12345 hotline incoming call service insight method and system

Abstract

The invention discloses a 12345 hotline incoming call service insight method and system, and the method comprises the steps: S1), carrying out the number recognition after an incoming call is accessed; s2) if the incoming call is the first incoming call, recording basic information, entering voice navigation, and skipping to S6) after the processing is finished; if the incoming call is not the first incoming call, generating an intelligent label, and judging a work order type; s3) if the work order is a work order with a problem to be solved, calling an intelligent work order management module for processing; s4) if no work order with the problem to be solved exists, judging whether the user is a vehicle moving work order user; s5) if the user is a vehicle moving work order user, automatically querying a previous vehicle moving record, and querying whether to add a call or not; if not, switching to a seat skill group intelligent distribution module for processing; and S6) calling a call data statistical feedback module for processing, and updating the user portrait. The application solves the problems of tedious process, low efficiency and the like of a traditional system, improves the working efficiency and the service quality of the 12345 hotline, improves the user experience, and reduces the pressure of customer service.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent seat voice calls, and particularly relates to a method and system for insight into incoming calls of the 12345 hotline. Background Art

[0002] In the current field of intelligent voice call technology, especially in applications related to government service hotlines, some technical solutions and products relatively close to the present invention have emerged.

[0003] For example, the patent named "A Personnel Identification and Sending System for Intelligent Work Order Call Traffic" (publication number CN 118798578 A, application date September 2024) applied by State Grid Siji Feitian (Lanzhou) Cloud Data Technology Co., Ltd. This system includes a voice collection module, a voice conversion module, an information processing module, a personnel identification module, a task scheduling module, and a work order allocation module. By identifying call traffic voices, it automatically generates work orders and sets priorities, reducing the workload of manual entry and management. It can also intelligently identify the task types in the work orders and extract relevant information, taking the most relevant maintenance type as the task entity of the work order, reducing the risk of misjudgment, comprehensively capturing the various requirements of the work order, and providing detailed basis for subsequent personnel matching and task allocation. The system constructs an intelligent and dynamically updated knowledge graph by mapping the task entities and personnel entities and the relationships between the entities, improving the accuracy of task matching, having strong adaptability and flexibility, and ensuring the optimal utilization of resources and the efficient completion of tasks. However, this patent mainly focuses on the matching of personnel and tasks in the work order processing flow, and has not reached the highly integrated and intelligent level expected by the present invention in terms of the comprehensiveness of insight into the intentions of user incoming calls and the personalized and automated services for different types of user incoming calls. It does not fully combine user portraits, making it difficult to achieve pre-judgment and proactive services for user incoming calls, and there are certain limitations in enhancing the user interaction experience and reducing the pressure on manual customer service.

[0004] For another example, the patent "A Processing Method and Application for Real - time Call Q&A Matching" applied by Tianjin Chezhijia Software Co., Ltd. (publication number CN 119107951 A, application date September 2024). This invention belongs to the technical field of call service processing. Through real - time speech - to - text conversion, the pre - processed speech data is sent to the ASR service for real - time conversion; real - time text processing and grouping extract, sort, and group the text through role detection and grouping and timestamp correction; after - call text processing matches the text with the groups already identified during the call. By dispersing the invocation of the TEXTCNN model during the call process, this patent significantly reduces the time required for centralized text processing after the call, ensuring that the performance requirement of outputting filing information within 5 seconds after hanging up can be met even during peak hours. However, the focus of this patent is on the processing method of call Q&A matching and the control of the response duration of filing information, lacking in - depth mining and utilization of the user's historical call data, unable to effectively insight into the current call intention based on the user's past call situations, difficult to achieve intelligent classification and efficient processing of call services, and unable to well meet the application scenario requirements such as the 12345 hotline that needs to quickly and accurately respond to various demands of citizens.

[0005] The following core defects exist in the prior art or products in the processing of 12345 hotline call services: 1. Insufficient ability to insight into user intentions, relying on simple speech recognition or task tags, unable to achieve proactive services of "predicting user needs as soon as the user calls", resulting in artificial customer service needing to repeatedly ask users for information, reducing efficiency.

[0006] For example, the patent of State Grid Siji Feitian (CN 118798578 A) only generates work orders and assigns tasks through speech recognition, lacking in - depth integration and analysis of the user's historical call data, work order records, and interaction behaviors, and unable to accurately predict the user's call intention (such as whether it belongs to repeated complaints, urgent demands, or emotionally sensitive users).

[0007] The patent of Tianjin Chezhijia (CN 119107951 A) focuses on the real - time processing of call Q&A matching, but does not use user portraits or historical data for call intention classification, resulting in service responses remaining at the passive processing level.

[0008] 2. Lack of personalized service processes, with fixed service processes, unable to meet the diverse needs of users, resulting in poor user experience (such as multiple transfers, repeated operations), and artificial customer service needs to handle a large number of low - value consultations.

