Omnichannel intelligent quality control system and monitoring method

Abstract

The invention relates to the technical field of intelligent customer service quality inspection, and discloses an omni-channel intelligent quality control system, and the system comprises a multi-source data normalization module which is used for obtaining a multi-mode original session data flow in real time through a preset interface, and converting the multi-mode original session data flow into a standardized format; the semantic feature analysis module is used for processing the standardized multi-modal original session data flow to generate an NLP semantic analysis result containing a time sequence tag; the quality inspection rule management module is used for storing and managing quality inspection rules, receiving NLP semantic analysis results and performing real-time mode recognition and risk assessment on the NLP semantic analysis results based on the quality inspection rules; the real-time quality inspection execution module is used for calculating a service quality index and a risk measurement index according to the risk assessment result returned by the quality inspection rule management module, and triggering real-time risk warning and monitoring large screen updating based on the risk measurement index; the application display module comprises a large monitoring screen; the display module is used for displaying required information.

Description

Technical Field

[0001] This manual relates to the field of intelligent customer service quality inspection technology, and in particular to an omnichannel intelligent quality control system and monitoring method. Background Technology

[0002] Intelligent customer service quality inspection, as a core component for improving customer service experience and managing business risks, hinges on accurate and timely monitoring and analysis of the entire omnichannel interaction process.

[0003] Existing technical solutions suffer from several shortcomings: First, traditional quality inspection systems often rely on post-event sampling from a single channel (such as telephone recordings alone), or can only simply aggregate data from different channels without achieving deep integration and real-time processing. This results in a one-sided quality inspection perspective, fragmented data, and difficulty in forming a globally unified view of service quality. Second, traditional quality inspections often focus on offline analysis and post-event judgment models, using a pre-set, slowly updated static rule base for result matching and scoring, ultimately generating periodic quality inspection reports. These solutions are essentially post-event auditing tools, with rigid decision-making logic and lengthy response cycles. They cannot support millisecond-level real-time assessment of service status and risk levels during conversations, and lack the ability to immediately trigger alarms and coordinate with business systems to enforce intervention when risks occur. Furthermore, the system's rules and models cannot self-optimize and dynamically evolve based on new business feedback and dialogue data, resulting in insufficient adaptability to new service problems and risk patterns.

[0004] Therefore, there is an urgent need for an omnichannel intelligent quality control system and monitoring method to overcome the channel isolation, lag in response, and static decision-making of traditional quality inspection, and to achieve real-time monitoring, accurate evaluation, and dynamic risk intervention of customer service quality. Summary of the Invention

[0005] In view of this, the present invention aims to propose an omnichannel intelligent quality control system and monitoring method to solve the problems of untimely quality monitoring risk response and weak operational decision support caused by the fragmentation of data sources, delayed analysis response, rigid decision-making mechanism and lack of risk intervention capability in existing customer service quality inspection systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A multi-channel intelligent quality control system, the system being used for quality monitoring and risk intervention in the service process of customer service channels, the system comprising: The multi-source data unification module is used to access telephone, online conversation and video channels in real time through preset interfaces to obtain multimodal raw conversation data streams, convert the multimodal raw conversation data streams into a standardized format, and output them to a message queue for downstream use.

[0007] The semantic feature analysis module is used to process the standardized multimodal raw session data stream in real time and generate structured NLP semantic analysis results containing time sequence labels.

[0008] The quality inspection rule management module is used to store and manage quality inspection rules, receive the NLP semantic analysis results, and perform real-time pattern recognition and risk assessment on the NLP semantic analysis results based on the quality inspection rules.

[0009] The real-time quality inspection execution module is used to calculate service quality indicators and risk measurement indicators based on the risk assessment results returned by the quality inspection rule management module, and to trigger real-time risk alarms and monitoring screen updates based on the risk measurement indicators.

[0010] The application display module includes a monitoring dashboard, which displays real-time service quality indicators and risk alerts across the entire platform.

[0011] The beneficial effects of this solution are as follows: In existing technologies, customer service quality monitoring generally suffers from isolated data sources, delayed processing responses, and the inability to proactively intervene during the process, thus hindering the real-time nature and effectiveness of customer service quality management and risk control. This system, through real-time collection and standardization of data from all channels, multimodal semantic feature parsing, real-time risk assessment driven by dynamic rules, and a dual-engine execution mechanism, achieves a leap from relying on manual post-event sampling to full-scale real-time intelligent control, effectively improving quality inspection coverage, risk response speed, and the accuracy of operational decisions.

[0012] Furthermore, the multi-source data normalization module includes: The data access unit is used to obtain voice streams from telephone channels through CTI interfaces or voice gateways; and to obtain text and image session data from online conversation channels and audio and video streams from video channels in real time by embedding SDKs or calling API interfaces.

