Intelligent platform system and construction method based on AI model

Abstract

The present invention belongs to the field of artificial intelligence technology, and in particular to an intelligent platform system and a construction method based on an AI model. The platform includes an interactive sub-platform, a simulation training subsystem and a practical training simulation feedback subsystem; the platform constructs a dynamic customer service portrait through a multimodal algorithm, optimizes the training effect by using scenario simulation and micro-expression analysis, intelligently recommends personalized courses and evaluates the training effectiveness in real time; the practical training simulation feedback subsystem simulates the real response scenario, combines preprocessing, scheduling, correction and evaluation modules, and effectively improves the quality and efficiency of customer service responses; the interactive sub-platform promotes seamless connection between users and systems, and realizes closed-loop management of training, practical training and feedback; the present invention provides an intelligent solution for customer service training and actual combat, and significantly improves service quality and efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to an intelligent platform system and a construction method based on an AI model. Background Art

[0002] In the customer service industry, especially in large call centers like customer service centers, there are widespread problems of high staff turnover and frequent business updates. Traditional training methods usually rely on dedicated training teams and fixed training materials, such as PPT and Word documents, and use standardized test questions and templates for assessment. This method not only consumes a lot of human resources, but also often leads to poor training results due to the lack of targeted and personalized training content. In addition, traditional training methods also have problems such as untimely feedback and difficulty in quantifying training results.

[0003] For example, the patent with publication number CN117745488A discloses a customer service training method and equipment, including: obtaining a customer service training request; determining case information corresponding to the customer service training request; generating a virtual order corresponding to the customer service training request based on the case information; using a preset large model to process the virtual order and case information, and generating a customer service training task for implementing the customer service training operation, wherein the customer service training task includes the session information of the virtual user.

[0004] For example, a patent with authorization announcement number CN108734379B discloses a differentiated online training method for customer service personnel, including: S1: conduct quality inspection on customer service personnel and extract problems in quality inspection; S2: count the specific categories of each problem generated in the quality inspection of each customer service personnel; S3: mark the customer service personnel according to the different categories of the quality inspection problems generated by the customer service personnel; S4: summarize the quality inspection problems generated by the customer service personnel into the training system; S5: the training system transmits the updated quality inspection problems to the database, and then obtains the solutions corresponding to each updated quality inspection problem from the database, and stores the updated quality inspection problems and solutions in the training system in a one-to-one correspondence; S6: the customer service personnel completes the online training through the training system.

[0005] The above-mentioned prior art has the following problems: the prior art is still insufficient in how to perform accurate response scheduling and targeted training evaluation and feedback training during the training process. In order to solve the above problems, the present invention provides an intelligent platform system and construction method based on AI model construction. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes an intelligent platform system and construction method based on AI model, the platform includes an interactive sub-platform, a simulation training subsystem and a practical training simulation feedback subsystem; the platform constructs a dynamic customer service portrait through a multimodal algorithm, optimizes the training effect by scenario simulation and micro-expression analysis, intelligently recommends personalized courses and evaluates the training effectiveness in real time; the practical training simulation feedback subsystem simulates the real response scenario, combines the preprocessing, scheduling, correction and evaluation modules, and effectively improves the quality and efficiency of customer service response; the interactive sub-platform promotes seamless connection between users and systems, and realizes closed-loop management of training, practical training and feedback; the present invention provides an intelligent solution for customer service training and actual combat, significantly improving service quality and efficiency.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The intelligent platform system built based on the AI ​​model includes: a simulation training subsystem and a practical training simulation feedback subsystem; the simulation training subsystem includes a scenario simulation module, an intelligent course library module, a recommendation engine module, and a training evaluation module;

[0009] The recommendation engine module recommends matching training courses to each customer service representative through the recommendation algorithm based on the secondary dynamic customer service portrait library obtained by the scenario simulation module and the customer service training database obtained by the intelligent course library module. The training evaluation module uses the built-in AI-agent1 to conduct real-time evaluation of the training results of the customer service representative based on the recommended training courses, obtain the corresponding customer service training scores, and output the areas in which the customer service representative is good and not good at, as well as the degree of proficiency in the corresponding areas. The areas in which the customer service representative is good and not good at, as well as the degree of proficiency in the corresponding areas, are fed back to the corresponding customer service node in the secondary dynamic customer service portrait library for storage.

