Customer service quality inspection method and system

Abstract

The invention discloses a customer service quality inspection method and system, and the method comprises the steps: obtaining an interaction record of a customer service and a user and associated auxiliary information, and generating a dialogue situation portrait; based on the dialogue situation portrait, calling a plurality of pre-configured special intelligent agents with decoupling functions, and performing parallel analysis on the interaction record, each special intelligent agent being used for generating a dimension score from a corresponding specific function dimension and executing abnormal event monitoring related to the dimension; and summarizing the dimension score generated by each special agent and the monitored abnormal event, carrying out comprehensive calculation, and generating a quality inspection report. According to the method, multi-dimensional deep analysis is carried out on 100% customer service dialogues through parallel cooperation of a plurality of special agents, so that the quality inspection precision and objectivity are greatly improved. And the function of the special agent is decoupled, so that the flexible configuration of the architecture is realized, and the special agent with the corresponding function can be expanded or configured according to requirements.

Description

Technical Field

[0001] This invention relates to the field of intelligent customer service technology, and in particular to a customer service quality inspection method and system. Background Technology

[0002] Current customer service quality inspection primarily relies on manual spot checks and rule-based automated systems. Manual inspection is inefficient, has low coverage, and is heavily influenced by subjective factors. Traditional automated quality inspection systems often depend on pre-set keyword matching and simple speech-to-text analysis, failing to deeply understand the contextual semantics of the dialogue. Furthermore, existing systems often employ single, fixed analytical logic, making them unsuitable for complex and varied dialogue scenarios and unable to dynamically adjust quality inspection strategies for different business contexts. This results in unreasonable resource allocation and insufficient identification of key risks. Summary of the Invention

[0003] To address the shortcomings of existing customer service quality inspection technologies, such as low efficiency, superficial analysis dimensions, inability to verify in conjunction with business context, lack of real-time performance, rigid system architecture, and fixed strategies that cannot dynamically adapt to different scenarios, this invention proposes a customer service quality inspection method and system based on intelligent agents. This method not only enables full and accurate in-depth post-event analysis but also allows for real-time monitoring of the service process and the issuance of alerts when serious anomalies are detected, thereby achieving multi-dimensional and full-process management of service quality.

[0004] Firstly, the present invention proposes a customer service quality inspection method, comprising the following steps: S1. Obtain the interaction records and related auxiliary information between customer service and users to generate a dialogue context profile; S2. Based on the dialogue context profile, multiple pre-configured functionally decoupled specialized intelligent agents are invoked to perform parallel analysis on the interaction records. Each specialized intelligent agent is used to generate a dimension score from the corresponding specific functional dimension and to perform abnormal event monitoring related to that dimension. S3. Summarize the dimensional scores generated by each specialized intelligent agent and the abnormal events detected, perform comprehensive calculations, and generate a quality inspection report.

[0005] Preferably, S2 further includes: Based on the dialogue context profile, load the corresponding predefined quality inspection strategy template; The quality inspection strategy template predefines the weight configuration, scoring rules, and judgment criteria for each functional dimension for different dialogue scenario profiles. The specialized intelligent agent analyzes and detects anomalies in the interaction records based on the quality inspection strategy template.

[0006] Preferably, the multiple functional dimensions include at least two of the following: service process and information accuracy assessment, customer service attitude and professionalism evaluation, and user emotion recognition and reassurance effect analysis.

[0007] Preferably, the plurality of the specialized intelligent agents include: The process compliance intelligent agent is configured to assess the standardization of customer service actions based on a pre-built service process knowledge base. The information verification intelligent agent is configured to call the application programming interface of an external business system to verify the authenticity of the business information provided by customer service. A service attitude assessment agent is configured to analyze customer service language patterns and professionalism, and to use historical service data to evaluate the effectiveness of solutions. An emotion recognition agent is configured to identify changes in a user's emotional state during a conversation and to assess the effectiveness of customer service reassurance measures.

[0008] Preferably, in S3, the comprehensive calculation uses the following formula: in, This indicates the final overall score. This represents the weight of the i-th functional dimension; This represents the score for the i-th dimension given by the specialized intelligent agent.

[0009] Preferably, the "interaction records between customer service and users" in S1 include historical interaction records of ended conversations and ongoing real-time conversation streams.