[0013] The data standardization unit is used to convert the voice stream, text and image session data and audio and video streams acquired by the data access unit into a preset format and output them through the Kafka message queue.

[0014] Beneficial effects: The multi-source data unification module effectively solves the problem of data silos with scattered data sources and heterogeneous formats in customer service scenarios by building a unified, real-time, omnichannel data access and standardized pipeline. It provides high-quality, standardized multimodal data stream input for downstream real-time semantic analysis and intelligent decision-making, thus laying the data foundation for the system to achieve full-volume, real-time quality inspection and risk intervention, and significantly improving the timeliness and consistency of data processing.

[0015] Furthermore, the semantic feature analysis module includes: An automatic speech recognition service cluster is used to transcribe the speech stream and the audio stream extracted from the audio and video stream in real time to obtain transcribed text with timestamps and speaker separation identifiers.

[0016] The Natural Language Processing (NLP) service cluster is used to parse the transcribed text, the image and text conversation data, and the text information extracted from the audio and video streams, and output NLP semantic analysis results. The NLP semantic analysis results include at least sentiment analysis scores, customer intent recognition results, keyword hit information based on a preset lexicon, customer service semantic similarity, and sensitive entity recognition results.

[0017] Beneficial effects: By integrating automatic speech recognition and natural language processing clusters, the multimodal raw conversation data stream is transformed in real time into structured semantic analysis results rich in temporal and role information. This enables quantitative analysis of conversation content in multiple dimensions such as sentiment, intent, compliance, and risk entities, thereby providing accurate and comprehensive machine-understandable semantic features for subsequent rule matching and intelligent decision-making, greatly improving the system's ability to recognize and judge complex conversation scenarios.

[0018] Furthermore, the quality inspection rule management module includes: The visual rule configuration unit provides a graphical interface for users to define or modify quality inspection rules by dragging and dropping components.

[0019] The hot deployment and execution unit is used to support the real-time loading of newly configured or modified quality inspection rules without interrupting services, and to apply them to all real-time and historical session data.

[0020] The quality inspection rule template library contains pre-set general quality inspection rule templates.

[0021] Beneficial effects: By integrating visual zero-code configuration, hot deployment mechanism and pre-built rule template library, the system achieves efficient customization and uninterrupted updates of quality inspection rules, effectively solving the pain points of traditional quality inspection system rule maintenance relying on development, long update cycle and inability to trace back applications. This enables business personnel to directly participate in and lead the continuous optimization of quality inspection strategies, greatly improving the system's flexibility, response speed and operational efficiency.

[0022] Furthermore, the real-time quality inspection execution module includes a real-time session insight unit and a real-time risk control intervention unit configured in parallel.

[0023] The real-time session insight unit is used to call the rules in the quality inspection rule management module, calculate the service quality indicators based on the risk assessment results, and drive the monitoring dashboard to update.

[0024] The real-time risk control intervention unit is used to call the rules in the quality inspection rule management module, determine whether the risk level exceeds the preset threshold based on the risk assessment result, and generate and trigger a real-time risk alarm if the preset threshold is exceeded.

[0025] Beneficial Effects: By configuring real-time session insight units and real-time risk control intervention units in parallel, a dual-track parallel processing mechanism was achieved for quantitative assessment of service quality and immediate control of high-risk behaviors. This architecture upgrades the traditional post-event analysis and reporting model to a proactive operational model of in-event monitoring and intervention. On the one hand, it provides managers with a global dynamic view through real-time indicator calculation and large-screen driving; on the other hand, it enables proactive interception and immediate response to business risks through millisecond-level risk judgment and alarm triggering, thereby significantly improving the real-time performance and effectiveness of quality control.

[0026] Furthermore, the real-time risk alarm information triggered by the real-time risk control intervention unit includes a first type of alarm instruction and a second type of alarm instruction.

[0027] The first type of alarm instruction is a notification instruction, which is used to push alarm information containing risk session identifier and risk type to designated personnel.

[0028] The second type of alarm instruction is a session control instruction, which is sent to an external session control system to trigger at least one operation, including forced disconnection, mute, and supervisor intervention.

[0029] Beneficial Effects: By employing a dual-channel alarm mechanism that distinguishes between notification commands and session control commands, a complete closed loop from risk identification to intervention is achieved. This setup not only ensures managers' right to know in real time, but more importantly, it can directly link with external session control systems. When high-risk situations are detected, it automatically executes rigid intervention measures such as forced disconnection, muting, or supervisor intervention. This upgrades risk management from passive "early warning" to proactive "handling," effectively avoiding business losses and compliance risks caused by human delays or oversights, and significantly enhancing the system's proactive defense and real-time control capabilities.

[0030] Furthermore, the application demonstration module also includes: The management backend provides functions such as session retrieval and review, quality inspection task management, and quality inspection report generation.