[0009] Existing systems usually adopt fixed IVR menus or general work order processing procedures, without designing differentiated service paths for different user types (such as first - time callers, high - frequency complaint users, parking lot transfer work order users).

[0010] For example, the patent of State Grid Siji Feitian does not distinguish user types and only matches personnel according to task entities, and cannot automatically push the work order progress for "users with problems to be solved" or preferentially transfer "users to be comforted emotionally" to the psychological counseling seat.

[0011] 3. The efficiency of intelligent work order processing and seat allocation is low. The work order processing relies on manual intervention, and the seat matching accuracy is low, resulting in multiple transfers for complex problems and extending the processing cycle.

[0012] The patent of Tianjin Chezhijia only optimizes the response speed of the information filed after the call, but does not realize the intelligent processing of work orders (such as automatically obtaining the work order progress and dynamically updating the status).

[0013] Seat allocation relies on simple skill tags or busy / idle status, lacking a comprehensive evaluation of the complexity of user needs, the professional ability of seats, and the historical service quality (such as the patent of State Grid Siji Feitian).

[0014] 4. The data-driven system optimization is insufficient. It is impossible to dynamically optimize the line allocation strategy, user classification algorithm, or seat skill training direction through data feedback. The system iteration relies on manual experience, and the intelligent level is limited.

[0015] Most systems only collect basic data such as call duration and connection rate, and do not establish a correlation analysis model between call data, user portraits, work order processing, and seat capabilities.

[0016] 5. The multi-module collaboration ability is weak, and the problem of information silos is significant. Customer service personnel need to switch between multiple systems, reducing service efficiency.

[0017] Modules such as user portraits, work order management, and seat allocation are independent of each other, and data interaction is not smooth. For example, the patent of State Grid Siji Feitian does not synchronize user portrait information to the seat terminal, resulting in customer service personnel having to manually query the user's historical data.

[0018] Overall, the current existing technologies have not yet formed a complete, efficient, and intelligent system in the processing of 12345 hotline incoming calls, and have not organically combined functions such as insight into user incoming call intentions, provision of personalized services, intelligent work order processing, and seat allocation to achieve all-round optimization of incoming call services. Summary of the Invention

[0019] Objective of the Invention: To overcome the above deficiencies, the objective of the present invention is to provide a method and system for 12345 hotline incoming call service insight. When a citizen makes a call, the call first accesses this system. The system intelligently matches the user's historical calls, records, and work orders through the incoming call number, combines the user portrait and intelligent tags, and uses large language models for analysis and reasoning to insight the call intention and automatically provide services for citizens. The present invention solves problems such as the cumbersome process, low efficiency, and low accuracy of manual matching in the traditional call center IVR voice navigation system, improves the work efficiency and service quality of the 12345 hotline, improves the user experience, and reduces the work pressure of customer service staff.

[0020] Technical Solution: To achieve the above objective, the present invention provides a method for 12345 hotline incoming call service insight, including the following steps: S1): After the incoming call is connected, call the incoming and outgoing call management module for number identification and select a line according to user classification. S2): If it is the first call, record the basic information and generate a user ID, enter the voice navigation, and jump to S6) after the processing is completed; if it is not the first call, call the user portrait module to generate intelligent tags in combination with user portrait data, quickly match the current user type according to the existing user type, and determine whether there is a work order with problems to be solved; when making the first call, the user does not need to be asked for identity information manually, reducing the complexity of the work process. The generation of intelligent tags by the user portrait module in combination with user portrait data includes: S201): Analyze the user's historical conversation text and real-time conversation text based on a large language model, extract key semantic features, and generate emotion tags and business tags. S202): Generate intelligent tags through a rule engine in combination with user portrait data. S3): If there is a work order with problems to be solved, call the intelligent work order management module for intelligent processing. The specific intelligent processing by the intelligent work order management module in S3) is to automatically obtain the real-time progress of the work order to be solved from the database and broadcast it to the user. S4): If there is no work order with problems to be solved, determine whether the user is a parking lot transfer work order user. S5): If it is a user of a vehicle relocation work order, automatically query the previous vehicle relocation record and ask whether to add a call; if it is not a user of a vehicle relocation work order, switch to the intelligent allocation module of the agent skill group for intelligent agent allocation and select a human agent for processing; for users with problems to be solved, automatically obtain the progress of the work order and broadcast it through the voice robot, and the user can independently choose whether to transfer to a human agent; at the same time, since vehicle relocation work orders account for 30% of the daily processed work orders, additional classification of vehicle relocation work order users is added. If it is a user of a vehicle relocation work order, the system will automatically add a call to the vehicle owner who has not relocated, and the new work order will be directly transferred to the automatic vehicle relocation robot for processing, greatly reducing the processing time of vehicle relocation work orders; The agent intelligent allocation in S5) includes: S501): Calculate the complexity of user requirements; S502): Calculate the comprehensive allocation score of the agent; S503): Select the agent with the highest comprehensive allocation score and assign the incoming call to this agent; the intelligent allocation of the agent skill group matches user requirements with agent skills based on the Hungarian algorithm and dynamically adjusts according to the current load of the agent; S6): Call the call data statistics and feedback module for data statistics, generate a report, and call the user portrait module to update the user portrait. It deeply integrates the user's historical call records (such as call frequency, work order type), work order content, and interaction behavior data to construct a multi-dimensional user portrait, greatly improving the accuracy of predicting the needs of non-first-time incoming call users; at the same time, based on the large language model, analyze the user's historical conversations, extract key semantic features, and combine intelligent tags to achieve accurate classification of incoming call intentions.