[0010] Preferably, when the interaction record is a real-time dialogue stream, the method further includes: In the parallel analysis of S2, the abnormal events identified by any specialized intelligent agent are classified according to the predefined abnormal event judgment criteria in the quality inspection strategy template: If an abnormal event meets the severity level of the preset judgment criteria, an early warning message will be generated and issued in real time, and the special intelligent agent that identified the abnormal event will record and score it according to the scoring rules. If an abnormal event meets the general level of the preset judgment criteria, the specialized intelligent agent that identified the abnormal event will record and score it according to the scoring rules.

[0011] Secondly, the present invention also provides a customer service quality inspection system, the system including a central routing controller and multiple functionally decoupled specialized intelligent agents; The central routing controller is used to obtain the interaction records between customer service and users and related auxiliary information, generate dialogue context profiles, and schedule special intelligent agents based on the dialogue context profiles. The specialized intelligent agent is used to perform parallel analysis on the interaction records to output dimension scores and abnormal event monitoring results, which are then sent to the central routing controller. The central routing controller is also used to summarize the dimensional scores and abnormal event monitoring results and perform comprehensive calculations to generate the final quality inspection report.

[0012] This invention utilizes multiple specialized intelligent agents working in parallel and collaboratively to perform multi-dimensional, in-depth analysis of 100% of customer service conversations, significantly improving the accuracy and objectivity of quality inspection. The decoupling of specialized intelligent agent functions enables flexible architecture configuration, allowing for the expansion or configuration of specialized intelligent agents with corresponding functions as needed. Attached Figure Description

[0013] Figure 1 This is a flowchart of the customer service quality inspection method proposed in this invention. Detailed Implementation

[0014] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] Reference Figure 1 This invention proposes a customer service quality inspection method, which is executed by a customer service quality inspection system. The method includes: S1. Acquire customer service and user interaction records and related auxiliary information in real time or in batches. Use natural language processing technology to transform unstructured interaction records into structured text, and perform cleaning, sentence segmentation and role separation to reconstruct the complete dialogue context.

[0016] Auxiliary information refers to various related information beyond the interaction log, used to assist in determining the business scenario of the dialogue, the user's identity, or the type of question. This includes, but is not limited to: Interactive channel information: such as APP complaint channels, user calls; User attribute information: such as user ID, membership level, user profile, etc.; Business-related information: such as the order number associated with this interaction record, historical work order records, etc.; After acquiring the interaction records, the initial content of the auxiliary data and interaction records is analyzed at the beginning of the dialogue. For example, based on user information, all user information within the system is retrieved, such as user profiles and historical orders. This allows for the rapid generation of a dialogue context profile. The dialogue context profile is then built using current interaction information and auxiliary information such as user profiles.

[0017] For example, when a user reports a problem through the app's online customer service, the opening line might be "XXX is terrible." In this case, the user profile can be used to quickly identify the conversation context as a "general complaint."

[0018] For example, when a user makes a call through the hotline, the opening line might be "I would like to know about XXX". In this case, the conversation scenario profile can be identified as "a general business inquiry".

[0019] S2. Based on the dialogue context profile, call multiple pre-configured functionally decoupled specialized intelligent agents, and select the corresponding specialized intelligent agent to perform the quality inspection sub-task.

[0020] For example, several specialized intelligent agents include: The process compliance intelligent agent is configured to assess the standardization of customer service actions based on a pre-built service process knowledge base (SOP library).

[0021] The information verification intelligent agent is configured to call the application programming interface (API) of external business systems to verify the authenticity of business information provided by customer service, thereby determining the accuracy of the information provided. In existing technologies, traditional quality inspection systems are closed data silos, making it difficult to cross-validate information based on real-time access to external business data (such as user orders, flight status, and ticket information) according to the dialogue scenario. The information verification intelligent agent, by verifying external business data, can significantly improve the accuracy of quality inspection.

[0022] The service attitude assessment agent is configured to analyze customer service representatives' language patterns for standardization and professionalism, and can access historical service data to evaluate the effectiveness of solutions and the improvement of the customer service representatives.