[0031] The real-time alarm center is connected to the real-time risk control intervention unit and is used to receive and manage the real-time risk alarm information and push the real-time risk alarm information to designated personnel.

[0032] Beneficial effects: By integrating the management backend and the real-time alarm center, a visualized command platform integrating operations, analysis, and emergency response has been built. The management backend provides refined operational tools for all sessions, supports problem backtracking and task management, and realizes the systematization and measurability of quality inspection work; the real-time alarm center serves as a risk handling hub, ensuring that high-risk information is captured in real time, significantly improving operational efficiency and risk collaborative handling capabilities.

[0033] Furthermore, it also includes an intelligent learning and optimization module; the intelligent learning and optimization module includes an active learning unit, a model optimization unit, and a manual annotation platform.

[0034] The active learning unit is used to filter out session cases with confidence levels lower than preset standards based on the service quality indicators and risk measurement indicators, and push the session cases to be labeled to the manual labeling platform.

[0035] The model optimization unit is used to perform iterative training and version updates for the semantic feature analysis module based on the feedback annotation results of the manual annotation platform for the session cases to be annotated.

[0036] Beneficial effects: This module can accurately locate marginal cases where the system's judgment is ambiguous or has low confidence, guide manual intervention for annotation, and use high-quality feedback data for continuous training and optimization of semantic analysis models. This gradually improves the system's recognition and judgment accuracy for complex scenarios, new jargon, and unknown risks, enabling the system to dynamically adapt to business changes and achieve a leap from static rule execution to dynamic intelligent evolution. It effectively solves the problem of performance decay caused by the solidification of rules and models in traditional quality inspection systems.

[0037] A method for intelligent quality monitoring across all channels, the method comprising: S1. Access telephone, online conversation and video channels in real time through preset interfaces, obtain multimodal raw conversation data streams, and output the multimodal raw conversation data streams after converting them into a standardized format.

[0038] S2. Perform real-time processing on the standardized multimodal raw session data stream to generate structured NLP semantic analysis results containing time-series labels.

[0039] S3. Based on preset quality inspection rules, perform real-time pattern recognition and risk assessment on the NLP semantic analysis results; and calculate service quality indicators and risk measurement indicators based on the risk assessment results.

[0040] S4. Based on the aforementioned risk measurement indicators, trigger real-time risk alarms and monitor screen updates, and visualize the real-time service quality indicators and risk alarm information of the entire platform through the monitor screen.

[0041] Beneficial Effects: This method achieves systematic and intelligent monitoring and control of customer service quality by constructing a closed-loop process from real-time access to omnichannel data and multimodal semantic parsing to dynamic rule evaluation and dual-channel execution. It integrates the traditional discrete and lagging quality inspection process into a unified, real-time data-driven decision-making flow, significantly expanding the coverage and timeliness of quality monitoring. Furthermore, through proactive risk alerts and visualization, it elevates quality monitoring from passive auditing to proactive operation, thereby comprehensively improving service reliability, risk control capabilities, and operational decision-making efficiency. Attached Figure Description

[0042] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is an exemplary structural diagram of an omnichannel intelligent quality control system; Figure 2 This is an exemplary flowchart of an omnichannel intelligent quality monitoring method. Detailed Implementation

[0043] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0044] As indicated in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0045] Flowcharts are used in this specification to illustrate the operations performed by the system according to embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0046] The following detailed explanation illustrates the specific implementation methods: Example 1: Figure 1 This is an exemplary structural diagram of an omnichannel intelligent quality control system, such as... Figure 1 As shown, an omnichannel intelligent quality control system is used to monitor the quality and intervene in risks of the service process in customer service channels. The system includes: a multi-source data normalization module, a semantic feature analysis module, a quality inspection rule management module, a real-time quality inspection execution module, and an application display module.

[0047] The multi-source data unification module is used to access telephone, online conversation and video channels in real time through preset interfaces to obtain multimodal raw conversation data streams, convert the multimodal raw conversation data streams into a standardized format and output them to a message queue for downstream use.

[0048] Multimodal raw session data streams refer to real-time or near real-time communication data containing at least two modalities, including audio, text, images, and video. In this embodiment, the multi-source data unification module interfaces with the data output of an external customer service system through a preset interface to obtain multimodal raw session data streams.

[0049] Pre-defined interfaces are pre-defined application programming interfaces (APIs) or protocol interfaces used for data communication with external customer service systems. For example, real-time voice streams can be obtained from telephone exchanges via a CTI interface, text and image chat logs can be obtained from external customer service systems via WebSocket API or HTTP RESTful API, and audio and video streams can be obtained from video customer service systems via video streaming protocols such as RTMP, WebRTC, or a dedicated SDK.

[0050] External customer service systems refer to third-party software platforms or hardware integrations that provide customer service interaction functions. Examples include Avaya, Genesys, and Asterisk telephone exchange systems; Tencent Qidian, Zhichi Customer Service, and NetEase Qiyu online customer service platforms; and AnyChat, Zoom video conferencing systems, or their customized video customer service systems.

