business process template required to execute the current customer intention according to the process map, records the status of each task node, and forms a full-process log chain;

[0028] The historical process tracing analysis module analyzes events in the entire process log chain, simultaneously obtains historical process tracing anomaly information, calculates the historical process tracing anomaly coefficient, and evaluates the degree of anomaly in the system's historical process tracing;

[0029] The closed-loop state determination module detects whether the logical end of the process has reached the expected state, simultaneously obtains abnormal closed-loop failure information, calculates the abnormal closed-loop failure coefficient, and evaluates the degree of misjudgment of the system's closed-loop judgment;

[0030] The pseudo-closed-loop fraud response identification module builds a pseudo-closed-loop fraud response error model based on the anomaly coefficient and abnormal closed-loop failure coefficient of historical process tracing. It then outputs a pseudo-closed-loop fraud response error index to assess the current pseudo-closed-loop fraud response error level of the customer service full-process digital system.

[0031] The pseudo closed-loop risk marking module compares the pseudo closed-loop deception response error index with a preset pseudo closed-loop deception response error index threshold and marks the pseudo closed-loop risk process.

[0032] Technical effects and advantages of the present invention:

[0033] 1. The present invention realizes multimodal and precise understanding of customer requests and initialization of task processes through steps S1-S3. By introducing a deep semantic parsing model to extract the customer's main intention and secondary intention, a task execution association graph is constructed based on the integration of historical context, so that process establishment no longer relies on a single request representation, and has higher robustness and business adaptability. It is particularly suitable for high-frequency scenarios such as unclear customer expressions and repeated feedback from multiple channels, and significantly improves the accuracy of process modeling and pre-synchronization capabilities. Furthermore, steps S4 and S5 innovatively introduce two key calculation indicators: the "historical process tracing anomaly coefficient" and the "abnormal closed loop failure coefficient", to achieve a dual quantitative assessment of the customer service system's execution stability and closed loop judgment accuracy. The above two coefficients are organically integrated into the "pseudo-closed loop deception response error index", and a unified quantitative model is established to determine whether the customer service system currently has a high-risk pseudo-closed loop phenomenon. This breaks through the traditional judgment method based on single-point log status and achieves a comprehensive assessment of process-level integrity, state continuity, and the overlap of abnormal patterns. This index not only has the advantages of high precision and strong generalization, but also can provide the system with a clear and unambiguous risk level stratification mechanism, providing a reliable basis for subsequent policy intervention, manual confirmation, and service optimization. Based on the comparison of the pseudo-closed loop deception response error index with the set threshold, real-time intelligent tagging of pseudo-closed loop processes is achieved, high-risk processes are accurately identified, and early warning or re-judgment mechanisms are dynamically triggered. This processing strategy effectively prevents problems such as a surge in customer complaints, data pollution, and service logic confusion caused by "false closed loops", and improves the system's automatic repair capabilities and response credibility.

[0034] 2. In typical business scenarios of small and medium-sized financial institutions, especially under resource-constrained conditions, this invention can achieve stable operation of customer service systems and improve service reliability through optimal data fusion mechanisms and high-precision process modeling capabilities. Its significant advantages are: on the one hand, it achieves cross-channel process consistency modeling and improves the ability to identify process context integrity; on the other hand, by introducing dual anomaly coefficients and a comprehensive discrimination model, it constructs a pseudo-closed-loop identification and risk intervention closed-loop, significantly reducing the pseudo-closed-loop risk rate, enhancing the system's self-diagnosis capabilities and customer satisfaction, and providing a universal and practical intelligent upgrade path for the financial industry's customer service systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0036] Figure 1 This is a flow chart of the method of Example 1 of the present invention;

[0037] Figure 2 This is a flow chart of the system of Example 2 of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Example 1: Figure 1 The present invention provides a method for digitalizing the entire customer service process of small and medium-sized financial institutions, including the following steps:

[0040] Step S1: Intervene in customer requests through a multimodal approach, identify the user's main intention and secondary intention using a deep semantic parsing model, and initialize the process nodes;