[0023] An emotion-recognition agent is configured to track and identify changes in a user's emotional state during conversations and assess the effectiveness of customer service reassurance measures.

[0024] Each specialized intelligent agent can collaborate in parallel to analyze interaction records. Each specialized intelligent agent generates a dimension score from the corresponding specific functional dimension and performs anomaly event monitoring related to that dimension.

[0025] It should be noted that each specialized intelligent agent is centrally scheduled and controlled by a central routing controller. This central routing controller is used to invoke the corresponding specialized intelligent agents based on the dialogue context profile, and to summarize and comprehensively calculate the parallel analysis results of each specialized intelligent agent. The resulting system structure is flexible and can be expanded to include new specialized intelligent agents to cope with any new quality inspection dimensions.

[0026] S3. The analysis output of each specialized intelligent agent is quantified into dimensional scores. The dimensional scores generated by each specialized intelligent agent and the abnormal events detected are summarized. The central routing controller calculates the multi-dimensional scores based on the dynamic weights corresponding to the context profile, and finally generates a comprehensive score and the corresponding quality inspection report.

[0027] In this embodiment, the comprehensive calculation uses the following formula: in, This indicates the final overall score. This represents the weight of the i-th functional dimension; it is dynamically loaded based on the quality inspection strategy template corresponding to the dialogue scenario profile. This represents the score for the i-th dimension made by the specialized intelligent agent, which is calculated by the corresponding specialized intelligent agent based on the scoring details in the quality inspection strategy template and by comprehensively considering the abnormal events identified within this dimension.

[0028] Using the above method, this invention can achieve 100% full-scale in-depth quality inspection. Combined with external data verification and multi-dimensional quantitative scoring, the analysis accuracy and objectivity far exceed traditional methods. Furthermore, the parallel processing mechanism of multiple specialized intelligent agents significantly improves the system's processing efficiency, scalability, and analysis depth in complex, long-dialogue scenarios.

[0029] In a preferred embodiment, S2 further includes: Based on the dialogue context profile, load the corresponding predefined quality inspection strategy template; The quality inspection strategy template is a structured configuration file that predefines the weight configuration of each functional dimension, scoring details, and judgment criteria for various types of abnormal events for different dialogue context profiles. The specialized intelligent agent analyzes the interaction records based on the quality inspection strategy template.

[0030] For example, in one specific embodiment, the core functional dimensions include at least two of the following: service process and information accuracy assessment, customer service attitude and professionalism evaluation, and user emotion recognition and reassurance effect analysis.

[0031] For example, the functional dimensions of service process and information accuracy judgment can be completed collaboratively by service attitude assessment agents and information verification agents. For instance, in the airline ticketing industry, when a user inquires about flight delay and rebooking policies, the information verification agent automatically calls the "flight dynamics interface" to obtain order data and compares it with customer service's response to verify its accuracy. If the verified information fundamentally contradicts the customer service response (such as denying the existence of the order), it is recorded as a severe anomaly.

[0032] For example, the functional dimensions of customer service attitude and professionalism assessment can be implemented through a service attitude assessment agent.

[0033] For example, the functional dimensions of user emotion recognition and soothing effect analysis can be achieved through an emotion recognition intelligent agent.

[0034] In one optional implementation, the predefined weight configuration, scoring details, and criteria for judging abnormal events in the quality inspection strategy template are as follows: 1. Scoring calculation for functional dimensions

[0035] 1.1 Service process and information accuracy assessment

[0036] ①Calculation logic: =Process compliance score + Information accuracy score.

[0037] ② Process compliance score: Score = (Number of covered nodes / Total number of nodes to be covered) × 100 The process compliance intelligent agent checks the completeness of key process nodes based on a service process knowledge base. For example, nodes that should be covered may include greetings, confirmation of issues, solutions, and closing remarks.

[0038] ③ Information accuracy score: Score = 100 - ∑(Severity coefficient of each validation error × Deduction benchmark) This score is provided by the information verification agent. For example, customer service provides an incorrect rule (severity = 1.0, deduct 50 points); customer service provides an outdated policy (severity = 0.7, deduct 35 points).

[0039] 1.2 Customer Service Attitude and Professionalism Assessment

[0040] Calculation logic: =Basic etiquette score + communication skills score + professional knowledge demonstration score.