[0051] Message queues are middleware used to temporarily store and buffer data and enable asynchronous communication between upstream and downstream modules. Here, the multimodal raw session data stream is converted into a standardized format and output to the message queue for downstream use to avoid data congestion and achieve ordered data transmission and asynchronous processing.

[0052] Furthermore, the multi-source data unification module includes a data access unit and a data standardization unit.

[0053] The data access unit is used to obtain voice streams from telephone channels through CTI interfaces or voice gateways; and to obtain text and image session data from online conversation channels and audio and video streams from video channels in real time by embedding SDKs or calling API interfaces.

[0054] The data standardization unit is used to convert the voice stream, text and image session data, and audio and video streams acquired by the data access unit into a preset format and output them through the Kafka message queue.

[0055] In this embodiment, the preset format is a standard data format that is preset by the system and can be uniformly recognized by downstream modules.

[0056] In this embodiment, the preset format of the audio stream extracted from the voice stream and audio / video stream is PCM or MP3 format; the preset format of the transcribed text, the text-to-image conversation data, and the text information extracted from the audio / video stream is JSON format; the preset format of the video stream extracted from the audio / video stream is H.264 / H.265 encoded video stream format, and keyframe information of the video is extracted and the screen attributes are recorded in structured JSON format.

[0057] Kafka message queue is a high-throughput, low-latency distributed message middleware, mainly used for asynchronous transmission of large amounts of real-time data between system modules, ensuring that data is not lost and is transmitted in an orderly manner.

[0058] In this embodiment, technicians configure the Kafka cluster connection address and Topic in the data standardization unit; and encapsulate the converted voice streams, text and image session data, and audio / video streams into Kafka-supported message formats according to preset Topic categories; based on the encapsulated Kafka-supported message formats, the encapsulated messages are further sent to the corresponding Topic in the cluster in real time through Kafka client tools. At this time, downstream modules subscribe to the corresponding Topic and automatically pull this data from Kafka for processing.

[0059] The message format includes, but is not limited to, data content, timestamp, and identifier ID; the Kafka client tool can be a Java client.

[0060] In this embodiment, the multi-source data unification module effectively solves the problem of data silos with scattered data sources and heterogeneous formats in customer service scenarios by constructing a unified, real-time, omnichannel data access and standardized pipeline. This provides high-quality, standardized multimodal data stream input for downstream real-time semantic analysis and intelligent decision-making, thereby laying the data foundation for the system to achieve full-volume, real-time quality inspection and risk intervention, and significantly improving the timeliness and consistency of data processing.

[0061] The semantic feature analysis module is used to process the standardized multimodal raw session data stream in real time and generate structured NLP semantic analysis results containing time sequence labels.

[0062] NLP semantic analysis results refer to the set of structured features output after the standardized multimodal raw conversation data stream has been processed by natural language processing.

[0063] In this embodiment, the method for generating NLP semantic analysis results can be found in the following description.

[0064] Furthermore, the semantic feature analysis module includes: An automatic speech recognition service cluster is used to transcribe audio streams extracted from speech streams and audio / video streams in real time, resulting in transcribed text with timestamps and speaker separation identifiers.

[0065] In this embodiment, the automatic speech recognition service cluster uses an automatic speech recognition engine deployed on it to perform real-time speech-to-text conversion on the audio stream extracted from the speech stream and audio / video stream, thereby outputting transcribed text; wherein, the automatic speech recognition engine integrates at least one of the following enhanced processing capabilities during the transcription process: speaker separation, automatic noise reduction, and dialect recognition.

[0066] The Natural Language Processing (NLP) service cluster is used to parse text extracted from transcribed text, image-text conversation data, and audio / video streams, and output NLP semantic analysis results. These results include at least sentiment analysis scores, customer intent recognition results, keyword matching information based on a pre-defined lexicon, customer service semantic similarity, and sensitive entity recognition results.

[0067] Sentiment analysis scores are continuous or discrete numerical indicators used to quantify the emotional state of the person expressing the emotion.

[0068] Customer intent identification results are determined by analyzing customer statements and classifying their core needs or purposes into predefined categories. Examples include inquiries, complaints, and transaction processing.

[0069] Keyword hit information based on a preset thesaurus refers to the matching of specific words in the preset thesaurus in terms of their position, frequency, etc., during a conversation.

[0070] Customer service semantic similarity refers to the degree to which the actual customer service response is close to or conforms to the standard script at the semantic level.

[0071] Sensitive entity identification results refer to specific types of information fragments that involve personal privacy, trade secrets, or compliance risks.

[0072] In this embodiment, after receiving the text information, the natural language processing service cluster will perform unified text cleaning on the text information, including word segmentation and part-of-speech tagging; and perform multi-source text alignment and fusion based on a unified timeline and session identifier to form a unified text sequence with time sequence and source markers.

