A hot-line traffic grading intervention method, system, electronic device and storage medium

Abstract

The application relates to the field of artificial intelligence and communication technology, and particularly discloses a hotline traffic hierarchical intervention method and system, an electronic device and a storage medium. The application collects bidirectional voice streams, carries out noise reduction, transcription, sentence segmentation and desensitization processing, extracts multi-dimensional call features in combination with traffic labels, carries out fusion analysis by using the analytic hierarchy process to distribute weights, simultaneously carries out space-time dimension correlation analysis based on time stamps to capture feature change trends and correlation, inputs the fusion analysis result, the change trend and the correlation into an intervention suggestion model, outputs suggestions containing intervention types, timing and priorities, and triggers intelligent monitoring, intelligent cipher or intelligent takeover actions accordingly, solves the problem that in the prior art, monitoring and cipher intervention completely depend on manual initiation, and automatic control of takeover only considers emotional features, leading to single triggering conditions, realizes real-time hierarchical intervention on hotline traffic, and improves the comprehensiveness, timeliness and scene adaptability of intervention measures.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and communication technology, and in particular to a method, system, electronic device and storage medium for tiered intervention of hotline calls. Background Technology

[0002] In government hotline call services, the experience and competence of operators vary greatly, easily leading to issues such as unfamiliarity with procedures, inappropriate wording, or emotional outbursts. Without timely intervention, this not only affects the efficiency of resolving user needs but also increases customer complaints and reduces hotline service satisfaction. Currently, call intervention methods mainly include monitoring, whispering, and takeover. However, existing technology only automates takeover scenarios; non-takeover intervention measures such as monitoring and whispering still rely entirely on manual initiation, resulting in low intervention efficiency and difficulty in meeting real-time requirements.

[0003] Furthermore, existing automated control in takeover scenarios only considers emotional characteristics, neglecting key features in actual call scenarios such as business type, response status, and dialogue. This makes the triggering conditions for automated intervention too simplistic, making it difficult to adapt to complex and ever-changing actual call environments. Consequently, the implementation of applications is greatly limited, and the potential of automated intervention in improving call service quality cannot be fully realized.

[0004] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method, system, electronic device, and storage medium for tiered intervention of hotline call volume.

[0006] Firstly, the present invention provides a method for hierarchical intervention of hotline call volume, the technical solution of which is as follows: Collect two-way voice streams between the agent and the caller, and associate them with call tags; perform noise reduction processing on the two-way voice streams, transcribe the noise-reduced voice data into text data, perform sentence segmentation processing on the text data to obtain text segments with timestamps, and desensitize sensitive information in each text segment. Extract multidimensional call features based on each text segment and the call tag; Based on the call scenario type and the urgency of the work order, the analytic hierarchy process (AHP) is used to assign weights to the multidimensional call features, and a fusion analysis is performed based on the multidimensional call features and their corresponding weights to obtain the fusion analysis results. Based on the timestamps corresponding to the multidimensional call features, spatiotemporal correlation analysis is performed on the multidimensional call features to obtain the feature change trend and the correlation between features; The fusion analysis results, the feature change trends, and the correlations between the features are input into the intervention suggestion model, and intervention suggestions are output. The intervention suggestions include intervention type, intervention timing, and intervention priority. The intervention type is intelligent listening, intelligent whispering, or intelligent takeover. According to the intervention timing and priority, the corresponding intervention action is triggered based on the intervention type: if the intervention type is intelligent monitoring, a monitoring reminder is pushed to the supervisor's terminal and an independent monitoring channel is established; if the intervention type is intelligent whispering, a handling suggestion is retrieved from the knowledge base, converted into voice, and pushed to the agent's terminal; if the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, and a call channel between the supervisor and the complainant is established.

[0007] The beneficial effects of the hotline call traffic tiered intervention method of the present invention are as follows: The method of this invention collects two-way voice streams and performs noise reduction, transcription, sentence segmentation, and desensitization processing. It extracts multi-dimensional call features by combining call tags, uses the analytic hierarchy process (AHP) to assign weights for fusion analysis, and performs spatiotemporal correlation analysis based on timestamps to capture feature change trends and correlations. The fusion analysis results, change trends, and correlations are input into an intervention suggestion model, which outputs suggestions including intervention type, timing, and priority. Based on these suggestions, it triggers intelligent monitoring, intelligent whispering, or intelligent takeover actions. This solves the problem that in the prior art, monitoring and whispering interventions rely entirely on manual initiation, and takeover automation control only considers emotional features, resulting in a single trigger condition. It realizes real-time hierarchical intervention for hotline calls, improving the comprehensiveness, timeliness, and scenario adaptability of intervention measures.

[0008] Based on the above scheme, the hotline call tiered intervention method of the present invention can be further improved as follows.

[0009] In one alternative approach, the call handling tags include basic tags and extended tags. The basic tags include work order number, agent ID, source of request, and acceptance time. The extended tags include agent real-time status, agent capability profile tags, work order urgency level, and the caller's historical request records. All call handling tags are associated through session IDs.

[0010] The advantages of adopting the above optional method are: further associating the basic tag with the extended tag through the session ID to form a complete call data link, and subsequent multi-dimensional call feature extraction can directly call the associated tag information, avoiding data silos and improving the accuracy and completeness of feature extraction.

[0011] In one alternative approach, the step of transcribing the denoised speech data into text data includes: An automatic speech recognition model is used to transcribe the noise-reduced speech data into initial text data. After transcription, the initial text data is corrected using a hotline terminology database and a script database. Typos are corrected and missing words are added to obtain the corrected text data.

[0012] The advantages of adopting the above-mentioned optional methods are as follows: after the automatic speech recognition and transcription is completed, a hotline terminology database and a dialogue database are introduced for error correction. The misspellings in professional terms and common dialogues are corrected and the missing words are filled in, which improves the matching degree between the text transcription results and the actual business scenario of the hotline, and provides a more accurate text foundation for subsequent feature extraction.

[0013] In one alternative approach, the multidimensional call features include service type features, response status features, speech features, caller emotional features, and agent capability features. The business type characteristics include the area of ​​the complaint, the business subcategory, the urgency of the complaint, the difficulty of resolving the complaint, and the relevance of historical complaints. The response characteristics include the completeness of the response, the accuracy of the policy basis, the timeliness of the response, and the satisfaction of the response result. The discourse characteristics include the standardization of the discourse, the use of polite language, the accuracy of professional terminology, and the rationality of the communication logic. The emotional characteristics of the complainant include the type of emotion, the intensity of emotion, the trigger point of emotion, and the form of emotion expression. The agent's ability characteristics include business proficiency, policy understanding, emotional coping ability, and communication adaptability.

[0014] The beneficial effects of adopting the above optional methods are as follows: a multi-dimensional call feature system is further constructed from five dimensions: business type, response status, dialogue, caller's emotions, and agent's ability. This system covers key influencing factors in call scenarios, making the basis for intervention judgment more comprehensive, reducing the one-sidedness caused by single-dimensional analysis, and improving the fit between intervention suggestions and actual situation.

[0015] In one alternative approach, the step of assigning weights to the multidimensional call features using the analytic hierarchy process includes: A judgment matrix is ​​constructed based on the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics. The maximum eigenvalue and the corresponding eigenvector of the judgment matrix are calculated. The eigenvector is normalized to obtain the weights of the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics. A consistency index is calculated based on the maximum eigenvalue of the judgment matrix and the order of the judgment matrix. A consistency ratio is calculated based on the consistency index and the average random consistency index. When the consistency ratio is less than a preset threshold, a consistency test is passed. The weights of each dimension feature are used as the weight allocation result of the multi-dimensional call features. Specifically, in the scenario of urgent public appeals, the weight of the appealer's emotional characteristics is increased; in the scenario of professional policy explanation, the weights of the explanation situation characteristics and the business type characteristics are increased.

[0016] The beneficial effects of adopting the above-mentioned optional methods are as follows: further use the analytic hierarchy process to assign weights to call features of different dimensions and pass the consistency test. At the same time, the weight ratio is dynamically adjusted according to different scenarios such as urgent public demands and professional policy explanations, so that the fusion analysis results are closer to the actual call scenario needs and the rationality and pertinence of intervention triggers are enhanced.

[0017] In one optional approach, the step of performing spatiotemporal correlation analysis on the multidimensional call features to obtain feature change trends and inter-feature correlations includes: Based on the timestamps corresponding to the business type characteristics, the response characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics, the temporal changes of each characteristic during the call are analyzed to obtain the characteristic change trend. Extract the correlation between the change of the emotional characteristics of the complainant and the change of the response characteristics, extract the matching degree correlation between the verbal characteristics and the business type characteristics, and take the correlation and the matching degree correlation as the correlation between the features.