[0041] The evaluation is provided by a service attitude assessment agent. Evaluation criteria include the frequency of polite language, the proportion of negative / negative vocabulary, sentence fluency, whether jargon / standard terminology is used, and the degree of improvement in solutions to similar historical problems (by comparing historical work order data).

[0042] 1.3 Analysis of User Emotion Recognition and Soothing Effects

[0043] Calculation logic: D3 = Score for suppressing emotional deterioration + Score for problem-solving orientation.

[0044] The analysis was provided by an emotion recognition agent. Evaluation criteria included both emotional trajectory and effectiveness of reassurance.

[0045] The emotional trajectory includes the peak and duration of the user's negative emotions during the conversation, as well as whether the emotions stabilized after customer service intervention. The effectiveness of reassurance includes whether customer service recognized the user's emotions, used reassuring language, and steered the conversation towards problem-solving.

[0046] In a specific implementation, the event severity levels for abnormal events are classified as follows: ① Serious abnormal events (triggering real-time alerts), including: Business authenticity: Providing information that fundamentally contradicts the data returned by the application interface (e.g., denying the existence of an order).

[0047] Service red line: Customer service representatives use insulting or threatening language.

[0048] Compliance risks: Promising high compensation or leaking internal data without verification.

[0049] ② Common abnormal events, including: Inaccurate information: Providing incorrect policy details or non-real-time data.

[0050] Missing process: Key confirmation or notification steps are missing.

[0051] Poor attitude: repeatedly interrupting the user or showing impatience.

[0052] Each specialized intelligent agent scores corresponding dimensions based on identified anomalies within its respective functional dimension. For example, discrepancies in business authenticity or inaccurate information can be penalized by the information verification intelligent agent; customer service violations or poor attitudes can be penalized by the service attitude assessment intelligent agent; and customer service compliance risks or missing processes can be penalized by the process compliance intelligent agent. The central routing controller aggregates the dimensional scores from all specialized intelligent agents, assesses the accuracy of the penalty standards, and then performs a comprehensive calculation to obtain the final overall score.

[0053] 2. Context-aware dynamic weight configuration

[0054] Specifically, This mainly includes service processes and information accuracy. Customer service attitude and professionalism User emotion recognition and soothing effect For the same dialogue context profile, the total weight of the three functional dimensions is 1.

[0055] In specific implementations, the weight configuration for some common dialogue scenario profiles can be set as shown in the table below:

[0056] Among them, dialogue scenario profile 1 corresponds to high-value complaints, such as complaints from high-level customers or complaints about large orders; dialogue scenario profile 2 corresponds to ordinary business inquiries; dialogue scenario profile 3 corresponds to risk control sensitive conversations, such as account unblocking or orders placed by individuals with poor credit records; and dialogue scenario profile 4 corresponds to general complaints, such as order failures or payment failures.

[0057] For high-value complaints, it is crucial to prioritize improving the user experience; therefore, the weight given to user emotions and the effectiveness of reassurance should be carefully considered. To the highest level, avoid customer churn.

[0058] For general business inquiries, the main focus is on balancing efficiency and attitude, emphasizing accuracy and basic service standards; therefore, the weighting is... , Relatively high.

[0059] For risk-sensitive sessions, business security and compliance must be given utmost importance. The accuracy of processes and information is the bottom line and carries significant weight. Highest.

[0060] For general complaints, the focus is on resolving the actual problems quickly and accurately, therefore the weight is... Highest.

[0061] It should be noted that the specific weight configuration can be changed according to actual needs and is not limited to the example in the table above. The system dynamically loads matching quality inspection strategy templates through dialogue context profiling, automatically adjusts the analysis weights and anomaly thresholds of each functional dimension, and focuses quality inspection resources on key risk points, significantly improving the system's flexibility and the accuracy of risk identification.

[0062] It should be noted that the "interaction records between customer service and users" in S1 include historical interaction records of ended conversations as well as ongoing real-time conversation streams.