[0041] Step S2: calling the multi-channel context synchronization unit to integrate the customer's historical requests and processing trajectories on different channels, constructing the customer's personal process map, and obtaining a cross-channel task execution association map;

[0042] Step S3: Match the business process template required to execute the current customer intention according to the process map, record the status of each task node, and form a full-process log chain;

[0043] Step S4: parse the events in the entire process log chain, synchronously obtain the historical process tracing anomaly information, calculate the historical process tracing anomaly coefficient, and evaluate the degree of anomaly in the system's historical process tracing;

[0044] Step S5: Check whether the logical end of the process has reached the expected state, and simultaneously obtain abnormal closed-loop failure information to calculate the abnormal closed-loop failure coefficient to evaluate the misjudgment degree of the system closed-loop judgment;

[0045] Step S6: Build a pseudo-closed-loop fraud response error model based on the historical process tracing anomaly coefficient and the abnormal closed-loop failure coefficient, output a pseudo-closed-loop fraud response error index, and evaluate the current pseudo-closed-loop fraud response error level of the customer service full-process digital system;

[0046] Step S7, comparing the pseudo closed-loop deception response error index with a preset pseudo closed-loop deception response error index threshold, and marking the pseudo closed-loop risk process;

[0047] Step S1: Intervene in customer requests through multimodal means (including voice, text, and images), and extract customer request metadata, including request time, access channel, customer ID, device type, and historical request ID, for subsequent process context association. A deep semantic parsing model (based on pre-trained Chinese language models such as BERT-CRF or ERNIE-PLM, combined with named entity recognition and sequence labeling) is used to identify the user's primary intent (main demand) and secondary intent (supplementary requests or constraints). A mapping is established between intent labels and business process templates, such as: reporting a lost bank card → Flow_ID_001; changing a mobile phone number → Flow_ID_007; checking the repayment date → Flow_ID_013.

[0048] Generate a standard process node sequence based on the matched business process template (e.g., "authentication → information confirmation → action execution → receipt generation"), mark the first node to be executed as the "current node," and append the intent recognition result as a process context input parameter to complete the process node initialization.

[0049] Step S2: Invoke the multi-channel context synchronization unit to access and standardize customer request data from different channels (app, webpage, WeChat, phone, counter, etc.). Fuzzy matching and trusted field joint modeling are used to generate a unified customer identity ID. For example, "Wang Ming - 123456 mobile phone" in the app and "W.Ming - 456 ID card" in the phone are identified as the same customer. The customer's historical requests and processing traces across different channels are integrated, sorted, and aligned to construct a customer interaction timeline, marking the causal and state transition relationships of key business nodes.

[0050] Based on the business type, processing action, and system response information in historical requests, process nodes are extracted from text records. Each node includes the following information: node ID (such as "N5"), action (such as "modify bound mobile phone number"), status (such as "execution successful"), channel (such as "APP"), and timestamp. Using the customer's ID as the primary key, the nodes are concatenated in chronological order. If there are jumps (such as requesting unlock without verification), nonlinear jump edges are established to complete the construction of the customer's personal process map and obtain a cross-channel task execution association graph.

[0051] In step S3, the business process template required to execute the current customer intention is matched according to the process map. Specifically, the user's intention is represented as a vector (based on BERT / BiLSTM representation) and the intent similarity is calculated between it and the starting intent vector in the business process template. : ,in is the user intention vector, The intent vector is the template; the intent similarity is compared with the preset intent similarity threshold. If the intent similarity is greater than the intent similarity threshold, the currently matched business process template is considered as a candidate template; the process nodes in the candidate template are matched with the process map, and the process path deviation rate is calculated: deviation rate = number of non-matching nodes / total number of template nodes; the template with the lowest deviation rate is selected as the current task execution template; a status listener is set for each task node to record the creation, execution, completion, failure and other status of each task node in real time. The status change is triggered by the backend service and database transaction;