[0021] Further, the update of the user portrait in S6) includes: S601): Data collection, obtain relevant data from different data sources, including but not limited to historical incoming call records, work order information, and interaction behavior data of the call data statistics and feedback module; S602): Data cleaning, clean the collected data, including but not limited to removing duplicate data, handling missing values, and correcting incorrect data; S603): Data association, associate the data in different data sources through the user's unique identifier (such as mobile phone number, user ID); S604): Feature extraction and transformation, including but not limited to calculating behavior statistics, text mining, and data standardization; S605): Data fusion, integrating the extracted and transformed features into a unified user portrait model to form a multi-dimensional user portrait. By combining the user's historical behaviors (such as repeated complaints, work order processing progress), it predicts the incoming call intention, avoiding the need for manual customer service to repeatedly ask for information and improving work efficiency. At the same time, by constructing a multi-dimensional user portrait, it is possible to understand the user more comprehensively and provide strong support for the accurate prediction of incoming call intentions.

[0022] Further, the multi-dimensional user portrait described in S605) includes a basic information dimension, a behavior preference dimension, and a business demand tendency dimension; The basic information dimension includes demographic information and contact information; The behavior preference dimension includes call behavior, interaction behavior, and work order behavior; The business demand tendency dimension includes topic clustering of work order texts and call records to identify the business areas and problem types that the user is concerned about, and obtaining the user's satisfaction and expectations for different services through sentiment analysis of texts and voices. In addition to the three main dimensions of basic information, behavior preference, and business demand tendency, each dimension is further divided into multiple specific features. Through such a user portrait, it is possible to understand the user more comprehensively and provide strong support for the accurate prediction of incoming call intentions.

[0023] Further, the S201) includes: S20101): Using historical dialogue texts, user portrait data, and real-time interaction data as basic data for text preprocessing; S20102): Tokenizing and vectorizing the text after text preprocessing; S20103): Using semantic tags generated by a large language model and mapping them to the user portrait tag system to extract key semantic features, thereby generating emotion tags and business tags. By combining historical data with real-time analysis, it improves the accuracy of intention classification, avoids service biases caused by simple voice keyword matching, and reduces the misjudgment rate.

[0024] Further, the S202) specifically realizes the mapping alignment of emotion tags and business tags with user portrait data tags through a fast matching algorithm, maps the emotion tags and business tags into the user portrait data tag categories, and generates intelligent tags. Through the mapping alignment of emotion tags and business tags, it can more accurately identify the user's needs and emotional states, thereby providing more personalized services, greatly improving the matching accuracy and response speed, and at the same time reducing the misjudgment rate.

[0025] Further, for the S501), the user demand complexity is calculated comprehensively from the complaint level, work order urgency, and problem complexity. The specific method is to use weighted average, and the formula is as follows: User requirement complexity = Quantification value of complaint level × Weight of complaint level + Quantification value of work order urgency × Weight of work order urgency + Quantification value of problem complexity × Weight of problem complexity; Among them, the quantification value of the complaint level is divided into five levels: 1, 2, 3, 4, and 5 according to the complaint level; the quantification value of the work order urgency is divided into three levels: 1, 2, and 3 according to the time requirement for resolving the work order; the quantification value of the problem complexity is evaluated according to factors such as the business scope and technical difficulty involved in the problem, and is divided into three levels: 1, 2, and 3. In actual user requirements, a single factor often cannot fully reflect the situation comprehensively. Therefore, a comprehensive evaluation is designed from three key dimensions: complaint level, work order urgency, and problem complexity, which can more comprehensively capture all aspects of user requirements and avoid misjudgment caused by one-sidedness.

[0026] Furthermore, the comprehensive allocation score of the agent in step S502) is calculated from the user requirement complexity, the agent skill matching degree, and the quantification value of the agent's current load. The specific method is weighted average, and the formula is as follows: Comprehensive allocation score = User requirement complexity × Weight of user requirement complexity + Agent skill matching degree × Weight of agent skill matching degree + (4 - Quantification value of agent's current load) × Weight of agent's current load; Among them, the agent skill matching degree is 1 if the agent has the ability to handle problems of this type of user, and 0 if not; the quantification value of the agent's current load is divided into three levels: 1, 2, and 3 according to the number of work orders being processed by the agent. By comprehensively considering three key factors: user requirement complexity, agent skill matching degree, and quantification value of the agent's current load, the allocation situation of the agent is comprehensively evaluated; through the quantification operation of the comprehensive allocation score for problems and agents, complex problems are directly assigned to high-skill agents, avoiding multiple transfers and improving the professionalism of the service at the same time.