[0018] The beneficial effects of adopting the above-mentioned optional methods are as follows: further time-series change analysis and correlation extraction of multi-dimensional call features based on timestamps, capturing the dynamic correlation between the caller's emotions and the response, and between the wording and the business type, and incorporating time-series changes and correlations into the fusion analysis, so that intervention judgment can reflect the real-time evolution of the call process.

[0019] In one alternative approach, the intervention suggestion further includes intervention support information and triggering core feature criteria; the step of triggering the corresponding intervention action according to the intervention timing and intervention priority and the intervention type further includes: If the intervention type is intelligent monitoring, then based on the triggering core characteristics, a monitoring reminder containing the seat ID, the request content and the triggering reason is pushed to the squad leader terminal, an independent monitoring channel is established and the real-time transcribed text is displayed; If the intervention type is intelligent whisper, then the treatment suggestion is retrieved from the knowledge base according to the intervention assistance information, converted into voice and pushed to the agent terminal, and the text content of the treatment suggestion is sent to the display screen of the agent terminal at the same time. If the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, a call channel between the team leader and the complainant is established, a reassuring voice message is sent to the complainant, and the reason for the takeover, the policy basis, and the follow-up requirements are sent to the original agent. The data from the entire intervention process will be collected into a big data platform. The data from the entire intervention process includes the voice data, the text data, the intervention actions, the treatment suggestions, and the agent feedback. The data from the entire intervention process is mined, and the recognition accuracy and emotion judgment precision of the automatic speech recognition model are optimized based on the mining results. The intervention threshold is also dynamically adjusted according to the characteristics of the demands in the hotline field.

[0020] The beneficial effects of adopting the above-mentioned optional methods are as follows: further supplementing the intervention recommendations with auxiliary information and triggering basis, so that various intervention actions have clear execution guidelines, while transmitting the data of the entire intervention process back to the big data platform for mining, continuously optimizing the accuracy of speech recognition and emotion judgment, dynamically adjusting the intervention threshold, and forming a self-improvement mechanism for intervention capabilities.

[0021] Secondly, this invention provides a hotline call tiered intervention system, the technical solution of which is as follows: The processing module is used to collect two-way voice streams between the agent and the caller, and associate them with call tags; to perform noise reduction processing on the two-way voice streams, to transcribe the noise-reduced voice data into text data, to perform sentence segmentation processing on the text data to obtain text segments with timestamps, and to desensitize sensitive information in each text segment. The extraction module is used to extract multidimensional call features based on each text segment and the call tag; The fusion module is used to assign weights to the multidimensional call features according to the call scenario type and the urgency of the work order, and to perform fusion analysis based on the multidimensional call features and their corresponding weights to obtain the fusion analysis results. The analysis module is used to perform spatiotemporal correlation analysis on the multidimensional call features based on the timestamps corresponding to the multidimensional call features, so as to obtain the feature change trend and the correlation between features; The output module is used to input the fusion analysis results, the feature change trends, and the correlation between the features into the intervention suggestion model and output intervention suggestions. The intervention suggestions include intervention type, intervention timing, and intervention priority. The intervention type is intelligent listening, intelligent whispering, or intelligent takeover. The triggering module is used to trigger corresponding intervention actions according to the intervention timing and priority, and based on the intervention type: if the intervention type is intelligent monitoring, a monitoring reminder is pushed to the supervisor's terminal and an independent monitoring channel is established; if the intervention type is intelligent whispering, a handling suggestion is retrieved from the knowledge base, converted into voice, and pushed to the agent's terminal; if the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, and a call channel between the supervisor and the complainant is established.

[0022] The beneficial effects of the hotline call tiered intervention system of the present invention are as follows: The system of this invention collects two-way voice streams and performs noise reduction, transcription, sentence segmentation, and desensitization processing. It extracts multi-dimensional call features by combining caller ID tags, uses the analytic hierarchy process (AHP) to assign weights for fusion analysis, and performs spatiotemporal correlation analysis based on timestamps to capture feature change trends and relationships. The fusion analysis results, change trends, and relationships are input into an intervention suggestion model, which outputs suggestions including intervention type, timing, and priority. Based on these suggestions, it triggers intelligent monitoring, intelligent whispering, or intelligent takeover actions. This solves the problem that existing technologies rely entirely on manual initiation for monitoring and whispering interventions, and that automated takeover control only considers emotional features, resulting in single triggering conditions. It achieves real-time hierarchical intervention for hotline calls, improving the comprehensiveness, timeliness, and scenario adaptability of intervention measures.

[0023] Thirdly, the technical solution of an electronic device according to the present invention is as follows: It includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the hotline traffic tiered intervention method of the present invention.

[0024] Fourthly, the technical solution of a computer-readable storage medium provided by the present invention is as follows: The computer-readable storage medium stores instructions that, when read, cause the computer-readable storage medium to perform the steps of the hotline traffic classification intervention method of the present invention.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0026] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1This is a flowchart illustrating an embodiment of a hotline call traffic tiered intervention method according to the present invention. Figure 2 This is a schematic diagram of the overall process; Figure 3 A flowchart illustrating the process of scenario-based artificial intelligence intelligent analysis; Figure 4 A schematic diagram illustrating the process of triggering and executing hierarchical intelligent intervention; Figure 5 This is a schematic diagram of an embodiment of a hotline call tiered intervention system according to the present invention; Figure 6 This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0027] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0028] Figure 1 This diagram illustrates a flowchart of an embodiment of a hotline traffic tiered intervention method provided by the present invention. This method can be executed by electronic devices such as terminal devices or servers. The terminal device can be any fixed or mobile terminal, such as user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, or wearable device. The server can be a single server or a server cluster consisting of multiple servers. Any electronic device can implement the hotline traffic tiered intervention method by having its processor call computer-readable instructions stored in its memory. Figure 1 As shown, it includes the following steps: S1. Collect the two-way voice stream between the agent and the caller, and associate it with call tags; perform noise reduction processing on the two-way voice stream, transcribe the noise-reduced voice data into text data, perform sentence segmentation processing on the text data to obtain text segments with timestamps, and perform desensitization processing on sensitive information in each text segment.

[0029] Two-way voice stream refers to the real-time audio data stream generated between the agent and the caller during a call, containing voice signals from two independent channels. For example, agent A (employee ID S001) and caller B (phone number 138****) begin a call at 10:15 AM on June 1, 2026, simultaneously collecting the voice recordings of both agent A and caller B to form two independent audio streams. Call traffic tags refer to structured metadata associated with each call session, used to describe the basic attributes, participant status, and business context of that session. For example, caller B's work order number is T20260601001, agent ID is S001, the source of the call is a telephone channel, the processing time is 10:15 AM on June 1, 2026, agent A's real-time status is busy, agent A's capability profile tag is moderate policy understanding, the work order urgency is moderate, and caller B's historical call records show that they called to inquire about a similar matter three months ago. All of this information serves as the call traffic tag for this session.

[0030] The denoised speech data refers to the clean speech signal obtained after the environmental noise, echo, and current noise in the original two-way speech stream are suppressed by an algorithm. For example, if the complainant B is using a mobile phone outdoors, the original speech stream contains vehicle noise and wind noise. The environmental noise is removed by using time-domain denoising algorithms and frequency-domain denoising algorithms to obtain the denoised speech data, which retains only the speech content of the agent A and the complainant B.

[0031] The text data refers to the text content transcribed from the denoised speech data using an automatic speech recognition model, including the text on the agent's side and the text on the complainant's side. For example, agent A says, "Hello, how can I help you?" and complainant B replies, "I would like to inquire about the policy processing procedure." These two speech segments are transcribed into two text data segments: "Hello, how can I help you?" and "I would like to inquire about the policy processing procedure." A text segment refers to an independent sentence unit with a timestamp obtained after automatically segmenting the transcribed text data according to semantic units. For example, agent A's continuous speech transcribed text "Hello, how can I help you?" is segmented into two text segments: "Hello" and "How can I help you?", each timestamped with 10:15:03 and 10:15:05 on June 1, 2026, respectively.

[0032] S2. Extract multidimensional call features based on each text segment and the call tag.