[0052] It should be noted that each business process template consists of several standardized task nodes, with the following structure:

[0053] "Process ID": "F001";

[0054] "Process Name": "Bank Card Lost Report Process";

[0055] "Original Intention": "Report Lost";

[0056] "Process Node": [{"Node ID":"N1", "Name":"Authentication", "Input Requirements":["ID Card","Mobile Verification Code"], "Output":"Verification Passed"};

[0057] {"Node ID":"N2", "Name":"Loss / Loss Registration", "Dependent Node":"N1", "Output":"Loss / Loss Successful"};

[0058] {"node ID":"N3", "name":"Notify Customer", "dependent node":"N2", "output":"Completion Feedback"}];

[0059] Based on the current customer request, under the selected template, the state sequence executed from the beginning to the end is used as a full-process log chain, represented as an ordered linked list: , each It is an execution log of a task node;

[0060] Step S4: parse the events of the entire log chain, as follows:

[0061] Traverse the process log chain step by step to check whether it meets the legal state transition set Σ={initialization→executing→completed / failed};

[0062] Use depth-first search (DFS) to detect whether there are broken links in the process graph;

[0063] Synchronously obtain historical process tracing anomaly information to calculate the historical process tracing anomaly coefficient and evaluate the degree of anomaly in the system's historical process tracing;

[0064] The historical process tracing anomaly coefficient in this invention is a quantitative indicator used to measure the occurrence of process tracing errors, process state breakage, execution path anomalies, and closed-loop information loss in the customer service systems of small and medium-sized financial institutions. This coefficient reflects the system's ability to recover past task chains and the integrity of its logical backtracking when processing customer requests. The high or low historical process tracing anomaly coefficient directly indicates whether the system can accurately restore the user's historical interaction path, effectively identify node states, and properly connect contextual semantics. A high historical process tracing anomaly coefficient indicates significant problems in the customer service system's restoration of historical processes, such as missing process node logs, unexplained state jumps, broken dependency paths, and misjudgment of abnormal logical closed-loops. This indicates that the system's ability to trace past events is weak, making it difficult to make decisions based on the actual historical state when responding to current requests. This increases the risk of "false closed-loop" situations, where the system mistakenly believes a service process has been properly closed or the problem has been resolved, when in fact the customer request has not been completed or the problem has not been truly addressed. This can lead to serious consequences such as repeated complaints, suspended service, and business misjudgments. In this case, the customer service system can be easily "tricked" by erroneous process logs or false data, resulting in deceptive responses and misleading subsequent business judgment logic. On the contrary, if the historical process tracing anomaly coefficient is small, it means that the system has good process tracing capabilities when processing historical process maps, and can completely and accurately restore every process node and state transition requested by the customer. It relies on path coherence, stable state transmission, and no large-scale logical breaks or false closed-loop judgments. The system can provide high-quality intelligent responses to current requests based on the true historical state, significantly reducing the probability of false closed-loop responses and improving the consistency and accuracy of customer service. Based on the dynamic trend of the anomaly coefficient in historical process tracing, the system can perform risk assessments on potential pseudo-closed-loop fraudulent response errors, thereby achieving the following beneficial effects in actual deployment: First, it improves the system's sensitivity and ability to discern abnormal process behavior, enabling the system to not only detect whether there are anomalies in the current process, but also trace back to the source to discover the root cause of the anomaly; second, it provides a stable and quantitative basic input variable for the subsequent construction of the pseudo-closed-loop fraudulent response error index, making the index more credible and dynamically adjustable; third, it helps the customer service system have stronger process logic restoration capabilities when facing multiple concurrent requests and cross-channel task intersections, thereby strengthening process consistency, closed-loop accuracy, and customer satisfaction from the system architecture level; fourth, it provides a clear guidance path for manual supervision or audit intervention - the system can automatically mark processes with high levels of historical tracing anomalies for manual inspection, reducing the risk of business vulnerability spread; fifth, during the continuous learning and iterative optimization process of the customer service system, it provides a feedback mechanism to drive the adaptive evolution of the system model towards "low anomalies, strong traceability, and high closed-loop".