[0027] The present invention also provides a system for insight into the incoming calls of the 12345 hotline, which is used to implement a method for insight into the incoming calls of the 12345 hotline, including: an incoming and outgoing call management module, a user profile module, an intelligent work order management module, an intelligent allocation module for the seat skill group, and a call data statistics and feedback module; the incoming and outgoing call management module, the user profile module, the intelligent work order management module, the intelligent allocation module for the seat skill group, and the call data statistics and feedback module are connected in sequence, and the call data statistics and feedback module is respectively connected to the incoming and outgoing call management module and the user profile module; the incoming and outgoing call management module is used to identify numbers and select lines according to user classification; the user profile module is used to update and construct user profiles; the intelligent work order management module is used to intelligently process work orders with problems to be solved; the intelligent allocation module for the seat skill group is used to intelligently allocate seats for users; the call data statistics and feedback module is used for data statistics, generating reports, and providing data to the user profile module for updating. Each module synchronizes information in real time through a unified data bus (RESTful API, WebSocket). For example, the user profile module pushes and synchronizes tags to the seat terminal through the WebRTC protocol; the seat terminal displays key user information in real time (such as "has made 3 historical complaints" in red), and there is no need to switch systems to query, eliminating the problem of information silos and greatly improving the accuracy of cross-system operations.

[0028] As can be seen from the above technical solutions, the present invention has the following beneficial effects: 1. The method and system for insight into the incoming calls of the 12345 hotline according to the present invention integrate the user's historical call records, work order information, and interaction behavior data, construct a multi-dimensional profile including basic information, behavior preferences, and business demand tendencies, deeply integrate the user profile with historical data, and achieve accurate prediction of incoming call intentions; 2. The method and system for insight into the incoming calls of the 12345 hotline according to the present invention, through the design of a differentiated service process, reduce manual intervention and repeated interactions, avoid users repeating problem descriptions, reduce the number of transfers, and improve satisfaction; 3. The method and system for insight into the incoming calls of the 12345 hotline according to the present invention, with the design of dynamic intelligent work order management and intelligent seat allocation, achieve accurate matching of "requirements - skills", and greatly improve the response efficiency; 4. The method and system for insight into the incoming calls of the 12345 hotline according to the present invention, the call data statistics and feedback module statistically analyzes the data and generates reports, providing data support for the decision-making of the system; 5. The method and system for insight into the incoming calls of the 12345 hotline according to the present invention, with a multi-module collaborative architecture, eliminates information silos and improves service quality and efficiency. Description of the Drawings

[0029] Figure 1 This is the step diagram of a method for 12345 hotline call business insight according to the present invention; Figure 2 This is the overall flow chart of a method for 12345 hotline call business insight according to the present invention; Figure 3 This is the flow chart of user portrait generation in a method for 12345 hotline call business insight according to the present invention; Figure 4 This is the flow chart of intelligent label generation in a method for 12345 hotline call business insight according to the present invention; Figure 5 This is the module schematic diagram of a 12345 hotline call business insight system according to the present invention; Figure 6 This is the data statistics and feedback mechanism diagram of the call data statistics and feedback module in a 12345 hotline call business insight system according to the present invention. Detailed implementation manners

[0030] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present invention, but should not be construed as limiting the present invention.

[0031] Embodiment 1 In this embodiment, as Figure 1 and Figure 2 , the present invention discloses a method for 12345 hotline call business insight, including the following steps: S1): After the call is connected, call the incoming and outgoing call management module to identify the number, and select the line according to the user classification; S2): If it is the first call, record the basic information and generate a user ID, enter the voice navigation, and jump to S6) after the processing is completed; if it is not the first call, call the user portrait module to generate intelligent labels in combination with the user portrait data, quickly match the current user type according to the existing user types, and determine whether it is a work order with problems to be solved; The generation of intelligent labels by the user portrait module in combination with the user portrait data includes: S201): Analyze the user's historical conversation text and real-time conversation text based on the large language model, extract the key semantic features, and generate emotion labels and business labels; S202): Combine the user portrait data and generate intelligent labels through the rule engine.

[0032] S3): If it is a work order with problems to be solved, call the intelligent work order management module for intelligent processing; The intelligent processing by the intelligent work order management module in S3) is specifically to automatically obtain the real-time progress of the work order to be solved from the database and report it to the user through voice broadcast; S4): If it is not a work order with problems to be solved, determine whether the user is a parking lot relocation work order user; S5): If it is a parking lot relocation work order user, automatically query the previous parking lot relocation record and ask whether to add a call; if it is not a parking lot relocation work order user, switch to the agent skill group intelligent allocation module for intelligent agent allocation and select an artificial agent for processing; The agent intelligent allocation in S5) includes: S501): Calculate the complexity of user requirements; S502): Calculate the comprehensive allocation score of the agent; S503): Select the agent with the highest comprehensive allocation score and allocate the incoming call to this agent; S6): Call the call data statistics and feedback module for data statistics, generate a report, and call the user portrait module to update the user portrait.