[0010] The training simulation feedback subsystem includes a training simulation module, an answer preprocessing module, and an answer scheduling module; the training simulation module is used to simulate incoming call voice information in different answering scenarios, emotional states, and fields based on the collected historical call information through AI-agent2 with a built-in role-playing model; the answer scheduling module builds an answer buffer pool through the secondary dynamic customer service portrait library and the secondary evaluation scores obtained by the evaluation sharing module, and according to the constructed answer buffer pool and the fields and problems in the call information obtained by the answer preprocessing module, the customer service in the answer buffer pool is scheduled through the enhanced scheduling model to answer the call information in the corresponding field.

[0011] Specifically, the training evaluation module outputs the specific steps of the corresponding customer service's strengths and weaknesses in the response areas, including:

[0012] A1. Obtain the training process and result data of M field training courses recommended by N customer service staff according to the recommendation engine module, and pre-process the acquired data; the training process and result data include: the completion rate of each field course, the correct rate of answering questions in each field, the corresponding rate of answering questions in each field, and the micro-expression state score of the customer service staff answering questions in each field in the scenario simulation;

[0013] A2. Based on the preprocessed training process and result data, Q response evaluation indicators corresponding to each field are constructed. According to the constructed response evaluation indicators, the customer service training evaluation set is obtained. ;

[0014] A3. Build an evaluation model based on the fuzzy evaluation algorithm and embed the built model into AI-agent1. Input this into AI-agent1 to obtain the customer service training score of the nth customer service representative in the mth field. ;

[0015] A4. Set the three-level assessment threshold and and , set the fourth-level proficiency level assessment interval according to the third-level assessment threshold, specifically: When When When , it means that the nth customer service is more proficient in the mth field. When , it means that the nth customer service representative is not good at the mth field;

[0016] A5: The four-level proficiency evaluation interval set in A4 is built into AI-agent1. Based on the customer service training score obtained by real-time evaluation, AI-agent1 outputs the proficiency of the nth customer service in the mth field, and obtains the corresponding proficient and non-proficient response fields of the nth customer service based on the proficiency and customer service training score. When the corresponding customer service training scores in M ​​fields meet the requirements, If the conditions are not met, the corresponding customer service staff will not be allowed to enter the practical training stage.

[0017] Specifically, the specific steps of role-playing model construction include:

[0018] B1. Obtaining the incoming call text information by using the voice-to-text extraction algorithm based on the historical incoming call voice information;

[0019] B2. Automatically label the acquired incoming call text information with domain labels according to the association rules, and construct an incoming call text library using the labeled incoming call text information;

[0020] B3, input the acquired historical incoming call voice information into the pre-trained voice extraction sub-model, acquire different types of basic voice features in the corresponding incoming call voice information, and construct an incoming call voice basic feature library using the acquired basic voice features;

[0021] B4, input the acquired historical incoming call voice information into the pre-trained emotion extraction sub-model, obtain different types of voice emotion features and emotion types in the corresponding incoming call voice information, and use the acquired voice emotion features and emotion types to build an incoming call emotion feature library;

[0022] B5. Use a random algorithm to build and pre-train a random generation sub-model, and build a built-in answering difficulty level into the random generation sub-model. Input the features in the incoming call information text library, incoming call voice basic feature library, and incoming call emotional feature library into the random generation sub-model to simulate the generation of incoming call voice information at different answering difficulty levels.

[0023] Specifically, the specific steps of the response scheduling module to build the response buffer pool include:

[0024] C1. Build M domain response buffer pools based on the customer service response domains stored in the secondary dynamic customer service portrait library, and store the corresponding response domain customer service training scores that meet The customer service information of the conditions is stored in the constructed corresponding field response buffer pool;

[0025] C2. Sort the customer service staff in the mth answer buffer pool by the proficiency level evaluation interval, and update the sorting of all answer buffer pools by the obtained secondary evaluation scores, and perform scheduling and answer evaluation settings for all answer buffer pools after the sorting.