[0065] The logic for obtaining the anomaly coefficient for historical process tracing is as follows:

[0066] Get historical process map collection Reconstruction process map set with system tracing is the total number of atlas sets; calculate the historical process atlas Reconstructing process flow chart with system tracing Flowchart of edit distance: ,in is the flow graph edit distance, Historical process map Convert to system traceability and reconstruction process map The number of edit operations (the edit operations include insertion, deletion, and replacement of nodes / edges), Historical process map The number of nodes, Reconstructing process maps for system traceability The number of nodes; calculate the execution path offset distance: ,in To perform path offset distance, Historical process map The length of the execution path in Reconstructing process maps for system traceability The length of the execution path in is the longest common subsequence length, ,in Historical process map The u-th execution operation in the execution path, Reconstructing process maps for system traceability The vth execution operation of the execution path;

[0067] Calculate the node frequency contribution of the historical process graph: ,in is the node frequency contribution, is the number of visits to the mth node in the execution path of the historical process map, and M is the number of all different nodes in the execution path; the node frequency contribution of the system retrospective reconstruction process map is calculated based on the same method. , calculate the contribution difference: ,in is the difference in contribution;

[0068] Calculate the anomaly coefficient of historical process tracing: ,in To trace the abnormal coefficient of historical process, are the preset proportional coefficients of flowchart edit distance, execution path offset distance, and contribution difference, respectively, and All greater than 0;

[0069] It should be noted that the above formulas are all dimensionless and numerical calculations. Common dimensionless methods include Min-Max normalization and Z-Score normalization, which will not be described here. Set it according to the actual situation. For example, adopt the expert empowerment method, that is, invite experts in related fields to determine the preset proportion coefficients of various indicators through professional opinion surveys and comprehensive evaluations, for example, It can be 0.3, 0.4, 0.3;

[0070] In step S5, a set of task nodes that must be completed in the end is extracted from the business process template. Each end node should correspond to the "completed" or "closed" status defined by the business rules, and the process logic end is checked to see if it has reached the required status.

[0071] Synchronously obtain abnormal closed-loop failure information to calculate the abnormal closed-loop failure coefficient and evaluate the misjudgment degree of the system closed-loop judgment;

[0072] The Abnormal Closed-Loop Failure Coefficient in this invention measures the degree of abnormality in which the customer service full-process digital system fails to achieve the expected state at the logical endpoint of a business process. This coefficient comprehensively reflects the system's accuracy in determining process closure, the legitimacy of process node states, and the consistency of process endpoint states. It serves as a comprehensive quantitative indicator of the customer service system's execution integrity, process closure compliance, and tolerance for misjudgment. A large Abnormal Closed-Loop Failure Coefficient indicates that many business processes in the current system fail to complete effectively or terminate in unexpected states. This may be due to process interruptions, abnormal node states, or branch jump deviations, resulting in the process logic failing to close and the system mistakenly identifying it as completed. In this case, the system is very likely to provide customers with misleading feedback of "processed" or "completed." In reality, the customer request has not been closed-loop processed, ultimately leading to decreased customer satisfaction, repeated backlogs of issues, and delayed risk resolution, resulting in a typical "pseudo-closed-loop deceptive response" scenario. Conversely, a small Abnormal Closed-Loop Failure Coefficient indicates that the system has a strong ability to identify the process endpoint state and can correctly perceive and determine whether the business process is truly closed. The process node status is legal, the endpoint is logically continuous, and the system's response after task completion is highly reliable. Customer feedback is effectively addressed, ensuring high process consistency and service integrity. A low coefficient value corresponds to high closed-loop quality, which also means that the system has high decision-making accuracy and exception tolerance in determining process completion. Incorporating the abnormal closed-loop failure coefficient into the customer service system's process tracking and response mechanism effectively supports the dynamic assessment of the pseudo-closed-loop fraudulent response error index. This coefficient, along with the historical process tracing anomaly coefficient, constitutes two key structural indicators in the system's error response model. The former focuses on assessing the "logical completeness of the current process," while the latter focuses on the "reliability of historical process tracing." The combined effect of the two constructs a comprehensive picture of customer service response quality. By dynamically monitoring the abnormal closed-loop failure coefficient, the system can flag processes with potential pseudo-closed-loop risk in real time, proactively block the output of error information, and trigger a compensation mechanism to reschedule, re-evaluate, and verify the closed-loop status of unfinished tasks. This mechanism is particularly critical in scenarios such as finance, which have extremely high requirements for customer feedback accuracy and process auditability. It can significantly improve the intelligent closed-loop response capabilities of the customer service system, reduce the frequency of misleading responses, and improve the credibility of business process automation. It has extremely high practical value and promotion significance.