[0073] Multiple NLP tasks are performed in parallel on the preprocessed unified text sequence to extract structured semantic features.

[0074] First, a pre-trained sentiment analysis model is used to calculate the sentiment polarity score of a statement or conversation segment. For example, it can output a continuous value between -1 and 1. -1 represents extreme negativity, and 1 represents extreme positivity.

[0075] Sentiment analysis models can be fine-tuned models based on BERT. The customer service dialogue text is input into the model, and the final hidden state at the [CLS] position is used as the sentence representation. Then, a regression layer or classification layer is added. End-to-end optimization is performed by minimizing the loss function (such as mean squared error loss or cross-entropy loss) between the predicted sentiment score and the real sentiment label. Finally, a dedicated model that can accurately understand the sentiment expression in the customer service domain and output continuous sentiment scores is obtained.

[0076] Secondly, a text classification model is used to classify the intent of the text information, and the preset intent categories and their confidence levels are output. .

[0077] The text classification model can be a Transformer-based classifier. After inputting the client's sentence, the model uses the final hidden state at the [CLS] position as the semantic representation of the entire sentence, followed by a Softmax classification layer. Through training, the model learns to map the semantic representation of the input sentence to a predefined probability distribution of intent categories. The class with the highest probability is output as the intent to be recognized, and its corresponding probability value is the confidence score. The training objective of the model is to maximize the predicted probability of the correct intent category, which is typically achieved by minimizing the cross-entropy loss function.

[0078] Third, based on a pre-set business terminology database, such as prohibited words, product names, and service terms, it performs rapid scanning and matching, and records the matched keywords and their locations.

[0079] Use NER models, such as BiLSTM-CRF or Span-based recognition models, to identify sensitive entities and business entities in text.

[0080] Sensitive entities include PERSON (person's name), PHONE (telephone number), and ID (ID card number); business entities include ORDER_ID (order number) and PRODUCT (product model number).

[0081] Fourth, for customer service response text, a sentence encoding model is used to convert it into a vector representation. and vectors in the standard speech library The semantic similarity is derived by calculating the cosine similarity; the calculation logic is as follows: ; In the formula, It is a vector of the customer service response text; It is a vector from the standard speech library.

[0082] This formula is calculated. and The cosine similarity between them measures their directional consistency, which is used to assess the semantic closeness between the actual customer service response and the standard script.

[0083] as a result The range of values ​​for is [-1, 1]. If This indicates that the two vectors are in exactly the same direction, meaning the actual response is semantically identical to the standard response; if This indicates that the two vectors are orthogonal, meaning the actual response is semantically independent of the standard response; if This indicates that the two vectors are in completely opposite directions, meaning that the actual response is semantically completely opposite to the standard response.

[0084] All features extracted by the parallel tasks described above are encapsulated according to a predefined structured architecture, and timestamps and speaker identifiers of the corresponding text segments are added. The final output is a complete, machine-readable NLP semantic analysis result document.

[0085] In this embodiment, by integrating an automatic speech recognition and natural language processing cluster, the multimodal raw conversation data stream is transformed in real time into structured semantic analysis results rich in temporal and role information. This enables quantitative analysis of conversation content in multiple dimensions such as emotion, intent, compliance, and risk entities, thereby providing accurate and comprehensive machine-understandable semantic features for subsequent rule matching and intelligent decision-making, and greatly improving the system's ability to recognize and judge complex conversation scenarios.

[0086] The quality inspection rule management module is used to store and manage quality inspection rules, receive NLP semantic analysis results, and perform real-time pattern recognition and risk assessment on the NLP semantic analysis results based on the quality inspection rules.

[0087] Furthermore, the quality inspection rule management module includes: The visual rule configuration unit provides a graphical interface for users to define or modify quality inspection rules by dragging and dropping components.

[0088] In this embodiment, the visual rule configuration unit provides draggable logic components and configurable input controls on the front-end interface, allowing business personnel to build rule logic flows by connecting these components on a graphical canvas. After the rule configuration is completed, the editor backend compiles the graphical logic flow into a machine-executable JSON structured rule description, thereby enabling the customization, verification, and deployment of complex quality inspection rules without writing program code.

[0089] The draggable logic components can include "if", "and", and "or" to represent conditions, "then trigger" to represent operations, and nodes to represent data fields and operators.

[0090] Configurable input controls can include selecting an emotion threshold, entering keywords, and selecting an entity type.

[0091] The hot deployment and execution unit is used to support the real-time loading of newly configured or modified quality inspection rules without interrupting services, and to apply them to all real-time and historical session data.