[0033] Specifically, referring to Figure 4 , the following is an exemplary scenario for generating possible intelligent tags: Scenario: User A dials the 12345 hotline, (1) Data acquisition: Historical conversation text: "The construction noise downstairs in my house is very serious. I have complained 3 times, and every time they said they would handle it but nothing has been solved!" Real-time data: Emotion value 92 points, speech rate 220 words / minute, interrupted 2 times; (2) LLM analysis results: Emotion tendency: Angry (confidence level 95%), problem type: Noise complaint (confidence level 98%), entity recognition: Construction noise (location: Chaoyang District); (3) Tag generation: Intelligent tags: ["Users with urgent demands", "Users with high-frequency complaints"].

[0034] Specifically, according to the data of the 12345 hotline in a certain province, the proportion of parking lot relocation work orders reaches 28% (average daily processing volume of more than 1200), and the repeated complaint rate is as high as 35% (multiple transfers are required in the traditional process). Therefore, parking lot relocation work order users are classified independently and a customized process is set.

[0035] In this embodiment, as Figure 3 , the update of the user portrait in S6) includes: S601): Data collection, obtaining relevant data from different data sources, including but not limited to historical call records, work order information, and interaction behavior data from the call data statistics feedback module; the historical call records are stored in a MySQL database, containing information such as call time, call duration, call topic, etc.; the work order information is stored in Elasticsearch, covering work order creation time, processing progress, processing result, problem type, etc.; the interaction behavior data includes but not limited to user key operations, voice emotion values, speech rate, volume, etc., from real-time call records and behavior log systems.

[0036] S602): Data cleaning, cleaning the collected data, including but not limited to removing duplicate data, handling missing values, and correcting incorrect data; for missing values, they can be processed by filling with mean, median or deletion. S603): Data association, associating data from different data sources through a user unique identifier (such as mobile phone number, user ID) to ensure that different data of the same user can be accurately corresponded. S604): Feature extraction and transformation, including but not limited to calculating behavior statistics, text mining, and data standardization; calculating behavior statistics includes but not limited to the average call duration of users, complaint frequency, work order resolution time, etc.; text mining includes performing sentiment analysis and topic clustering on work order texts and call records to extract key information; data standardization is to standardize features of different scales. S605): Data fusion, fusing the extracted and transformed features into a unified user portrait model to form a multi-dimensional user portrait; weighted average is used as a preferred method for data fusion.

[0037] In this embodiment, the multi-dimensional user portrait described in S605) includes a basic information dimension, a behavior preference dimension, and a business demand tendency dimension. The basic information dimension includes demographic information and contact information. The behavior preference dimension includes call behavior, interaction behavior, and work order behavior. The business demand tendency dimension includes performing topic clustering on work order texts and call records to identify the business areas and problem types that users are concerned about, and obtaining the satisfaction and expectations of users for different services through sentiment analysis of text and voice.

[0038] Specifically, the demographic information includes user registration information and historical records, such as age, gender, region, etc.; the contact information includes mobile phone number, email, etc.

[0039] Specifically, call behaviors include call time patterns (such as peak hours, weekday / weekend preferences), call duration distribution, call frequency, etc.; interaction behaviors include users' key operations, voice emotion values, speech rates, volumes, etc. For example, users with a fast speech rate, high volume, and large fluctuations in emotion values may be more inclined to direct and urgent communication methods; ticket behaviors include users' ticket submission frequencies, problem type preferences, ticket handling satisfaction, etc.

[0040] Specifically, topic analysis includes performing topic clustering on ticket texts and call records to identify the business areas and problem types that users are concerned about. For example, if a user's tickets mainly focus on product fault reporting and after-sales service consultation, it indicates that they have a high demand for these services; sentiment analysis understands users' satisfaction and expectations for different services through sentiment analysis of texts and voices. Positive sentiment may indicate users' recognition of a service, while negative sentiment may imply areas for improvement.

[0041] In this embodiment, as Figure 4 , the step S201) includes: S20101): Using historical dialogue texts, user profile data, and real-time interaction data as basic data for text preprocessing; S20102): Performing word segmentation and vectorization on the text after text preprocessing; S20103): Using semantic tags generated by a large language model and mapping them to the user profile tag system to extract key semantic features, thereby generating emotion tags and business tags.

[0042] Specifically, integrating Baidu Voice / iFlytek's TTS / ASR engines to convert real-time call voices into texts, and performing sentiment classification and topic classification on the texts generated by ASR based on the fine-tuned BERT model combined with prompt engineering, and outputting emotion tags (such as anger, satisfaction), business tags, and confidence levels (such as anger confidence level 92%).