[0033] Among them, multidimensional call features refer to a set of quantitative or categorical indicators extracted from text segments and call tags to describe the multidimensional attributes of a call; for example, five dimensions of features were extracted from the caller B's call: business type features (the call area is people's livelihood, the business sub-category is policy consultation, and the urgency of the call is moderate), response features (the completeness of the response is complete), speech features (the speech standardization is compliant), caller's emotional features (emotion type is calm, and emotion intensity is low), and agent's ability features (business proficiency is moderate).

[0034] S3. Based on the call scenario type and the urgency of the work order, the multidimensional call features are assigned weights using the analytic hierarchy process (AHP), and a fusion analysis is performed based on the multidimensional call features and their corresponding weights to obtain the fusion analysis results.

[0035] The call scenario type refers to the classification of the service context to which the current call belongs based on the content of the request, the business area, and the interaction mode. For example, if caller B is inquiring about the policy processing procedure, the call scenario type is classified as a professional policy explanation scenario. The urgency level of the work order refers to the classification of the urgency of processing the current work order based on the content of the request, historical request records, and business rules. For example, if caller B's description of the situation does not fall under the category of life and property safety or basic living security requests, the urgency level of the work order is classified as general.

[0036] The fusion analysis result refers to the quantitative evaluation vector obtained by weighting the multi-dimensional call features with the weights assigned by the analytic hierarchy process. This vector reflects the overall performance of the current call's various dimensions. For example, the feature values ​​of the five dimensions of caller B's call—service type, response status, dialogue style, caller's emotional state, and agent's ability—are multiplied by their corresponding weights and then summed to obtain a fusion analysis result of 0.85. The response status feature contributes the most, indicating that the current response quality is good.

[0037] S4. Based on the timestamps corresponding to the multidimensional call features, perform spatiotemporal correlation analysis on the multidimensional call features to obtain the feature change trend and the correlation between features.

[0038] The timestamp refers to the time information that identifies the moment a text segment appears during a call, used to achieve a precise correspondence between the text content and the call sequence. For example, the timestamp corresponding to the text segment "I would like to inquire about the policy processing procedure" is 10:15:08 on June 1, 2026, indicating that the time when the complainant B said this sentence was 8 seconds after the call started.

[0039] Among them, the characteristic change trend refers to the evolution direction and fluctuation pattern of each dimension of call characteristics over time during the call. For example, the emotional characteristics of the complainant B are calm in the first 2 minutes of the call, turn slightly anxious after 3 minutes, and the emotional intensity shows a slow upward trend. At the same time, the response situation characteristic decreases slightly after 3 minutes because the agent A hesitates a little when citing policy clauses. The two show an inverse change direction. The correlation between characteristics refers to the mutual influence, synergistic change, or dependency between different dimensions of call characteristics during the call. For example, the emotional intensity of the complainant B changes with the fluctuation of the accuracy of the agent A's response, and there is a correlation between the two. There is a matching degree correlation between the accuracy of professional terminology in the agent A's speech characteristics and the difficulty of resolving the complaint in the business type characteristics.

[0040] S5. Input the fusion analysis results, the feature change trends, and the correlation between the features into the intervention suggestion model, and output the intervention suggestion. The intervention suggestion includes the intervention type, intervention timing, and intervention priority. The intervention type is intelligent listening, intelligent whispering, or intelligent takeover.

[0041] The intervention suggestion model refers to a pre-trained machine learning model that can output intervention decision suggestions based on the results of fusion analysis, feature change trends, and correlations between features. For example, the fusion analysis result score of 0.85, the slowly increasing trend of the complainant B's emotional intensity, and the correlation between emotional intensity and the accuracy of the response are all input into the intervention suggestion model. The model outputs a suggestion for intelligent whispered intervention. The intervention suggestion refers to the decision information output by the intervention suggestion model, including the intervention type, timing, and priority. For example, the intervention suggestion model outputs the following intervention suggestion for the caller B: intervention type is intelligent whispered intervention, intervention timing is 3 minutes and 20 seconds into the call, and intervention priority is medium.

[0042] The intervention type refers to the category of intervention methods suggested for the current call situation, including intelligent monitoring, intelligent whispering, and intelligent takeover. For example, the intervention suggestion model outputs intelligent whispering as the intervention type, indicating that a policy interpretation prompt should be sent to agent A. The intervention timing refers to the suggested time point or stage of the call to trigger the intervention action. For example, the intervention suggestion model outputs intervention timing as within 1 second after complainant B finishes saying "What specific materials are needed?", i.e., when the call reaches 3 minutes and 20 seconds. The intervention priority refers to the processing order of the current intervention suggestion among multiple concurrent pending intervention requests. For example, if the model is simultaneously processing a whispering intervention request from complainant B and a takeover intervention request from another complaint call, it determines the intervention priority of the other complaint call is urgent, and the intervention priority of complainant B's call is medium, and processes them in order of urgency.

[0043] Intelligent monitoring refers to: automatically pushing monitoring reminders to the supervisor's terminal and establishing an independent monitoring channel after supervisor confirmation, allowing the supervisor to listen to the call content and view real-time transcribed text and analysis tags without being aware of the agents or the complainant. For example, if a potential deviation in agent A's statement of policy terms is detected, a monitoring reminder is automatically pushed to the supervisor's terminal. After supervisor C confirms, an independent monitoring channel is established. Neither agent A nor complainant B is aware of being monitored. Supervisor C's terminal simultaneously displays the real-time transcribed text and the analysis tag "Policy statement suspected of being incomplete." Intelligent whispering refers to: retrieving corresponding policy interpretations or handling suggestions from the knowledge base, converting them into concise speech through speech synthesis, and pushing them to the agent's terminal through an independent audio channel. At the same time, detailed text content is sent to the agent's terminal display screen. The complainant cannot hear or see this information. For example, retrieving a list of materials required for policy processing from the knowledge base, converting "informing the complainant that they need to provide identification and an application form" into a 3-second speech, and pushing it to agent A through an independent audio channel. At the same time, the detailed list of materials is sent to agent A's display screen. The complainant B cannot hear this prompt at all. Intelligent takeover refers to: cutting off the original call channel between the caller and the complainant, establishing a call channel between the caller and the supervisor or expert, and sending a standardized reassuring message to the complainant, while sending the original caller the reason for the takeover, the policy basis, and follow-up requirements; for example, cutting off the call channel between caller A and complainant B, establishing a call channel between expert D and complainant B, playing a reassuring message to complainant B, "Hello, to better solve your problem, a dedicated expert is now at your service," while simultaneously sending caller A the reason for the takeover as "incomplete policy interpretation leading to complainant's doubts," the policy basis as "relevant policy documents," and the follow-up requirement as "attend policy training."

[0044] S6. According to the intervention timing and the intervention priority, trigger the corresponding intervention action according to the intervention type: if the intervention type is intelligent monitoring, push the monitoring reminder to the supervisor's terminal and establish an independent monitoring channel; if the intervention type is intelligent whispering, retrieve the handling suggestion from the knowledge base, convert it into voice and push it to the agent's terminal; if the intervention type is intelligent takeover, cut off the call channel between the agent and the complainant, and establish a call channel between the supervisor and the complainant.

[0045] Intervention actions refer to the specific sequence of operations performed according to the type of intervention. For example, performing intelligent whisper intervention actions specifically includes retrieving the policy processing material list text from the knowledge base, converting the text into speech, pushing it to agent A's terminal through an independent audio channel, and sending detailed text content to agent A's display screen.

[0046] The technical solution of this embodiment acquires two-way voice streams and performs noise reduction, transcription, sentence segmentation, and desensitization processing. It extracts multi-dimensional call features by combining call tags, uses the analytic hierarchy process (AHP) to assign weights for fusion analysis, and performs spatiotemporal correlation analysis based on timestamps to capture feature change trends and correlations. The fusion analysis results, change trends, and correlations are input into the intervention suggestion model, which outputs suggestions including intervention type, timing, and priority. Based on these suggestions, it triggers intelligent monitoring, intelligent whispering, or intelligent takeover actions. This solves the problem that in the prior art, monitoring and whispering interventions rely entirely on manual initiation, and the automated takeover control only considers emotional features, resulting in a single trigger condition. It realizes real-time hierarchical intervention for hotline calls, improving the comprehensiveness, timeliness, and scenario adaptability of intervention measures.

[0047] In one alternative approach, the call handling tags include basic tags and extended tags. The basic tags include work order number, agent ID, source of request, and acceptance time. The extended tags include agent real-time status, agent capability profile tags, work order urgency level, and the caller's historical request records. All call handling tags are associated through session IDs.