[0063] In a preferred embodiment, when the interaction record is a real-time dialogue stream, the method further includes: In the parallel analysis of S2, the abnormal events identified by any specialized agent are classified according to the predefined abnormal event judgment criteria in the quality inspection strategy template: If an abnormal event meets the severity level of the preset judgment criteria, and an abnormal event that may seriously damage the customer experience or the company's interests is detected (such as the information provided by customer service being inconsistent with the data returned by the application interface corresponding to the order information; or customer service using inappropriate language when the user is emotionally agitated), then the specialized intelligent agent that identified the abnormal event will generate and issue an early warning message in real time, notifying management personnel to intervene and handle the matter, thereby nipping the problem in the bud.

[0064] If an abnormal event meets the general level of the preset judgment criteria, the special intelligent agent will score and evaluate it in the corresponding functional dimension and record it as an abnormal deduction item.

[0065] All abnormal events will be identified and recorded and scored by the specialized intelligent agent for the abnormal event according to the scoring rules, and used as the basis for comprehensive calculation by the central routing controller in S3.

[0066] When the interaction record is a historical interaction record of an ended dialogue, each specialized intelligent agent directly performs its own dimension scoring, which is then aggregated to the central routing controller.

[0067] The central routing controller aggregates the analysis results, anomaly records, and final unified comprehensive score from all specialized intelligent agents, generating a structured quality inspection report that includes the original dialogue text, business data evidence, and scoring details. This report clearly displays the scores, weighting configurations, and deduction criteria for each dimension, and is used for performance management and service optimization.

[0068] This method employs real-time detection and dynamic early warning capabilities to intervene promptly in severe anomalies based on the dialogue context, effectively preventing customer experience deterioration and potential business losses. Furthermore, the early warning logic is deeply integrated with the scoring model, ensuring consistent evaluation. This method addresses the common problems in existing technologies, such as the lack of effective real-time monitoring capabilities, the inability to promptly detect and intervene in problems during service, and the inability to effectively prevent the deterioration of customer experience by only addressing issues after they occur.

[0069] This invention also provides a customer service quality inspection system for implementing the above-mentioned method. The system is based on an intelligent agent collaborative network with logical reasoning and the ability to actively invoke external interfaces to process customer service interaction records. The system supports both real-time streaming processing and post-processing batch processing modes, and has the ability to dynamically route and adjust quality inspection strategies based on context awareness. Specifically, the system includes a central routing controller and multiple functionally decoupled specialized intelligent agents.

[0070] The central routing controller embeds or manages a profile generation rule engine, which is used to obtain customer service and user interaction records and related auxiliary information, and generate dialogue context profiles based on the interaction records and auxiliary information through predefined rules and models.

[0071] For example, the central routing controller can define a "high-value customer complaint" profile when keywords such as "dissatisfaction" or "complaint" appear in the conversation and the user's identity is found to be a high-value user.

[0072] Based on the generated dialogue context profile, the central routing controller loads the corresponding predefined quality inspection strategy template (this template defines the weights of each dimension). According to the quality inspection strategy template, the central routing controller dynamically determines the type of specialized intelligent agent to be invoked, and completes its instantiation and task distribution to ensure that each specialized intelligent agent performs analysis based on a unified strategy benchmark.

[0073] Each specialized intelligent agent is functionally decoupled, strictly limited to working within its own functional dimension, independent of each other, and does not interfere with each other. It is only responsible for executing analysis tasks and outputting raw results, and does not participate in any global scheduling or comprehensive decision-making.

[0074] The core responsibility of specialized intelligent agents is to perform in-depth analysis and anomaly monitoring. Each specialized intelligent agent continuously analyzes interaction records during the dialogue process (including real-time dialogue streams) and, based on policy details issued by the central routing controller, generates real-time dimensional scores from specific functional dimensions (such as service process, information accuracy, customer service attitude and professionalism, and user emotion recognition and reassurance effectiveness). And monitor for abnormal events related to this dimension.

[0075] When an abnormal event is identified and reaches a severe level, the specialized intelligent agent directly generates and issues early warning information (such as pop-ups, notifications, etc.) to achieve real-time intervention.

[0076] Taking real-time dialogue streaming as an example, the system's workflow after startup is as follows: 1. Process Startup: The system obtains the interaction records, and the central routing controller generates a dialogue context profile in real time based on the built-in rules.