[0043] In particular, the emotion tag generation process also extracts acoustic features as data expansion. In addition to collecting conventional features such as MFCC coefficients, fundamental frequency, and energy, it also focuses on extracting speech rate and volume features. The speech rate is determined by calculating the number of syllables per unit time, and the decibel value of the volume is calculated using an audio processing algorithm. At the same time, the timestamp of the conversation is analyzed to determine whether there is a situation of interrupting the other party. If one party interrupts the other party and the interval is less than the set threshold (such as 0.5 seconds), it is determined to be an interruption and the number of interruptions is recorded. The CNN-LSTM hybrid model is used to analyze the acoustic features. In addition to the original acoustic features, the speech rate, volume, and number of interruptions are used as additional input features. Based on the model learning, the relationship between these features and different emotions is predicted, and the emotion categories (such as anger, anxiety) and probability values ​​(such as 85% probability of anger) are predicted. For example, fast speech rate, high volume, and frequent interruptions are most likely related to anger.

[0044] Specifically, the text features are integrated with the acoustic features, and the text and acoustic analysis results are weighted averaged. At the same time, the weights of speaking speed, volume, and interruption are considered, and finally the emotion label is obtained.

[0045] In this embodiment, if Figure 4 , the S202) is specifically to realize the mapping alignment of emotion tags and business tags with user portrait data labels through a fast matching algorithm, map the emotion tags and business tags to the user portrait data label category, and generate smart tags.

[0046] Specifically, the fast matching algorithm selects cosine similarity matching as a preferred method.

[0047] In this embodiment, the user demand complexity in S501) is calculated by comprehensively calculating the complexity of the complaint level, the urgency of the work order and the complexity of the problem. The specific method is weighted average, and the formula is as follows: User demand complexity = complaint level quantified value × complaint level weight + work order urgency quantified value × work order urgency weight + problem complexity quantified value × problem complexity weight; Among them, the quantitative value of complaint level is divided into five levels: 1, 2, 3, 4, and 5 according to the complaint level; the quantitative value of work order urgency is divided into three levels: 1, 2, and 3 according to the time requirement for resolving the work order; the quantitative value of problem complexity is evaluated based on factors such as the business scope and technical difficulty involved in the problem, and is divided into three levels: 1, 2, and 3.

[0048] Specifically, the five levels of complaint level quantitative values ​​are: Level 1 complaint: Minor issues, such as general consultation misunderstandings, minor service defects, etc., with a quantitative value of 1; Secondary complaints: Medium-level problems, such as poor service attitude and unsmooth partial business processes, with a quantified value set to 2; Tertiary complaints: Relatively serious problems, such as product quality defects and important business processing mistakes, with a quantified value set to 3; Quaternary complaints: Major problems, such as those involving potential safety hazards and causing relatively large economic losses, with a quantified value set to 4; Quinary complaints: Extremely serious problems, such as systematic errors and group impacts, with a quantified value set to 5.

[0049] Specifically, the quantified values of the three levels of the urgency of the work order are as follows: Low urgency: Can be resolved within a relatively long time (such as more than 72 hours), with a quantified value set to 1; Medium urgency: Needs to be resolved within a certain time (such as 24 - 72 hours), with a quantified value set to 2; High urgency: Must be resolved within a relatively short time (such as within 24 hours), with a quantified value set to 3.

[0050] Specifically, the quantified values of the three levels of the problem complexity are as follows: Simple problems: Problems in a single business process with a clear processing flow, with a quantified value set to 1; Medium problems: Problems involving multiple business processes that require a certain degree of coordination and professional knowledge, with a quantified value set to 2; Complex problems: Cross-departmental and cross-system problems that require advanced professional skills and a large amount of coordination work, with a quantified value set to 3.

[0051] In this embodiment, the comprehensive allocation score of the agent seat is calculated from the quantified values of the user demand complexity, the agent seat skill matching degree, and the current load of the agent seat. The specific method is to use the weighted average, and the formula is as follows: Comprehensive allocation score = user demand complexity × user demand complexity weight + agent seat skill matching degree × agent seat skill matching degree weight + (4 - current load quantified value of the agent seat) × current load weight of the agent seat; Among them, the agent seat skill matching degree is based on whether the agent seat has the ability to handle problems of this type of user. If it has, it is 1; if not, it is 0. The current load quantified value of the agent seat is divided into three levels: 1, 2, and 3 according to the number of work orders being processed by the agent seat.

[0052] Specifically, in the comprehensive allocation score, (4 - current load quantified value of the agent seat) is selected. The lower the corresponding load, the more suitable it is to allocate a new work order, making the score positively correlated with the priority of allocation.

[0053] Specifically, the current load of the agent seat can be measured by the number of work orders being processed by the agent seat. The following is a preferred option: Low load: The number of work orders being processed is less than 30% of the total carrying capacity, and the quantization value is set to 1; Medium load: The number of work orders being processed is between 30% and 70% of the total carrying capacity, and the quantization value is set to 2; High load: The number of work orders being processed is greater than 70% of the total carrying capacity, and the quantization value is set to 3.