[0048] The basic tags refer to the basic metadata information directly associated with the call session. For example, the basic tags for caller B's call include work order number T20260601001, agent ID S001, the source of the request being a telephone call, and the acceptance time being 10:15 AM on June 1, 2026. The extended tags refer to supplementary information obtained through real-time collection or historical analysis in addition to the basic tags. For example, the extended tags for caller B's call include agent A's real-time status being busy, agent A's capability profile tag being moderate policy knowledge, the work order urgency being moderate, and caller B's historical request records showing that they had consulted on a similar matter three months prior.

[0049] The work order number refers to a unique identifier automatically generated for each incoming call; for example, after caller B calls, work order number T20260601001 is automatically assigned, and this number is used throughout the entire process of work order acceptance, processing, and follow-up. The agent ID refers to a unique identifier assigned to each agent; for example, agent A's agent ID is S001, which is used to link agent A's capability profile, historical performance, and real-time status data. The source of the request refers to the type of channel through which the caller initiated the call; for example, if caller B initiates the request by dialing a hotline, the source of the request is marked as telephone; if caller B initiates the inquiry through online customer service, the source of the request is marked as online. The acceptance time refers to the recorded start time of the call session; for example, the acceptance time of caller B's call is recorded as 10:15:03 on June 1, 2026.

[0050] Among them, the real-time status of the agent refers to the agent's current working status; for example, if agent A's real-time status is recorded as "busy" when caller B calls, it means agent A is on another call. Agent competency profile tags refer to assessment tags generated based on historical agent performance data, reflecting the agent's competency levels in various dimensions; for example, based on agent A's historical call records and performance evaluation data, agent competency profile tags include "medium" for business proficiency, "medium" for policy understanding, "good" for emotional handling ability, and "good" for communication adaptability. The caller's historical complaint records refer to the caller's past call records and processing results; for example, if a search reveals that caller B called three months ago to inquire about a similar matter, the work order number was T20260301056, and the processing result was "completed."

[0051] In the above-mentioned optional methods, the basic tags and extended tags are further associated through the session ID to form a complete call data link. Subsequent multi-dimensional call feature extraction can directly call the associated tag information, avoiding data silos and improving the accuracy and completeness of feature extraction.

[0052] In one alternative approach, the step of transcribing the denoised speech data into text data includes: An automatic speech recognition model is used to transcribe the noise-reduced speech data into initial text data.

[0053] The automatic speech recognition model refers to a deep learning model that converts speech signals into text. For example, an automatic speech recognition model pre-trained using historical hotline recordings transcribes Agent A's statement "You need to provide identification and an application form" into the initial text data "You need to provide identification and an application form." The initial text data refers to the raw text content directly output by the automatic speech recognition model without subsequent error correction. For example, the automatic speech recognition model outputs the initial text data "You need to provide identification and an application form" from Agent A's speech, where "identity verification" might be mistakenly transcribed as "identity verification" due to unclear pronunciation.

[0054] After transcription, the initial text data is corrected using a hotline terminology database and a script database. Typos are corrected and missing words are added to obtain the corrected text data.

[0055] The hotline terminology database and the script database refer to database collections that store professional terms, standard script templates, and their correct textual forms in the field of hotline services. For example, the hotline terminology database stores standard terms such as "identity certificate" and "application form," and the script database stores standard script templates such as "Hello, how can I help you?" and "Please provide some relevant materials." The hotline terminology database and the script database are used to correct "identity certificate" in the initial text data.

[0056] In the above-mentioned optional methods, after the automatic speech recognition and transcription is completed, a hotline terminology database and a dialogue database are introduced for error correction. This corrects typos in professional terms and common dialogues and fills in missing words, thereby improving the matching degree between the text transcription results and the actual business scenarios of the hotline and providing a more accurate text foundation for subsequent feature extraction.

[0057] In one alternative approach, the multidimensional call features include service type features, response status features, speech patterns features, caller emotional features, and agent capability features.

[0058] Among them, business type characteristics refer to a set of dimensional indicators that describe the business field, category, urgency, difficulty of resolution, and historical relevance of the request. For example, the business type characteristics of the caller B include: the request field is people's livelihood, the business subcategory is policy consultation, the urgency of the request is average, the difficulty of resolving the request is medium, and the historical relevance is that there was a similar request three months ago.

[0059] Among them, the response characteristics refer to a set of dimensional indicators describing the completeness, accuracy, timeliness, and satisfaction of the agent's response to the complainant's questions. For example, the response characteristics of agent A are as follows: the completeness of the response is complete (the required materials and processing procedures were explained), the accuracy of the policy basis is accurate (the policy documents were correctly cited), the timeliness of the response is timely (responded within 3 seconds), and the satisfaction of the response result is initially determined to be satisfactory (the complainant expressed understanding).

[0060] Among them, the discourse characteristics refer to: a set of dimensional indicators describing the standardization, politeness, professionalism and logic of the agent's language; for example, the discourse characteristics extracted from agent A include: the standardization of the language is compliant (answering according to the standard process), the use of polite language is good (using polite language such as "you" and "please"), the accuracy of professional terminology is good (the policy terminology is clearly expressed), and the rationality of communication logic is good (the explanation process is clear and organized).

[0061] Among them, the emotional characteristics of the complainant refer to: a set of dimensional indicators describing the emotional state of the complainant during the call in terms of type, intensity, trigger point and expression form; for example, the emotional characteristics of complainant B are as follows: the emotional type is calm, the emotional intensity is low, the emotional trigger point is when agent A says "You need to provide relevant materials", and the emotional expression form is a calm tone and normal questioning.

[0062] Among them, the agent capability characteristics refer to a set of dimensional indicators describing the agent's business proficiency, policy understanding, emotional response, and communication adaptability. For example, the agent capability characteristics extracted for agent A include: business proficiency is medium, policy understanding is medium, emotional response is good, and communication adaptability is good.

[0063] Specifically, the characteristics of the business type include the area of ​​the request, the sub-category of the business, the urgency of the request, the difficulty of resolving the request, and the relevance of the historical requests; the characteristics of the response include the completeness of the response, the accuracy of the policy basis, the timeliness of the response, and the satisfaction of the response result; the characteristics of the communication style include the standardization of the communication style, the use of polite language, the accuracy of professional terminology, and the rationality of the communication logic; the characteristics of the complainant's emotions include the type of emotion, the intensity of emotion, the trigger point of emotion, and the form of emotion expression; and the characteristics of the agent's ability include business proficiency, policy understanding, emotional coping ability, and communication adaptability.

[0064] The terms and conditions are as follows: **Claim Area:** This refers to the relevant business department or industry category to which the claim pertains. For example, if claimant B inquires about policy processing procedures, the claim area is classified as "People's Livelihood." **Business Subcategory:** This refers to a further subdivision of the business type within the claim area. For example, within the "People's Livelihood" category, claimant B's call is classified as "Policy Consultation." **Claim Urgency:** This refers to the level of impact of the claim on the claimant's life, property safety, and basic living security. For example, if claimant B's inquiry about policy processing procedures does not involve life, property safety, or basic living security, the urgency level is classified as "moderate." **Claim Resolution Difficulty:** This refers to the estimated workload required to resolve the claim based on its complexity, the clarity of the policy basis, and past experience. For example, if the policy processing procedure involves multiple documents and steps, indicating complexity, the resolution difficulty is classified as "medium." **Historical Relevance:** This refers to the similarity between the current claim and the claimant's past claims in terms of business area and content. For example, if claimant B inquired about a similar matter three months ago, which falls under the same "People's Livelihood" category as this policy consultation, the historical relevance is classified as "medium."

[0065] The criteria for "completeness of answer" include: whether the agent provides a comprehensive and thorough answer to the caller's questions; for example, agent A informing caller B, "You need to provide identification and an application form, and go to the service hall to process it," covering the required materials and processing location, would be considered complete. The accuracy of policy basis refers to whether the policy documents and legal provisions cited by the agent during the answer process are accurate; for example, agent A's accurate citation of relevant policy documents would be considered accurate. The timeliness of problem response refers to whether the time interval between each caller's question and response meets service standards; for example, agent A responds within 3 seconds after each of caller B's questions, meeting the timeliness requirement, and the problem response is considered timely. The satisfaction with the answer result refers to the initial assessment of the caller's satisfaction with the answer result based on subsequent verbal feedback; for example, caller B saying "Okay, I understand" after hearing agent A's answer would be initially considered satisfactory.