[0073] The logic for obtaining the abnormal closed-loop failure coefficient is as follows:

[0074] For each process Perform closed-loop state consistency judgment and obtain the closed-loop failure judgment value: ,in The closed loop fails to reach the judgment value. For the process The actual terminal state reached, For the process The expected closed-loop state set of When the value is equal to 1, obtain the process in the historical full process log chain The number of times the type has been successfully compensated for closed loops and process Total number of all abnormal closed-loop records of type , calculation process The probability of successfully closing the loop : , is a minimum constant to prevent division by zero (usually ); Calculate the abnormal closed loop failure coefficient: ,in is the abnormal closed-loop failure coefficient, is the total number of processes;

[0075] It should be noted that the above formulas are all dimensionless and numerical calculations. Common dimensionless methods include Min-Max normalization and Z-Score normalization, which will not be described here.

[0076] Step S6: Build a pseudo-closed-loop fraud response error model based on the historical process tracing anomaly coefficient and the abnormal closed-loop failure coefficient, output a pseudo-closed-loop fraud response error index, and evaluate the current pseudo-closed-loop fraud response error level of the customer service full-process digital system;

[0077] According to the historical process tracing anomaly coefficient and abnormal closed-loop failure coefficient, a pseudo-closed-loop deception response error model is constructed, and a pseudo-closed-loop deception response error index is output. The formula for the pseudo-closed-loop deception response error model is as follows: , where is the pseudo closed-loop deception response error index, To trace the abnormal coefficient of historical process, is the abnormal closed-loop failure coefficient, They represent the preset proportional coefficients of the historical process tracing anomaly coefficient and the abnormal closed loop failure coefficient, respectively, and All greater than 0;

[0078] It should be noted that the above formulas are all dimensionless and numerical calculations. Common dimensionless methods include Min-Max normalization and Z-Score normalization, which will not be described here. Set it according to the actual situation. For example, adopt the expert empowerment method, that is, invite experts in related fields to determine the preset proportion coefficients of various indicators through professional opinion surveys and comprehensive evaluations, for example, It can be 0.5, 0.5;

[0079] From the above calculation expression, it can be seen that the larger the historical process tracing anomaly coefficient and the larger the abnormal closed-loop failure coefficient, the larger the pseudo-closed-loop deception response error index, indicating that the customer service full-process digital system may have a higher risk of pseudo-closed-loop deception response at the current stage. Conversely, the smaller the historical process tracing anomaly coefficient and the smaller the abnormal closed-loop failure coefficient, the smaller the pseudo-closed-loop deception response error index, indicating that the customer service full-process digital system has a higher service closed-loop authenticity and response credibility in its current state, the system has a more accurate understanding of user intentions, good task process traceability capabilities, and a sound closed-loop logic judgment mechanism, and the overall operation is closer to the service goals of authenticity, efficiency, and consistency;

[0080] In step S7, the pseudo closed-loop deception response error index is compared with a preset pseudo closed-loop deception response error index threshold, and the pseudo closed-loop risk process is marked as follows:

[0081] If the false closed-loop deception response error index is greater than the false closed-loop deception response error index threshold, it indicates that the current process has significant anomalies in historical trajectory tracing and closed-loop achievement judgment. There may be a false closed-loop response risk where an unclosed-loop process is mistakenly judged as "service completed". The current process is marked as a high false closed-loop risk process.