[0054] Embodiment 2 Based on Embodiment 1, in this embodiment, as Figure 5 , the present invention also discloses a 12345 hotline incoming call service insight system, including: an incoming and outgoing call management module, a user portrait module, an intelligent work order management module, a seat skill group intelligent allocation module, and a call data statistics and feedback module; The incoming and outgoing call management module, the user portrait module, the intelligent work order management module, the seat skill group intelligent allocation module, and the call data statistics and feedback module are connected in sequence, and the call data statistics and feedback module is respectively connected to the incoming and outgoing call management module and the user portrait module; The incoming and outgoing call management module is used for number identification and selecting a line according to user classification; the user portrait module is used for updating and constructing a user portrait; the intelligent work order management module is used for intelligently processing work orders with problems to be solved; the seat skill group intelligent allocation module is used for intelligently allocating seats to users; the call data statistics and feedback module is used for data statistics, generating reports, and providing data to the user portrait module for updating.

[0055] Specifically, the incoming and outgoing call management module includes number identification, line selection, and voice interaction functions; it uses a regular expression matching algorithm to identify the origin of the incoming call number, combines real-time line status monitoring (such as busy line rate, response delay), and selects the optimal access line through a dynamic programming algorithm (such as preferentially allocating local lines according to the city of origin); at the same time, it integrates a TTS / ASR engine to achieve real-time interaction between the voice robot and the user; and communicates with the seat terminal through the WebSocket protocol to support call transfer and information push.

[0056] Specifically, the user portrait module includes data collection, label generation, and portrait storage functions; it captures historical call records (stored in a MySQL database), work order texts (stored in Elasticsearch), and user interaction behavior logs (such as key operations, voice emotion values) in real time to obtain data, integrates the user call frequency, work order processing results, etc. through a decision tree algorithm, outputs "user type labels", and finally generates a user portrait; Specifically, the intelligent work order management module includes work order synchronization, intelligent handling, and exception handling functions. It pulls the latest work order status from the work order system through a scheduled task (Cron expression) and realizes asynchronous update based on the Apache Kafka message queue. For "users with problems to be solved", it calls the NLP model to parse the work order text, extracts key information (such as the processing department and the estimated completion time), and generates structured broadcast content. If the work order system interface times out, it triggers an alternative solution (such as manual synchronization after manual confirmation).

[0057] Specifically, the intelligent allocation module for agent skill groups includes a skill map, dynamic allocation, and terminal push functions. Based on the agent's historical service data (such as work order resolution rate and user rating), it calculates the proficiency of each agent in different business types through the Analytic Hierarchy Process (AHP) (such as the complaint handling ability value of 90 / 100). It uses the Hungarian algorithm to match the user's needs with the agent's skills, and combines the agent's current call volume (Redis real-time counting) to select the optimal agent with "skill matching degree ≥ 85% and call volume ≤ 3".

[0058] Specifically, such as Figure 6 , the call data statistics and feedback module includes data collection, analysis model, and policy synchronization functions. It embeds Prometheus monitoring metrics in the call link to collect data such as connection rate and average processing time in real time. It uses the XGBoost algorithm to predict the connection rate trend of each line, dynamically adjusts the line priority, and at the same time mines user behavior association rules through the Apriori algorithm (such as transferring "high-frequency complaint users + night calls" to senior agents). It generates an optimization strategy report every day at midnight and automatically synchronizes it to the incoming call management module (such as adjusting the night line allocation) and the user portrait module (such as updating the label weights).

[0059] Specifically, data transfer between modules is achieved through RESTful API, Kafka message queue, and Redis cache.

[0060] Specifically, the technical effects are compared as shown in the following table: The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements can still be made, and these improvements should also be regarded as the protection scope of the present invention.

Claims

1. A method for business insight of 12345 hotline incoming calls, characterized by: include: S1): After the call is received, the incoming and outgoing call management modules are called to identify the number and select the line according to the user classification; S2): If it is the first call, record the basic information and generate a user ID, enter the voice navigation, and jump to S6 after the processing is completed); If it is not the first call, the user portrait module is called to generate a smart tag based on the user portrait data, and the current user type is quickly matched according to the existing user type, and it is determined whether it is a work order with a problem to be solved; The user portrait module generates smart tags in combination with user portrait data, including: S201): Analyze the user's historical conversation text and real-time conversation text based on the large language model, extract key semantic features, and generate sentiment tags and business tags; S202): Combine user portrait data and generate smart tags through the rule engine; S3): If it is a work order with a problem to be solved, the intelligent work order management module is called for intelligent processing; The intelligent work order management module in S3) performs intelligent processing specifically by automatically obtaining the real-time progress of the work order to be solved from the database and reporting it to the user; S4): If it is not a work order with a problem to be solved, determine whether it is a car moving work order user; S5): If the user is a car moving work order user, the previous car moving record will be automatically queried, and the user will be asked whether to make an additional call; if the user is not a car moving work order user, the seat skill group intelligent allocation module will be switched to perform seat intelligent allocation, and a manual seat will be selected for processing; The intelligent seat allocation in S5) includes: S501): Calculate the complexity of user requirements; S502): Calculate the comprehensive seat allocation score; S503): Select the agent with the highest comprehensive allocation score and allocate the incoming call to the agent; S6): Call the call data statistics feedback module to perform data statistics, generate reports, and call the user portrait module to update the user portrait.