[0066] The criteria for evaluating call script standardization are as follows: **Script Standardization:** Whether the language used by the agent conforms to the requirements of the hotline service standard script template. For example, Agent A follows the standard procedure, thus qualifying as compliant. **Politeness Usage:** This refers to the agent's use of honorifics and polite expressions during the call. For example, Agent A uses polite expressions such as "you," "please," and "thank you for your call," thus qualifying as good. **Terminology Accuracy:** This refers to the accuracy of the agent's use of professional terminology during the answer process. For example, Agent A uses terms such as "identity certificate," "application form," and "service hall" clearly, thus qualifying as good. **Communication Logic Rationality:** This refers to the organization and logical coherence of the content during the answer process. For example, Agent A explains the policy processing procedure in the order of "material preparation—submission location—processing time limit," demonstrating clear logic, thus qualifying as good.

[0067] Here, "emotion type" refers to the category of emotions displayed by the caller during the call; for example, caller B's current emotion type is identified as calm. "Emotion intensity" refers to the level of intensity of the caller's emotional expression; for example, caller B's calm emotion intensity is determined to be low, manifested as a steady tone and normal speaking speed. "Emotional trigger point" refers to the specific event or content of the agent's speech that causes a change in the caller's emotion; for example, caller B's emotional trigger point is agent A saying, "You need to provide relevant materials," after which caller B's emotion remains calm. "Emotional expression form" refers to the external manifestation of the caller's emotion, including changes in tone, speaking speed, and questioning style; for example, caller B's calm emotion is expressed through a steady tone, normal speaking speed, and normal questioning style.

[0068] Among these, "Business Proficiency" refers to the agent's mastery of the knowledge, processes, and operational procedures within their respective business area; for example, based on Agent A's historical call records and performance data, their business proficiency is rated as moderate. "Policy Mastery" refers to the agent's understanding and mastery of relevant policies, regulations, and documents; for example, based on Agent A's accuracy and completeness in citing policy processing procedures, their policy mastery is rated as moderate. "Emotional Coping Ability" refers to the agent's ability to identify, understand, and soothe negative emotions of the caller; for example, based on Agent A's ability to provide a normal response when the caller B remains calm, their emotional coping ability is rated as good. "Communication Adaptability" refers to the degree of match between the agent's language expression and the caller's comprehension ability; for example, Agent A uses standard terminology such as "identity verification" and "application form," and the caller B expresses understanding, thus their communication adaptability is rated as good.

[0069] Among the above-mentioned optional methods, a multi-dimensional call feature system is further constructed from five dimensions: business type, response status, dialogue, caller's emotions, and agent's capabilities. This system covers key influencing factors in call scenarios, making the basis for intervention judgment more comprehensive, reducing the one-sidedness caused by single-dimensional analysis, and improving the fit between intervention suggestions and actual situations.

[0070] In one alternative approach, the step of assigning weights to the multidimensional call features using the analytic hierarchy process includes: A judgment matrix is ​​constructed based on the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics. The maximum eigenvalue and the corresponding eigenvector of the judgment matrix are calculated. The eigenvector is then normalized to obtain the weights of the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics.

[0071] The judgment matrix refers to a square matrix used in the analytic hierarchy process (AHP) to compare the importance of multiple dimensional features pairwise. The matrix elements represent the importance scale of the row dimension relative to the column dimension. For example, a judgment matrix is ​​constructed for the scenario of a caller B, where the importance scale of the business type feature relative to the caller's emotional feature is 3, indicating that in this scenario of professional policy interpretation, the business type feature is slightly more important than the caller's emotional feature. After pairwise comparison of all dimensions, a complete 5th-order judgment matrix is ​​formed.

[0072] A consistency index is calculated based on the maximum eigenvalue of the judgment matrix and the order of the judgment matrix. A consistency ratio is calculated based on the consistency index and the average random consistency index. When the consistency ratio is less than a preset threshold, a consistency test is passed. The weights of each dimension feature are used as the weight allocation result of the multi-dimensional call features. Specifically, in the scenario of urgent public appeals, the weight of the appealer's emotional characteristics is increased; in the scenario of professional policy explanation, the weights of the explanation situation characteristics and the business type characteristics are increased.

[0073] The consistency index refers to a quantitative indicator used to measure whether there is logical consistency between the pairwise comparison results of each dimension in the judgment matrix. For example, the consistency index is calculated as 0.035 based on the largest eigenvalue and the matrix order of the judgment matrix, and is used to calculate the consistency ratio. The consistency ratio is the ratio of the consistency index to the average random consistency index, used to determine whether the matrix passes the consistency test. For example, a consistency ratio of 0.031 is calculated, which is less than the preset threshold of 0.1, indicating that the pairwise comparisons of each dimension of the judgment matrix have satisfactory consistency. The preset threshold is a pre-set standard value used to determine whether the consistency ratio is qualified. For example, the preset threshold is set to 0.1, and the consistency ratio calculated from the judgment matrix is ​​0.031, which is less than 0.1, thus passing the consistency test.

[0074] Among the above-mentioned optional methods, the Analytic Hierarchy Process (AHP) is further used to assign weights to call features of different dimensions and pass consistency checks. At the same time, the weight ratio is dynamically adjusted according to different scenarios such as urgent public demands and professional policy explanations, so that the fusion analysis results are closer to the actual call scenario needs and the rationality and pertinence of intervention triggers are enhanced.

[0075] In one optional approach, the step of performing spatiotemporal correlation analysis on the multidimensional call features to obtain feature change trends and inter-feature correlations includes: Based on the timestamps corresponding to the business type characteristics, the response characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics, the temporal changes of each characteristic during the call are analyzed to obtain the characteristic change trend.

[0076] Among them, temporal change refers to the change in the value or state of the feature value of each dimension of the call as the call progresses over time; for example, the emotional characteristics of the complainant B remained calm from the beginning of the call to the end of the call, and the emotional intensity value was always 0.2, without showing obvious temporal change.

[0077] Extract the correlation between the change of the emotional characteristics of the complainant and the change of the response characteristics, extract the matching degree correlation between the verbal characteristics and the business type characteristics, and take the correlation and the matching degree correlation as the correlation between the features.

[0078] Among them, the correlation refers to the dependency relationship between two different dimensions of call features, which influence each other or change synergistically. For example, there is no obvious dependency between the emotional intensity of caller B and the completeness of agent A's answer, and there is no significant correlation between the two. The matching degree correlation refers to the degree of fit between the speech features and the business type features in terms of semantic content and logical rationality. For example, there is a matching degree correlation between the accuracy of professional terminology in agent A's speech features and the difficulty of resolving the caller B's request in the business type features. When the business type is professional policy explanation, the accuracy of professional terminology is highly suitable for this scenario.

[0079] Among the above-mentioned optional methods, further time-series change analysis and correlation extraction are performed on multi-dimensional call features based on timestamps to capture the dynamic correlation between the caller's emotions and the response, and between the wording and the business type. The time-series changes and correlations are incorporated into the fusion analysis so that the intervention judgment can reflect the real-time evolution of the call process.

[0080] In one alternative approach, the intervention suggestion further includes intervention support information and triggering core feature criteria; the step of triggering the corresponding intervention action according to the intervention timing and intervention priority and the intervention type further includes: If the intervention type is intelligent monitoring, then based on the triggering core characteristics, a monitoring reminder containing the seat ID, the request content, and the triggering reason is pushed to the squad leader terminal, an independent monitoring channel is established, and real-time transcribed text is displayed.

[0081] Real-time transcription refers to the process of instantly transcribing speech into text and displaying it on the relevant terminal during a call. For example, the conversation between agent A and complainant B is transcribed into text in real time and displayed on the supervisor C's terminal. The transcribed content is presented sentence by sentence in the form of a conversation stream, with a delay of no more than 1 second.

[0082] If the intervention type is intelligent whisper, then the treatment suggestion is retrieved from the knowledge base according to the intervention assistance information, converted into voice and pushed to the agent terminal, and the text content of the treatment suggestion is sent to the display screen of the agent terminal.

[0083] Among them, the agent terminal refers to an electronic device used by agents to answer calls, view information and receive prompts; for example, the agent terminal used by agent A is a computer with a headset and a display screen, used to answer the call from the complainant B and receive the secret message information.

[0084] If the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, a call channel between the team leader and the complainant is established, a reassuring voice message is sent to the complainant, and the original agent is sent the reason for the takeover, the policy basis, and the follow-up requirements.

[0085] Among them, the reassuring voice refers to the standardized voice prompt content played to the complainant in the intelligent takeover scenario; for example, the reassuring voice "Hello, in order to better solve your problem, a dedicated expert is now at your service" is played to complainant B, indicating that the call to complainant B has been transferred to the expert's seat.