[0092] In this embodiment, the hot deployment and execution unit maintains a rule engine instance running in memory and listens for rule change events. When it receives a newly configured or modified quality inspection rule, the unit uses dynamic class loading or script engine technology to compile or interpret the new rule logic into executable code in real time and inject it into the running rule engine instance, replacing or updating the original rule set. This process does not restart the service process. At the same time, the unit associates a snapshot of the currently effective rule version with each session being processed and immediately applies the new rules to newly arrived real-time session data. For historical session data streams that have not yet been processed, the new rules are applied retrospectively in subsequent processing nodes through a version-aware mechanism, thereby ensuring that rule changes can seamlessly and in real time cover all data.

[0093] The quality inspection rule template library contains pre-set general quality inspection rule templates.

[0094] The general quality inspection rule template is a pre-designed and packaged template for common customer service scenarios and universal quality inspection needs. For example, the service prohibition language template may include: if the customer service script contains negative words such as "no" or "I don't know", it will be marked as "inappropriate language"; the basic service process template may include: if no greeting appears within 10 seconds after the start of the conversation, it will be marked as "failure to greet in a timely manner", etc.

[0095] In this embodiment, by integrating visual zero-code configuration, hot deployment mechanism and pre-built rule template library, efficient customization and uninterrupted updates of quality inspection rules are achieved. This effectively solves the pain points of traditional quality inspection systems, such as rule maintenance relying on development, long update cycles and inability to trace back applications. It enables business personnel to directly participate in and lead the continuous optimization of quality inspection strategies, greatly improving the system's flexibility, response speed and operational efficiency.

[0096] The real-time quality inspection execution module is used to calculate service quality indicators and risk measurement indicators based on the risk assessment results returned by the quality inspection rule management module, and to trigger real-time risk alarms and monitoring dashboard updates based on the risk measurement indicators.

[0097] Service quality metrics are numerical or graded results used to quantify the service level of a single or batch of sessions. Examples include service duration, average response speed, overall sentiment score, script compliance rate, and business resolution rate.

[0098] Risk metrics are numerical or graded results used to quantify the degree and urgency of risk exposure in a single or batch of sessions. Examples include the percentage of risky sessions, the count of high-risk events, the overall risk score, and the slope of risk trends.

[0099] Furthermore, the real-time quality inspection execution module includes a real-time session insight engine and a real-time risk control intervention engine configured in parallel: The real-time session insight unit is used to call the rules in the quality inspection rule management module, calculate service quality indicators based on risk assessment results, and drive the monitoring dashboard to update.

[0100] In this embodiment, the real-time session insight unit uses a streaming computing framework to consume atomic judgment events from the risk assessment results in real time, and calculates them into multi-dimensional service quality indicators based on predefined aggregation and mapping rules. The indicator data stream is then pushed to the monitoring dashboard to drive real-time updates of the view.

[0101] The real-time risk control intervention unit is used to call the rules in the quality inspection rule management module, determine whether the risk level exceeds the preset threshold based on the risk assessment results, and generate and trigger a real-time risk alarm if the preset threshold is exceeded.

[0102] In this embodiment, the real-time risk control intervention unit processes the risk assessment results in real time through a streaming processing engine, calls rules to determine whether the risk level exceeds a preset threshold, and immediately generates an alarm event command and triggers its execution to link with external systems or notify relevant personnel.

[0103] Furthermore, the real-time risk alarm information triggered by the real-time risk control intervention unit includes first-type alarm instructions and second-type alarm instructions.

[0104] The first type of alarm instruction is a notification instruction, which is used to push alarm information containing risk session identifier and risk type to designated personnel.

[0105] The second type of alarm command is a session control command, which is sent to an external session control system to trigger at least one operation, including forced disconnection, mute, and supervisor intervention.

[0106] In this embodiment, a dual-channel alarm mechanism, distinguishing between notification commands and session control commands, achieves a complete closed loop from risk identification to intervention. This setup not only ensures managers' right to know in real time, but more importantly, it can directly link with the external session control system. When high-risk situations are detected, it automatically executes rigid intervention measures such as forced disconnection, muting, or supervisor intervention, thereby upgrading risk management from passive "early warning" to proactive "handling." This effectively avoids business losses and compliance risks caused by human delays or oversights, significantly enhancing the system's proactive defense and real-time control capabilities.

[0107] The application display module includes a monitoring dashboard, which displays real-time service quality indicators and risk alerts across the entire platform.

[0108] Furthermore, the application demonstration module also includes: The management backend provides functions such as session retrieval and review, quality inspection task management, and quality inspection report generation.

[0109] Among them, the conversation retrieval and replay features support searching historical conversations based on multiple criteria, playing recordings, viewing transcribed text, and AI analysis results.

[0110] The quality inspection task management system manages random inspection tasks and can recommend high-risk, low-scoring sessions for priority random inspection.

[0111] The quality inspection report generation function generates multi-dimensional quality inspection reports for individuals, teams, and business lines.