[0082] If the false closed-loop deception response error index is less than or equal to the false closed-loop deception response error index threshold, it means that the execution trajectory of the current process is basically complete, the closed-loop state matching degree is high, the process map is logically continuous, and the task state changes are reasonable. No obvious false closed-loop response risk is found, and there is no need to mark the current process.

[0083] The present invention realizes multimodal and precise understanding of customer requests and initialization of task processes through steps S1-S3, extracts customer main intentions and secondary intentions by introducing a deep semantic parsing model, and constructs a task execution association graph based on the integration of historical context, so that process establishment no longer relies on a single request representation, and has higher robustness and business adaptability. It is particularly suitable for high-frequency scenarios such as unclear customer expressions and repeated feedback from multiple channels, and significantly improves the accuracy of process modeling and pre-synchronization capabilities. Furthermore, steps S4 and S5 innovatively introduce two key calculation indicators: the "historical process tracing anomaly coefficient" and the "abnormal closed loop failure coefficient", to achieve a dual quantitative assessment of the customer service system's execution stability and closed loop judgment accuracy. The above two coefficients are organically integrated into the "pseudo-closed loop deception response error index", and a unified quantitative model is established to determine whether the customer service system currently has a high-risk pseudo-closed loop phenomenon. This breaks through the traditional judgment method based on single-point log status and achieves a comprehensive assessment of process-level integrity, state continuity, and the overlap of abnormal patterns. This index not only has the advantages of high precision and strong generalization, but also can provide the system with a clear and unambiguous risk level stratification mechanism, providing a reliable basis for subsequent policy intervention, manual confirmation, and service optimization. Based on the comparison of the pseudo-closed loop deception response error index with the set threshold, real-time intelligent tagging of pseudo-closed loop processes is achieved, high-risk processes are accurately identified, and early warning or re-judgment mechanisms are dynamically triggered. This processing strategy effectively prevents problems such as a surge in customer complaints, data pollution, and service logic confusion caused by "false closed loops", and improves the system's automatic repair capabilities and response credibility.

[0084] In typical business scenarios of small and medium-sized financial institutions, especially under resource-constrained conditions, this invention can achieve stable operation of customer service systems and improve service reliability through optimal data fusion mechanisms and high-precision process modeling capabilities. Its significant advantages are: on the one hand, it achieves cross-channel process consistency modeling and improves the ability to identify process context integrity; on the other hand, by introducing dual anomaly coefficients and a comprehensive discrimination model, it constructs a pseudo-closed-loop identification and risk intervention closed-loop, significantly reducing the pseudo-closed-loop risk rate, enhancing the system's self-diagnosis capabilities and customer satisfaction, and providing a universal and practical intelligent upgrade path for the financial industry's customer service systems.

[0085] Example 2: This example is an introduction to a full-process digital customer service system for small and medium-sized financial institutions. Figure 2 As shown, it includes a customer intention recognition module, a business process modeling and execution module, a full-process log chain building module, a historical process tracing and analysis module, a closed-loop state determination module, a pseudo-closed-loop deception response recognition module, and a pseudo-closed-loop risk marking module;

[0086] The customer intent recognition module intervenes in customer requests in a multimodal manner, uses a deep semantic parsing model to identify the user's main intent and secondary intent, and initializes the process nodes;

[0087] The business process modeling and execution module calls the multi-channel context synchronization unit to integrate the customer's historical requests and processing trajectories across different channels, build a customer's personal process map, and obtain a cross-channel task execution association map;

[0088] The full-process log chain building module matches the business process template required to execute the current customer intention according to the process map, records the status of each task node, and forms a full-process log chain;

[0089] The historical process tracing analysis module analyzes events in the entire process log chain, simultaneously obtains historical process tracing anomaly information, calculates the historical process tracing anomaly coefficient, and evaluates the degree of anomaly in the system's historical process tracing;