[0086] The "reason for takeover" refers to the specific reasons for triggering the intelligent takeover operation. For example, the reason pushed to Agent A is "incomplete policy interpretation leading to the complainant's doubts," explaining the specific circumstances that triggered the takeover. The "policy basis" refers to the corresponding policy documents or business specifications that provide rule support for the takeover operation. For example, the policy basis pushed to Agent A is "relevant policy documents," explaining the rule basis for the takeover. The "follow-up requirements" refers to the arrangements for subsequent handling tasks proposed by the original agent after the intelligent takeover. For example, the follow-up requirement pushed to Agent A is "attend policy training," requiring Agent A to complete specialized policy training.

[0087] The data from the entire intervention process will be collected into a big data platform. This data includes the voice data, text data, intervention actions, treatment suggestions, and agent feedback.

[0088] Among them, the big data platform refers to a back-end system used to collect and store data from the entire intervention process and provide data mining and analysis capabilities; for example, the voice data, transcribed text, intelligent whisper intervention actions, handling suggestions retrieved from the knowledge base, and agent feedback confirmed by agent A for this call of complainant B are all collected and stored on the big data platform.

[0089] Among them, agent feedback refers to the confirmation information or processing result sent by the agent after receiving or performing an intervention action; for example, after agent A supplements the specific list of materials for policy processing to the complainant B according to the secret prompt, he clicks the "processed" button on the agent terminal, and the confirmation information is sent to the system as agent feedback.

[0090] The data from the entire intervention process is mined, and the recognition accuracy and emotion judgment precision of the automatic speech recognition model are optimized based on the mining results. The intervention threshold is also dynamically adjusted according to the characteristics of the demands in the hotline field.

[0091] The mining results refer to valuable information that can be used to optimize the model after statistical analysis of the intervention process data collected in the big data platform. For example, mining the intervention process data of the caller B's current call and similar policy consultation calls revealed that the accuracy rate of the policy basis cited by the callers in policy consultation calls was at a good level. This information was used as the mining results for subsequent optimization.

[0092] Among them, recognition accuracy refers to the proportion of speech correctly transcribed into text by the automatic speech recognition model, reflecting the accuracy of the model's transcription. For example, using the optimized automatic speech recognition model to transcribe policy consultation calls, the recognition accuracy increased from 90% to 95%, and the mistranscription rate of professional terms such as "identity verification" significantly decreased. Emotion judgment accuracy refers to the accuracy with which the emotion recognition model determines the type and intensity of the complainant's emotion. For example, after optimizing the emotion judgment accuracy, the recognition accuracy for complainant B's "calm" emotion type increased from 88% to 94%.

[0093] The "characteristics of the request" refers to the differences in the types of services, commonly used phrases, and common emotional expressions exhibited by requests in different hotline areas. For example, policy consultation requests are characterized by a high frequency of professional terminology and frequent policy citations, while complaint requests are characterized by strong emotional expressions and a high frequency of repeated questions. The "intervention threshold" refers to the critical value of the comprehensive score of multi-dimensional call characteristics required to trigger various intervention actions. For example, based on the characteristics of policy consultation requests, the intelligent whisper intervention threshold is adjusted from 0.65 to 0.60, making it easier to trigger whisper prompts when there are deviations in the answers to policy consultation requests.

[0094] Among the above-mentioned optional methods, supplementary information and triggering criteria are further added to the intervention recommendations to provide clear execution guidelines for various intervention actions. At the same time, the data of the entire intervention process is transmitted back to the big data platform for mining, continuously optimizing the accuracy of speech recognition and emotion judgment, dynamically adjusting the intervention threshold, and forming a self-improvement mechanism for intervention capabilities.

[0095] Figure 2 A general flowchart of this embodiment is shown. Figure 2As shown, the real-time voice data acquisition and preprocessing stage is responsible for acquiring the two-way voice stream between the agent and the complainant and completing noise reduction, format conversion and quality detection. The scenario-based artificial intelligence intelligent analysis stage is responsible for converting the voice into structured features and performing fusion analysis. The hierarchical intelligent intervention triggering stage is responsible for executing listening, whispering or takeover actions according to the intervention suggestions. The data accumulation and model optimization stage is responsible for collecting the data of the entire intervention process and using it for model updates.

[0096] In the real-time voice data acquisition and preprocessing stage, two-way voice streams between agents and complainants are acquired in real time and synchronously associated with call tags. Call tags include basic and extended tags. Basic tags include work order number, agent identification, source of complaint, and processing time. Extended tags include agent real-time status, agent capability profile tags, work order urgency, and complainant's historical complaint records. All call tags are associated through session identification. The acquired two-way voice streams undergo noise reduction processing, employing time-domain and frequency-domain noise reduction algorithms to remove environmental noise, echoes, and current noise. The noise-reduced voice undergoes voice quality detection, determining voice validity through voice signal-to-noise ratio and voice activity detection. Low-quality voice is marked in real time, triggering an automatic speech recognition transcription enhancement mode. The voice format is converted from the default pulse code modulation format to MP3 and OPUS formats. MP3 format is used for secure storage, and OPUS format is used for real-time transmission. The preprocessed voice data is synchronously transmitted to the automatic speech recognition model and secure storage unit.

[0097] Figure 3 A detailed flowchart of the contextualized AI intelligent analysis phase is shown. For example... Figure 3 As shown, this stage includes speech transcription and text optimization, text structuring processing, real-time desensitization of sensitive information, multi-dimensional feature extraction, dynamic allocation and fusion analysis of feature weights, and inference and output of intervention suggestion models.

[0098] In the speech-to-text and text optimization process, an automatic speech recognition model is used to transcribe denoised speech data into initial text data in real time. The automatic speech recognition model is specifically trained for different hotline types, including government hotlines and directory assistance hotlines, based on a general model. This specialized training uses labeled historical hotline data to train the general model. After transcription, the initial text data is corrected using a hotline terminology database and a dialogue script database, correcting typos and adding missing words to obtain the final text data.

[0099] In text structuring, the transcribed text data is automatically segmented based on a natural language processing model. After segmentation, text segments with timestamps are obtained, indicating the time when the text segment appeared during the call. The correspondence between text segments and timestamps is shown in Table 1.

[0100] Table 1: In the real-time desensitization of sensitive information, sensitive information in each text segment is masked and desensitized in real time. Sensitive information includes name, ID number, address and contact information.

[0101] In multidimensional feature extraction, based on the structured text data and combined with the caller ID associated with the session identity, five core features are extracted. Each feature is bound to a timestamp to achieve accurate spatiotemporal tracing. The five features are shown in Table 2.

[0102] Table 2: Business type characteristics include the area of ​​the complaint, the subcategory of the business, the urgency of the complaint, the difficulty of resolving the complaint, and the relevance to previous complaints. Response characteristics include the completeness of the response, the accuracy of the policy basis, the timeliness of the response, and the satisfaction with the resolution result. Communication style characteristics include the standardization of the communication style, the use of polite language, the accuracy of professional terminology, and the logical coherence of the communication. Complainant emotional characteristics include the type of emotion, the intensity of the emotion, the trigger point for the emotion, and the form of expression of the emotion; emotion types include calm, dissatisfaction, anger, and anxiety. Agent competence characteristics include business proficiency, policy understanding, emotional coping skills, and communication adaptability.

[0103] In the dynamic allocation and fusion analysis of feature weights, the Analytic Hierarchy Process (AHP) is used to dynamically allocate weights to five feature dimensions based on call scenario type and work order urgency. For urgent public service requests, the weight of the caller's emotional characteristics is increased; for requests seeking professional policy explanations, the weights of the explanation details and service type characteristics are increased. A spatiotemporal fusion analysis is performed on each feature dimension, combining the timestamps corresponding to the features to analyze the changing trends and correlations of the features.

[0104] The specific process of weight allocation using the Analytic Hierarchy Process (AHP) is as follows: A judgment matrix is ​​constructed based on characteristics of business type, response status, verbal style, caller's emotional state, and agent competence. The elements of the judgment matrix represent the importance scale of the row dimension relative to the column dimension. The largest eigenvalue and corresponding eigenvector of the judgment matrix are calculated. The eigenvectors are then normalized to obtain the weight vectors for each dimension. The weight vectors are expressed as follows: in, Represents the weight vector. The weights represent the characteristics of the business type. The weights represent the features of the solution. Weights representing the characteristics of discourse. The weighting of the emotional characteristics of the complainant. The weights represent the seat capacity characteristics.