[0112] The real-time alarm center is connected to the real-time risk control intervention unit to receive and manage the real-time risk alarm information and push the real-time risk alarm information to designated personnel.

[0113] The real-time alarm center pushes critical alarms to designated personnel in real time via API, SMS, DingTalk, or Lark messages.

[0114] In this embodiment, the core of the application demonstration module is to build a visual control hub that integrates panoramic monitoring, in-depth operation and real-time response. It transforms the complex real-time analysis, rule matching and risk decision results at the bottom layer into an interactive, operable and traceable business interface and command channel, thereby upgrading the intelligent quality inspection system from a back-end analysis tool to a front-end operation command system, realizing a closed loop from data insight to management actions.

[0115] Furthermore, it also includes an intelligent learning and optimization module; the intelligent learning and optimization module includes an active learning unit, a model optimization unit, and a manual annotation platform.

[0116] The active learning unit is used to filter out session cases with confidence levels below a preset standard based on service quality indicators and risk measurement indicators, and then push these session cases to the manual annotation platform. The preset standard is set by humans based on experience.

[0117] The model optimization unit is used to iteratively train and update the semantic feature analysis module based on the feedback annotation results of the session cases to be annotated from the manual annotation platform.

[0118] By constructing a collaborative evolutionary closed loop of "data screening - manual confirmation of rights - model iteration", the system has the ability to continuously learn and self-optimize from business feedback, thereby breaking through the limitations of traditional quality inspection systems that rely on fixed rules and static models, and realizing dynamic improvement in quality inspection accuracy and scenario adaptability.

[0119] In this embodiment, the system achieves a leap from relying on manual post-event sampling to full-scale real-time intelligent control by real-time collection and standardization of data from all channels, multimodal semantic feature parsing, real-time risk assessment driven by dynamic rules, and a dual-engine execution mechanism. This effectively improves the quality inspection coverage, risk response speed, and accuracy of operational decisions.

[0120] Example 2: Figure 2 This is an exemplary flowchart of an omnichannel intelligent quality monitoring method. Figure 2 As shown, an omnichannel intelligent quality monitoring method includes: S1. Access telephone, online conversation and video channels in real time through preset interfaces, obtain multimodal raw conversation data streams, and output the multimodal raw conversation data streams after converting them into a standardized format.

[0121] S2. Perform real-time processing on the standardized multimodal raw session data stream to generate structured NLP semantic analysis results containing time-series labels.

[0122] S3. Based on preset quality inspection rules, perform real-time pattern recognition and risk assessment on the NLP semantic analysis results; and calculate service quality indicators and risk measurement indicators based on the risk assessment results.

[0123] S4. Based on the aforementioned risk measurement indicators, trigger real-time risk alarms and monitor screen updates, and visualize the real-time service quality indicators and risk alarm information of the entire platform through the monitor screen.

[0124] This method achieves systematic and intelligent monitoring and control of customer service quality by constructing a closed-loop process from real-time access to omnichannel data and multimodal semantic parsing to dynamic rule evaluation and dual-channel execution. It integrates the traditional discrete and lagging quality inspection process into a unified, real-time data-driven decision-making flow, significantly expanding the coverage and timeliness of quality monitoring. Furthermore, through proactive risk alerts and visualization, it elevates quality monitoring from passive auditing to proactive operation, thereby comprehensively improving service reliability, risk control capabilities, and operational decision-making efficiency.

[0125] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

[0126] Furthermore, unless expressly stated in the claims, the order of processing elements and sequences, the use of numbers and letters, or other names described in this specification are not intended to limit the order of the processes and methods described herein. Although various examples have been discussed in the foregoing disclosure of some embodiments of the invention that are currently considered useful, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments; rather, the claims are intended to cover all modifications and equivalent combinations that conform to the spirit and scope of the embodiments described herein. For example, while the system components described above can be implemented using hardware devices, they can also be implemented solely using software solutions, such as installing the described system on existing servers or mobile devices.

[0127] Similarly, it should be noted that, in order to simplify the description disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments in this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0128] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of this specification are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0129] For each patent, patent application, patent application publication, and other material such as articles, books, specifications, publications, and documents referenced in this specification, the entire contents of which are incorporated herein by reference. This excludes historical application documents that are inconsistent with or conflict with the content of this specification, as well as documents that limit the broadest scope of the claims in this specification (currently or subsequently appended to this specification). It should be noted that in the event of any inconsistency or conflict between the descriptions, definitions, and / or terminology used in the supplementary materials to this specification and the content of this specification, the descriptions, definitions, and / or terminology used in this specification shall prevail.