[0090] The closed-loop state determination module detects whether the logical end of the process has reached the expected state, simultaneously obtains abnormal closed-loop failure information, calculates the abnormal closed-loop failure coefficient, and evaluates the degree of misjudgment of the system's closed-loop judgment;

[0091] The pseudo-closed-loop fraud response identification module builds a pseudo-closed-loop fraud response error model based on the anomaly coefficient and the abnormal closed-loop failure coefficient of historical process tracing. It then outputs a pseudo-closed-loop fraud response error index to assess the current pseudo-closed-loop fraud response error level of the customer service full-process digital system.

[0092] A pseudo closed-loop risk marking module compares the pseudo closed-loop deception response error index with a preset pseudo closed-loop deception response error index threshold to mark the pseudo closed-loop risk process;

[0093] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0094] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0095] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0096] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the system and method described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways.

[0098] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for digitizing the entire customer service process of small and medium-sized financial institutions, characterized by: The steps include: Step S1: Intervene in customer requests through a multimodal approach, identify the user's main intention and secondary intention using a deep semantic parsing model, and initialize the process nodes; Step S2: calling the multi-channel context synchronization unit to integrate the customer's historical requests and processing trajectories on different channels, constructing the customer's personal process map, and obtaining a cross-channel task execution association map; Step S3: Match the business process template required to execute the current customer intention according to the process map, record the status of each task node, and form a full-process log chain; Step S4: parse the events in the entire process log chain, synchronously obtain the historical process tracing anomaly information, calculate the historical process tracing anomaly coefficient, and evaluate the degree of anomaly in the system's historical process tracing; Step S5: Check whether the logical end of the process has reached the expected state, and simultaneously obtain abnormal closed-loop failure information to calculate the abnormal closed-loop failure coefficient to evaluate the misjudgment degree of the system closed-loop judgment; Step S6: Build a pseudo-closed-loop fraud response error model based on the historical process tracing anomaly coefficient and the abnormal closed-loop failure coefficient, output a pseudo-closed-loop fraud response error index, and evaluate the current pseudo-closed-loop fraud response error level of the customer service full-process digital system; Step S7, comparing the pseudo closed-loop deception response error index with a preset pseudo closed-loop deception response error index threshold, and marking the pseudo closed-loop risk process; The logic for obtaining the anomaly coefficient for historical process tracing is as follows: Get historical process map collection Reconstruction process map set with system tracing is the total number of atlas sets; calculate the historical process atlas Reconstructing process flow chart with system tracing Flowchart of edit distance: ,in is the flow graph edit distance, Historical process map Convert to system traceability and reconstruction process map The number of edit operations, Historical process map The number of nodes, Reconstructing process maps for system traceability The number of nodes; calculate the execution path offset distance: ,in To perform path offset distance, Historical process map The length of the execution path in Reconstructing process maps for system traceability The length of the execution path in is the longest common subsequence length, ,in Historical process map The u-th execution operation in the execution path, Reconstructing process maps for system traceability The vth execution operation of the execution path; Calculate the node frequency contribution of the historical process graph: ,in is the node frequency contribution, is the number of visits to the mth node in the execution path of the historical process map, and M is the number of all different nodes in the execution path; the node frequency contribution of the system retrospective reconstruction process map is calculated based on the same method. , calculate the contribution difference: ,in is the difference in contribution; Calculate the anomaly coefficient of historical process tracing: ,in To trace the abnormal coefficient of historical process, are the preset proportional coefficients of flowchart edit distance, execution path offset distance, and contribution difference, respectively, and All greater than 0; The logic for obtaining the abnormal closed-loop failure coefficient is as follows: For each process Perform closed-loop state consistency judgment and obtain the closed-loop failure judgment value: ,in The closed loop fails to reach the judgment value. For the process The actual terminal state reached, For the process The expected closed-loop state set of When the value is equal to 1, obtain the process in the historical full process log chain The number of times the type has been successfully compensated for closed loops and process Total number of all abnormal closed-loop records of type , calculation process The probability of successfully closing the loop : , is a very small constant to prevent division by zero; calculate the abnormal closed loop failure coefficient: ,in is the abnormal closed-loop failure coefficient, is the total number of processes.