[0105] The consistency index is calculated based on the largest eigenvalue of the judgment matrix and the order of the judgment matrix. The formula for calculating the consistency index is as follows: in, Indicators of consistency This represents the largest eigenvalue of the judgment matrix. This indicates the order of the judgment matrix. The consistency ratio is calculated based on the consistency index and the average random consistency index. The formula for the consistency ratio is: in, Indicates the consistency ratio. This represents the average random consistency index. When the consistency ratio is less than a preset threshold, the consistency test is passed, and the weights of each feature dimension are used as the weight allocation result of the multi-dimensional call features.

[0106] In the inference and output of the intervention suggestion model, the multidimensional feature data after fusion analysis is input into the pre-trained intervention suggestion model. Based on the feature data and preset intervention rules, the model outputs intervention suggestions. These suggestions include intervention type, timing, priority, and auxiliary information, while also outputting the triggering core feature basis. Intervention types include intelligent listening, intelligent whispering, or intelligent takeover.

[0107] Figure 4 A detailed flowchart illustrating the triggering and execution phases of tiered intelligent intervention is provided. Figure 4 As shown, intervention actions are triggered based on intervention recommendations. These actions include intelligent listening, intelligent whispering, and intelligent takeover. The three intervention actions operate independently and can be seamlessly switched.

[0108] During intelligent monitoring, when trigger conditions are met, a monitoring alert is automatically pushed to the supervisor's terminal. The alert includes the agent's identification, the content of the request, and the triggering reason. Triggering reasons include suspected errors in policy interpretation and emotional fluctuations exceeding a threshold. After confirmation by the supervisor, an independent monitoring channel is established, unnoticed by the agent or the person making the request, and real-time transcribed text and AI analysis tags are displayed simultaneously.

[0109] In the intelligent whisper mode, once triggered, the system retrieves the corresponding policy interpretation text or handling suggestion text from the knowledge base. The handling suggestion is then converted into a concise speech using speech synthesis and pushed to the agent terminal through an independent audio channel, inaudible to the complainant. Simultaneously, the policy interpretation text or handling suggestion text is sent to the corresponding display screen on the agent terminal.

[0110] During intelligent takeover, a seamless switchover occurs immediately upon triggering, cutting off the original agent's communication channel with the complainant and establishing a new channel between the team leader or expert and the complainant, simultaneously sending standardized reassuring messages to the complainant. The original agent is also informed of the takeover reason, policy basis, and follow-up requirements. Two modes are supported: an agent observer mode for training and a complete isolation mode for handling major risk complaints.

[0111] In data accumulation and model optimization, all intervention process data is collected into a big data platform. This data includes voice data, transcribed text, intervention actions, handling results, and agent feedback. The intervention process data is then integrated with historical work orders, grid data, and policy documents. Through data mining analysis, the keyword recognition accuracy and emotion judgment precision of the automatic speech recognition model are optimized. Intervention thresholds are dynamically adjusted based on the specific needs of different hotline areas.

[0112] Figure 5 A schematic diagram of an embodiment of a hotline call traffic tiered intervention system 200 provided by the present invention is shown. Figure 5 As shown, the hotline call handling tiered intervention system 200 includes: The processing module 201 is used to collect two-way voice streams between the agent and the caller, and associate them with call tags; to perform noise reduction processing on the two-way voice streams, to transcribe the noise-reduced voice data into text data, to perform sentence segmentation processing on the text data to obtain text segments with timestamps, and to desensitize sensitive information in each text segment. Extraction module 202 is used to extract multi-dimensional call features based on each text segment and the call tag; The fusion module 203 is used to assign weights to the multidimensional call features according to the call scenario type and the urgency of the work order, and to perform fusion analysis based on the multidimensional call features and the corresponding weights to obtain the fusion analysis result. Analysis module 204 is used to perform spatiotemporal correlation analysis on the multidimensional call features based on the timestamps corresponding to the multidimensional call features, and obtain the feature change trend and the correlation between features; Output module 205 is used to input the fusion analysis results, the feature change trend and the correlation between the features into the intervention suggestion model, and output intervention suggestions. The intervention suggestions include intervention type, intervention timing and intervention priority. The intervention type is intelligent listening, intelligent whispering or intelligent takeover. Trigger module 206 is used to trigger corresponding intervention actions according to the intervention timing and intervention priority, based on the intervention type: if the intervention type is intelligent monitoring, a monitoring reminder is pushed to the supervisor's terminal and an independent monitoring channel is established; if the intervention type is intelligent whispering, a handling suggestion is retrieved from the knowledge base, converted into voice, and pushed to the agent's terminal; if the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, and a call channel between the supervisor and the complainant is established.

[0113] In one alternative approach, the call handling tags include basic tags and extended tags. The basic tags include work order number, agent ID, source of request, and acceptance time. The extended tags include agent real-time status, agent capability profile tags, work order urgency level, and the caller's historical request records. All call handling tags are associated through session IDs.

[0114] In one alternative embodiment, the processing module 201 is specifically used for: An automatic speech recognition model is used to transcribe the noise-reduced speech data into initial text data. After transcription, the initial text data is corrected using a hotline terminology database and a script database. Typos are corrected and missing words are added to obtain the corrected text data.

[0115] In one alternative approach, the multidimensional call features include service type features, response status features, speech features, caller emotional features, and agent capability features. The business type characteristics include the area of ​​the complaint, the business subcategory, the urgency of the complaint, the difficulty of resolving the complaint, and the relevance of historical complaints. The response characteristics include the completeness of the response, the accuracy of the policy basis, the timeliness of the response, and the satisfaction of the response result. The discourse characteristics include the standardization of the discourse, the use of polite language, the accuracy of professional terminology, and the rationality of the communication logic. The emotional characteristics of the complainant include the type of emotion, the intensity of emotion, the trigger point of emotion, and the form of emotion expression. The agent's ability characteristics include business proficiency, policy understanding, emotional coping ability, and communication adaptability.

[0116] In one alternative embodiment, the fusion module 203 is specifically used for: A judgment matrix is ​​constructed based on the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics. The maximum eigenvalue and the corresponding eigenvector of the judgment matrix are calculated. The eigenvector is normalized to obtain the weights of the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics. A consistency index is calculated based on the maximum eigenvalue of the judgment matrix and the order of the judgment matrix. A consistency ratio is calculated based on the consistency index and the average random consistency index. When the consistency ratio is less than a preset threshold, a consistency test is passed. The weights of each dimension feature are used as the weight allocation result of the multi-dimensional call features. Specifically, in the scenario of urgent public appeals, the weight of the appealer's emotional characteristics is increased; in the scenario of professional policy explanation, the weights of the explanation situation characteristics and the business type characteristics are increased.

[0117] In an alternative embodiment, the analysis module 204 is specifically used for: Based on the timestamps corresponding to the business type characteristics, the response characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics, the temporal changes of each characteristic during the call are analyzed to obtain the characteristic change trend. Extract the correlation between the change of the emotional characteristics of the complainant and the change of the response characteristics, extract the matching degree correlation between the verbal characteristics and the business type characteristics, and take the correlation and the matching degree correlation as the correlation between the features.

[0118] In one alternative approach, the intervention suggestion further includes intervention support information and triggering core feature basis; the triggering module 206 is also used for: If the intervention type is intelligent monitoring, then based on the triggering core characteristics, a monitoring reminder containing the seat ID, the request content and the triggering reason is pushed to the squad leader terminal, an independent monitoring channel is established and the real-time transcribed text is displayed; If the intervention type is intelligent whisper, then the treatment suggestion is retrieved from the knowledge base according to the intervention assistance information, converted into voice and pushed to the agent terminal, and the text content of the treatment suggestion is sent to the display screen of the agent terminal at the same time. If the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, a call channel between the team leader and the complainant is established, a reassuring voice message is sent to the complainant, and the reason for the takeover, the policy basis, and the follow-up requirements are sent to the original agent. The data from the entire intervention process will be collected into a big data platform. The data from the entire intervention process includes the voice data, the text data, the intervention actions, the treatment suggestions, and the agent feedback. The data from the entire intervention process is mined, and the recognition accuracy and emotion judgment precision of the automatic speech recognition model are optimized based on the mining results. The intervention threshold is also dynamically adjusted according to the characteristics of the demands in the hotline field.