[0130] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A multi-channel intelligent quality control system, the system being used for quality monitoring and risk intervention in the service process of customer service channels, characterized in that, The system includes: The multi-source data unification module is used to access telephone, online conversation and video channels in real time through preset interfaces to obtain multimodal raw conversation data streams, convert the multimodal raw conversation data streams into a standardized format, and output them to a message queue for downstream use; The semantic feature analysis module is used to process the standardized multimodal raw session data stream in real time and generate structured NLP semantic analysis results containing time sequence labels. The quality inspection rule management module is used to store and manage quality inspection rules, receive the NLP semantic analysis results, and perform real-time pattern recognition and risk assessment based on the quality inspection rules. The real-time quality inspection execution module is used to calculate service quality indicators and risk measurement indicators based on the risk assessment results returned by the quality inspection rule management module, and to trigger real-time risk alarms and monitoring screen updates based on the risk measurement indicators. The application display module includes a monitoring dashboard, which displays real-time service quality indicators and risk alerts across the entire platform.

2. The system according to claim 1, characterized in that, The multi-source data normalization module includes: The data access unit is used to obtain voice streams from telephone channels through CTI interfaces or voice gateways; and to obtain text and image session data from online conversation channels and audio and video streams from video channels in real time by embedding SDKs or calling API interfaces. The data standardization unit is used to convert the voice stream, text and image session data and audio and video streams acquired by the data access unit into a preset format and output them through the Kafka message queue.

3. The system according to claim 2, characterized in that, The semantic feature analysis module includes: An automatic speech recognition service cluster is used to transcribe the speech stream and the audio stream extracted from the audio and video stream in real time to obtain transcribed text with timestamps and speaker separation identifiers; The Natural Language Processing (NLP) service cluster is used to parse the transcribed text, the image and text conversation data, and the text information extracted from the audio and video streams, and output NLP semantic analysis results. The NLP semantic analysis results include at least sentiment analysis scores, customer intent recognition results, keyword hit information based on a preset lexicon, customer service semantic similarity, and sensitive entity recognition results.

4. The system according to claim 3, characterized in that, The quality inspection rule management module includes: The visual rule configuration unit provides a graphical interface for users to define or modify quality inspection rules by dragging and dropping components. The hot deployment and execution unit is used to support the real-time loading of newly configured or modified quality inspection rules without interrupting services, and to apply them to all real-time and historical session data. The quality inspection rule template library contains pre-set general quality inspection rule templates.

5. The system according to claim 4, characterized in that, The real-time quality inspection execution module includes a real-time session insight unit and a real-time risk control intervention unit set in parallel. The real-time session insight unit is used to call the rules in the quality inspection rule management module, calculate the service quality indicators based on the risk assessment results, and drive the monitoring dashboard to update. The real-time risk control intervention unit is used to call the rules in the quality inspection rule management module, determine whether the risk level exceeds the preset threshold based on the risk assessment result, and generate and trigger a real-time risk alarm if the preset threshold is exceeded.

6. The system according to claim 5, characterized in that, The real-time risk alarm information triggered by the real-time risk control intervention unit includes a first type of alarm instruction and a second type of alarm instruction. The first type of alarm instruction is a notification instruction, used to push alarm information containing risk session identifier and risk type to designated personnel; The second type of alarm instruction is a session control instruction, which is sent to an external session control system to trigger at least one operation, including forced disconnection, mute, and supervisor intervention.

7. The system according to claim 6, characterized in that, The application demonstration module also includes: The management backend provides functions for session retrieval and review, quality inspection task management, and quality inspection report generation. The real-time alarm center is connected to the real-time risk control intervention unit and is used to receive and manage the real-time risk alarm information and push the real-time risk alarm information to designated personnel.

8. The system according to claim 7, characterized in that, It also includes an intelligent learning and optimization module; the intelligent learning and optimization module includes an active learning unit, a model optimization unit, and a manual annotation platform; The active learning unit is used to filter out session cases with confidence levels lower than preset standards based on the service quality indicators and risk measurement indicators, and push the session cases to be labeled to the manual labeling platform. The model optimization unit is used to perform iterative training and version updates for the semantic feature analysis module based on the feedback annotation results of the manual annotation platform for the session cases to be annotated.

9. A method for omnichannel intelligent quality monitoring, used to construct an omnichannel intelligent quality control system as described in any one of claims 1-8, characterized in that, The method includes: S1. Access telephone, online conversation and video channels in real time through preset interfaces, obtain multimodal raw conversation data streams, and output the multimodal raw conversation data streams after converting them into a standardized format; S2. Perform real-time processing on the standardized multimodal raw session data stream to generate structured NLP semantic analysis results containing time-series labels; S3. Based on preset quality inspection rules, perform real-time pattern recognition and risk assessment on the NLP semantic analysis results; and calculate service quality indicators and risk measurement indicators based on the risk assessment results. S4. Based on the aforementioned risk measurement indicators, trigger real-time risk alarms and monitor screen updates, and visualize the real-time service quality indicators and risk alarm information of the entire platform through the monitor screen.

Images