2. The method for obtaining business insight into incoming calls on the 12345 hotline according to claim 1, characterized in that: The updating of the user portrait in S6) includes: S601): Data collection, obtaining relevant data from different data sources, including but not limited to historical call records, work order information, and interactive behavior data of the call data statistics feedback module; S602): Data cleaning: cleaning the collected data, including but not limited to removing duplicate data, processing missing values ​​and correcting erroneous data; S603): Data association, associating data from different data sources through a user's unique identifier (such as a mobile phone number, user ID); S604): Feature extraction and transformation, including but not limited to computing behavioral statistics, text mining, and data standardization; S605): Data fusion, fusing the extracted and converted features into a unified user portrait model to form a multi-dimensional user portrait.

3. The method for obtaining business insight into the 12345 hotline caller service according to claim 2, characterized in that: S605) The multi-dimensional user portrait includes a basic information dimension, a behavior preference dimension, and a business demand tendency dimension; The basic information dimensions include demographic information and contact information; The behavior preference dimensions include call behavior, interaction behavior and work order behavior; The business demand tendency dimension includes subject clustering of work order texts and call records to identify business areas and problem types that users are concerned about, and sentiment analysis of texts and voices to derive users' satisfaction and expectations for different businesses.

4. The method for obtaining business insight into incoming calls from the 12345 hotline according to claim 1, characterized in that: The S201) includes: S20101): Perform text preprocessing based on historical conversation text, user portrait data and real-time interaction data; S20102): Segment and vectorize the text after text preprocessing; S20103): Use the semantic tags generated by the large language model and map them with the user portrait tag system to extract key semantic features, thereby generating emotional tags and business tags.

5. The method for obtaining business insight into the 12345 hotline caller service according to claim 1, characterized in that: The S202) is specifically to realize the mapping alignment of the emotion tags and business tags with the user portrait data labels through a fast matching algorithm, map the emotion tags and business tags to the user portrait data label categories, and generate smart tags.

6. The method for obtaining business insight into incoming calls from the 12345 hotline according to claim 1, characterized in that: The user demand complexity in S501) is calculated by comprehensively calculating the complexity of the complaint level, the urgency of the work order and the complexity of the problem. The specific method is weighted average, and the formula is as follows: User demand complexity = complaint level quantified value × complaint level weight + work order urgency quantified value × work order urgency weight + problem complexity quantified value × problem complexity weight; Among them, the quantitative value of complaint level is divided into five levels: 1, 2, 3, 4, and 5 according to the complaint level; the quantitative value of work order urgency is divided into three levels: 1, 2, and 3 according to the time requirement for resolving the work order; the quantitative value of problem complexity is evaluated based on factors such as the business scope and technical difficulty involved in the problem, and is divided into three levels: 1, 2, and 3.

7. The method for obtaining business insight into incoming calls on the 12345 hotline according to claim 5, characterized in that: The comprehensive seat allocation score in S502) is calculated based on the complexity of user requirements, the matching degree of seat skills and the quantitative value of the current load of the seat. The specific method is weighted average, and the formula is as follows: Comprehensive allocation score = user demand complexity × user demand complexity weight + agent skill matching × agent skill matching weight + (4 - agent current load quantification value) × agent current load weight; Among them, the agent skill matching degree is based on whether the agent is capable of handling the problems of this type of user, which is 1 if the agent is capable of handling the problems of this type of user, and 0 if the agent is not capable of handling the problems. The agent's current load quantification value is divided into three levels: 1, 2, and 3 according to the number of work orders the agent is handling.

8. A system for business insight of 12345 hotline incoming calls, used to implement the method for business insight of 12345 hotline incoming calls described in claims 1 to 7, characterized in that: include: Inbound and outbound call management module, user portrait module, intelligent work order management module, agent skill group intelligent allocation module and call data statistics feedback module; The incoming and outgoing call management module, the user portrait module, the intelligent work order management module, the seat skill group intelligent allocation module and the call data statistics feedback module are connected in sequence, and the call data statistics feedback module is connected to the incoming and outgoing call management module and the user portrait module respectively; The incoming and outgoing call management module is used to identify numbers and select lines according to user classification; The user portrait module is used to update and construct user portraits; The intelligent work order management module is used to intelligently handle work orders with problems to be resolved; the seat skill group intelligent allocation module is used to intelligently allocate seats to users; the call data statistics feedback module is used for data statistics, report generation and providing data to the user portrait module for updating.

Images