[0119] It should be noted that the beneficial effects of the hotline traffic tiered intervention system 200 provided in the above embodiments are the same as those of the above-described hotline traffic tiered intervention method, and will not be repeated here. Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0120] The hotline call tiered intervention system 200 of the present invention can be a computer program (including program code) running on a computer device. For example, the hotline call tiered intervention system 200 of the present invention is an application software that can be used to execute the corresponding steps in the hotline call tiered intervention method of the present invention.

[0121] In some embodiments, the hotline traffic classification intervention system 200 of the present invention can be implemented in a combination of hardware and software. As an example, the hotline traffic classification intervention system 200 of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the hotline traffic classification intervention method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0122] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0123] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned hotline traffic classification intervention methods. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the hotline traffic classification intervention method shown in any embodiment of the present invention by calling the computer program.

[0124] In one alternative embodiment, an electronic device is provided, such as Figure 6 As shown, Figure 6 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0125] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0126] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0127] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0128] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0129] Among them, electronic devices can also be terminal devices. A terminal device can be any terminal device that can install applications and access web pages through applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0130] It should be noted that, Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0131] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described hotline call tiered intervention methods.

[0132] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0133] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned hotline traffic tiered intervention method.

[0134] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0135] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0136] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0137] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0138] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0139] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0140] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0141] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for tiered intervention of hotline call volume, characterized in that, include: Collect two-way voice streams between the agent and the caller, and associate them with call tags; The bidirectional speech stream is denoised, the denoised speech data is transcribed into text data, the text data is segmented into sentences to obtain text segments with timestamps, and sensitive information in each text segment is desensitized. Extract multidimensional call features based on each text segment and the call tag; Based on the call scenario type and the urgency of the work order, the analytic hierarchy process (AHP) is used to assign weights to the multidimensional call features, and a fusion analysis is performed based on the multidimensional call features and their corresponding weights to obtain the fusion analysis results. Based on the timestamps corresponding to the multidimensional call features, spatiotemporal correlation analysis is performed on the multidimensional call features to obtain the feature change trend and the correlation between features; The fusion analysis results, the feature change trends, and the correlations between the features are input into the intervention suggestion model, and intervention suggestions are output. The intervention suggestions include intervention type, intervention timing, and intervention priority. The intervention type is intelligent listening, intelligent whispering, or intelligent takeover. According to the intervention timing and priority, the corresponding intervention action is triggered based on the intervention type: if the intervention type is intelligent monitoring, a monitoring reminder is pushed to the supervisor's terminal and an independent monitoring channel is established; if the intervention type is intelligent whispering, a handling suggestion is retrieved from the knowledge base, converted into voice, and pushed to the agent's terminal; if the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, and a call channel between the supervisor and the complainant is established.

2. The hotline call handling tiered intervention method according to claim 1, characterized in that, The call handling tags include basic tags and extended tags. The basic tags include work order number, agent ID, source of request, and acceptance time. The extended tags include agent real-time status, agent capability profile tags, work order urgency level, and the caller's historical request records. All call handling tags are associated through session ID.

3. The hotline call traffic tiered intervention method according to claim 1, characterized in that, The steps for transcribing the denoised speech data into text data include: An automatic speech recognition model is used to transcribe the noise-reduced speech data into initial text data. After transcription, the initial text data is corrected using a hotline terminology database and a script database. Typos are corrected and missing words are added to obtain the corrected text data.

4. The hotline call traffic tiered intervention method according to claim 1, characterized in that, The multi-dimensional call features include service type features, response status features, speech and utterance features, caller's emotional features, and agent's ability features; The business type characteristics include the area of ​​the complaint, the business subcategory, the urgency of the complaint, the difficulty of resolving the complaint, and the relevance of historical complaints. The response characteristics include the completeness of the response, the accuracy of the policy basis, the timeliness of the response, and the satisfaction of the response result. The discourse characteristics include the standardization of the discourse, the use of polite language, the accuracy of professional terminology, and the rationality of the communication logic. The emotional characteristics of the complainant include the type of emotion, the intensity of emotion, the trigger point of emotion, and the form of emotion expression. The agent's ability characteristics include business proficiency, policy understanding, emotional coping ability, and communication adaptability.

5. The hotline call traffic tiered intervention method according to claim 4, characterized in that, The steps of assigning weights to the multidimensional call features using the analytic hierarchy process (AHP) include: A judgment matrix is ​​constructed based on the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics. The maximum eigenvalue and the corresponding eigenvector of the judgment matrix are calculated. The eigenvector is normalized to obtain the weights of the business type characteristics, the response status characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics. A consistency index is calculated based on the maximum eigenvalue of the judgment matrix and the order of the judgment matrix. A consistency ratio is calculated based on the consistency index and the average random consistency index. When the consistency ratio is less than a preset threshold, a consistency test is passed. The weights of each dimension feature are used as the weight allocation result of the multi-dimensional call features. Specifically, in the scenario of urgent public appeals, the weight of the appealer's emotional characteristics is increased; in the scenario of professional policy explanation, the weights of the explanation situation characteristics and the business type characteristics are increased.

6. The hotline call traffic tiered intervention method according to claim 5, characterized in that, The steps of performing spatiotemporal correlation analysis on the multidimensional call features to obtain feature change trends and inter-feature correlations include: Based on the timestamps corresponding to the business type characteristics, the response characteristics, the dialogue characteristics, the complainant's emotional characteristics, and the agent's ability characteristics, the temporal changes of each characteristic during the call are analyzed to obtain the characteristic change trend. Extract the correlation between the change of the emotional characteristics of the complainant and the change of the response characteristics, extract the matching degree correlation between the verbal characteristics and the business type characteristics, and take the correlation and the matching degree correlation as the correlation between the features.

7. The hotline call tiered intervention method according to claim 6, characterized in that, The intervention recommendations also include intervention support information and triggering core feature basis; The step of triggering the corresponding intervention action according to the intervention type based on the intervention timing and the intervention priority further includes: If the intervention type is intelligent monitoring, then based on the triggering core characteristics, a monitoring reminder containing the seat ID, the request content and the triggering reason is pushed to the squad leader terminal, an independent monitoring channel is established and the real-time transcribed text is displayed; If the intervention type is intelligent whisper, then the treatment suggestion is retrieved from the knowledge base according to the intervention assistance information, converted into voice and pushed to the agent terminal, and the text content of the treatment suggestion is sent to the display screen of the agent terminal at the same time. If the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, a call channel between the team leader and the complainant is established, a reassuring voice message is sent to the complainant, and the reason for the takeover, the policy basis, and the follow-up requirements are sent to the original agent. The data from the entire intervention process will be collected into a big data platform. The data from the entire intervention process includes the voice data, the text data, the intervention actions, the treatment suggestions, and the agent feedback. The data from the entire intervention process is mined, and the recognition accuracy and emotion judgment precision of the automatic speech recognition model are optimized based on the mining results. The intervention threshold is also dynamically adjusted according to the characteristics of the demands in the hotline field.

8. A hotline call tiered intervention system, characterized in that, include: The processing module is used to collect two-way voice streams between the agent and the caller, and associate them with call tags; The bidirectional speech stream is denoised, the denoised speech data is transcribed into text data, the text data is segmented into sentences to obtain text segments with timestamps, and sensitive information in each text segment is desensitized. The extraction module is used to extract multidimensional call features based on each text segment and the call tag; The fusion module is used to assign weights to the multidimensional call features according to the call scenario type and the urgency of the work order, and to perform fusion analysis based on the multidimensional call features and their corresponding weights to obtain the fusion analysis results. The analysis module is used to perform spatiotemporal correlation analysis on the multidimensional call features based on the timestamps corresponding to the multidimensional call features, so as to obtain the feature change trend and the correlation between features; The output module is used to input the fusion analysis results, the feature change trends, and the correlation between the features into the intervention suggestion model and output intervention suggestions. The intervention suggestions include intervention type, intervention timing, and intervention priority. The intervention type is intelligent listening, intelligent whispering, or intelligent takeover. The triggering module is used to trigger corresponding intervention actions according to the intervention timing and priority, and based on the intervention type: if the intervention type is intelligent monitoring, a monitoring reminder is pushed to the supervisor's terminal and an independent monitoring channel is established; if the intervention type is intelligent whispering, a handling suggestion is retrieved from the knowledge base, converted into voice, and pushed to the agent's terminal; if the intervention type is intelligent takeover, the call channel between the agent and the complainant is cut off, and a call channel between the supervisor and the complainant is established.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory, the memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the hotline call tiered intervention method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, implements the hotline traffic tiered intervention method as described in any one of claims 1 to 7.

Images