[0077] 2. Policy Scheduling: The central routing controller loads quality inspection policy templates based on dialogue context profiles and schedules relevant specialized intelligent agents to begin work. Each specialized intelligent agent performs parallel analysis, generating dimensional scores in real time and monitoring for anomalies. A direct warning is issued upon detecting significant anomalies.

[0078] 3. Post-event adjudication: Each specialized intelligent agent will summarize the final dimensional scores and detected anomalies to the central routing controller. The central routing controller will perform comprehensive calculations and generate a final quality inspection report based on this.

[0079] This invention provides accurate, data-supported, quantifiable, and scenario-appropriate feedback and scoring, forming a complete quality control closed loop from "dynamic strategy adaptation" and "in-process early warning" to "post-event quantitative evaluation and optimization," which significantly improves the overall service quality management level and the scientific nature of decision-making.

[0080] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A customer service quality inspection method, characterized in that, Includes the following steps: S1. Obtain the interaction records and related auxiliary information between customer service and users to generate a dialogue context profile; S2. Based on the dialogue context profile, multiple pre-configured functionally decoupled specialized intelligent agents are invoked to perform parallel analysis on the interaction records. Each specialized intelligent agent is used to generate a dimension score from the corresponding specific functional dimension and to perform abnormal event monitoring related to that dimension. S3. Summarize the dimensional scores generated by each specialized intelligent agent and the abnormal events detected, perform comprehensive calculations, and generate a quality inspection report.

2. The customer service quality inspection method according to claim 1, characterized in that, S2 further includes: Based on the dialogue context profile, load the corresponding predefined quality inspection strategy template; The quality inspection strategy template predefines the weight configuration, scoring rules, and judgment criteria for each functional dimension for different dialogue scenario profiles. The specialized intelligent agent analyzes and detects anomalies in the interaction records based on the quality inspection strategy template.

3. The customer service quality inspection method according to claim 2, characterized in that, Multiple functional dimensions include at least two of the following: service process and information accuracy assessment, customer service attitude and professionalism evaluation, and user emotion recognition and reassurance effect analysis.

4. The customer service quality inspection method according to claim 3, characterized in that, Several specialized intelligent agents include: The process compliance intelligent agent is configured to assess the standardization of customer service actions based on a pre-built service process knowledge base. The information verification intelligent agent is configured to call the application programming interface of an external business system to verify the authenticity of the business information provided by customer service. A service attitude assessment agent is configured to analyze customer service language patterns and professionalism, and to use historical service data to evaluate the effectiveness of solutions. An emotion recognition agent is configured to identify changes in a user's emotional state during a conversation and to assess the effectiveness of customer service reassurance measures.

5. The customer service quality inspection method according to claim 2, characterized in that, In S3, the comprehensive calculation uses the following formula: in, This indicates the final overall score. This represents the weight of the i-th functional dimension; This represents the score for the i-th dimension given by the specialized intelligent agent.

6. The customer service quality inspection method according to claim 2, characterized in that, The "Interaction Records between Customer Service and Users" in S1 includes historical interaction records of ended conversations as well as ongoing real-time conversation streams.

7. The customer service quality inspection method according to claim 6, characterized in that, When the interaction record is a real-time dialogue stream, the method further includes: In the parallel analysis of S2, the abnormal events identified by any specialized intelligent agent are classified according to the predefined abnormal event judgment criteria in the quality inspection strategy template: If an abnormal event meets the severity level of the preset judgment criteria, an early warning message will be generated and issued in real time, and the special intelligent agent that identified the abnormal event will record and score it according to the scoring rules. If an abnormal event meets the general level of the preset judgment criteria, the specialized intelligent agent that identified the abnormal event will record and score it according to the scoring rules.

8. A customer service quality inspection system, characterized in that, The system includes a central routing controller and multiple functionally decoupled specialized intelligent agents; The central routing controller is used to obtain the interaction records between customer service and users and related auxiliary information, generate dialogue context profiles, and schedule special intelligent agents based on the dialogue context profiles. The specialized intelligent agent is used to perform parallel analysis on the interaction records to output dimension scores and abnormal event monitoring results, which are then sent to the central routing controller. The central routing controller is also used to summarize the dimensional scores and abnormal event monitoring results and perform comprehensive calculations to generate the final quality inspection report.

Images