2. A method for digitalizing the entire customer service process of small and medium-sized financial institutions according to claim 1, characterized in that: In step S3, the business process template required to execute the current customer intention is matched according to the process map, specifically as follows: the user's intention is represented by a vector and its intention similarity is calculated with the starting intention vector in the business process template : ,in is the user intention vector, is the template intent vector; the intent similarity is compared with the preset intent similarity threshold. If the intent similarity is greater than the intent similarity threshold, the currently matched business process template is regarded as a candidate template; the process nodes in the candidate template are matched with the process map, and the process path deviation rate is calculated: deviation rate = number of non-matching nodes / total number of template nodes; the template with the lowest deviation rate is selected as the current task execution template.

3. The method for digitalizing the entire customer service process of small and medium-sized financial institutions according to claim 1, characterized in that: According to the historical process tracing anomaly coefficient and abnormal closed-loop failure coefficient, a pseudo-closed-loop deception response error model is constructed, and a pseudo-closed-loop deception response error index is output. The formula for the pseudo-closed-loop deception response error model is as follows: , where is the pseudo closed-loop deception response error index, To trace the abnormal coefficient of historical process, is the abnormal closed-loop failure coefficient, They represent the preset proportional coefficients of the historical process tracing anomaly coefficient and the abnormal closed loop failure coefficient, respectively, and Both are greater than 0.

4. A method for digitalizing the entire customer service process of small and medium-sized financial institutions according to claim 3, characterized in that: The pseudo closed-loop deception response error index is compared with the preset pseudo closed-loop deception response error index threshold to mark the pseudo closed-loop risk process, as follows: If the false closed-loop deception response error index is greater than the false closed-loop deception response error index threshold, the current process is marked as a high false closed-loop risk process; If the pseudo closed-loop deception response error index is less than or equal to the pseudo closed-loop deception response error index threshold, there is no need to mark the current process.

5. A full-process digital customer service system for small and medium-sized financial institutions, used to implement the full-process digital customer service method for small and medium-sized financial institutions as described in any one of claims 1-4, characterized by: It includes customer intention recognition module, business process modeling and execution module, full-process log chain construction module, historical process tracing analysis module, closed-loop status determination module, pseudo-closed-loop deception response identification module, and pseudo-closed-loop risk marking module; The customer intent recognition module intervenes in customer requests in a multimodal manner, uses a deep semantic parsing model to identify the user's main intent and secondary intent, and initializes the process nodes; The business process modeling and execution module calls the multi-channel context synchronization unit to integrate the customer's historical requests and processing trajectories across different channels, build a customer's personal process map, and obtain a cross-channel task execution association map; The full-process log chain building module matches the business process template required to execute the current customer intention according to the process map, records the status of each task node, and forms a full-process log chain; The historical process tracing analysis module analyzes events in the entire process log chain, simultaneously obtains historical process tracing anomaly information, calculates the historical process tracing anomaly coefficient, and evaluates the degree of anomaly in the system's historical process tracing; The closed-loop state determination module detects whether the logical end of the process has reached the expected state, simultaneously obtains abnormal closed-loop failure information, calculates the abnormal closed-loop failure coefficient, and evaluates the degree of misjudgment of the system's closed-loop judgment; The pseudo-closed-loop fraud response identification module builds a pseudo-closed-loop fraud response error model based on the anomaly coefficient and abnormal closed-loop failure coefficient of historical process tracing. It then outputs a pseudo-closed-loop fraud response error index to assess the current pseudo-closed-loop fraud response error level of the customer service full-process digital system. The pseudo closed-loop risk marking module compares the pseudo closed-loop deception response error index with a preset pseudo closed-loop deception response error index threshold and marks the pseudo closed-loop risk process.

Images