[0026] Specifically, the training simulation feedback subsystem also includes a training evaluation module, which uses the built-in AI-agent3 to conduct real-time evaluation of the response process of the dispatched customer service, obtain the training evaluation score, and conduct secondary training judgment on the customer service according to the training evaluation score;

[0027] The specific steps for scheduling and setting response evaluation for all response buffer pools after hierarchical sorting include:

[0028] The number of response buffer pools of the nth customer service is determined according to the acquired answering field corresponding to the nth customer service. If the nth customer service has at least one response buffer pool, the specific settings include:

[0029] C211. Set the initial answer score and dispatch frequency of the nth customer service in the mth answer buffer pool to 0. When the nth customer service is not selected for dispatch, its current status in all corresponding answer buffer pools is "1", that is, the nth customer service is currently in an idle state;

[0030] C212. When the nth customer service is extracted from an existing response buffer pool for scheduling, its current status in all corresponding existing response buffer pools is represented by "0", that is, the nth customer service is in the scheduling state. When the scheduling response is completed, its current status in all corresponding existing response buffer pools is converted from "0" to "1", and the scheduling frequency and initial response score of the nth customer service in the corresponding response buffer pool are each increased by 1.

[0031] Specifically, the specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting also include:

[0032] C213, use AI-agent3 to evaluate the dispatch response process in C212 in real time and obtain the training evaluation score of the corresponding response process ;

[0033] C214, according to the obtained ,like , then the initial response score of the nth customer service in the mth response buffer pool remains unchanged. , then the initial response score of the nth customer service in the mth response buffer pool is increased by 1. , then the initial response score of the nth customer service in the mth response buffer pool is increased by 2, and when the nth customer service completes the response and obtains a "good review", the initial response score is increased by 1;

[0034] C215. Add the dispatch bonus points, response evaluation bonus points, and favorable comment bonus points obtained by the nth customer service in the process of C212-C214 to the customer service training score corresponding to the mth field, and obtain the customer service comprehensive score of the nth customer service in the mth field;

[0035] C216. Based on the obtained customer service comprehensive score and the four-level expertise level evaluation range, obtain the expertise level of the nth customer service in the mth field at the current moment, and perform a second update on the hierarchical ranking of the nth customer service in the mth response buffer pool based on the expertise level at the current moment.

[0036] Specifically, the specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting also include:

[0037] C218, if , then the initial answer score of the nth customer service in the mth answer buffer pool is reduced by 1, and when the nth customer service completes the answer and gets a "bad review", the initial answer score is directly reduced by 1;

[0038] C219, repeat the process of C215 to add the answer score obtained in C218 to the customer service training score corresponding to the m-th field, obtain the customer service comprehensive score of the n-th customer service in the m-th field, and repeat the process of C216 according to the obtained customer service comprehensive score to perform a second update on the hierarchical sorting;

[0039] C220, when the customer service comprehensive score corresponding to the proficiency level of the nth customer service in the mth response buffer pool is less than When the nth customer service is removed from the mth answer buffer pool, AI-agent3 monitors the answer problems obtained during the process of the nth customer service being extracted from the mth answer buffer pool, and feeds them back to the intelligent course library module to generate customer service training materials in the corresponding field, and conducts secondary training in the corresponding field. When the nth customer service's secondary customer service training score in the mth field meets the requirements, When the condition is met, the nth customer service will be automatically added to the mth response buffer pool;

[0040] C221, when the comprehensive scores of the customer service corresponding to the expertise level of the nth customer service in all the response buffer pools are less than When the nth customer service representative is directly excluded from the training process, the answer questions obtained by AI-agent3 monitoring are fed back to the intelligent course library module to generate customer service training materials in the corresponding field for retraining.

[0041] Specifically, the specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting also include:

[0042] C222. When the nth customer service is currently selected from the p answer buffer pools, the corresponding proficiency level of the nth customer service in the p answer buffer pools is obtained and compared, the answer buffer pool corresponding to the maximum proficiency level of the nth customer service is obtained, and the call information corresponding to the field of the nth customer service answer is selected from the answer buffer pool;

[0043] C223. When two or more of the nth customer service have the same expertise level and the highest expertise level in the p answer buffer pools, the customer service comprehensive scores of the nth customer service in the answer buffer pools with the same expertise level are compared, and the nth customer service is scheduled to answer the incoming call information in the corresponding field from the answer buffer pool with the highest customer service comprehensive score, and the current state of the nth customer service in the remaining answer buffer pool is changed to "0";

[0044] C224. Repeat C213-C221 to evaluate the response process of C222 and C223, and conduct secondary training based on the evaluation results.

[0045] Specifically, the training simulation feedback subsystem also includes a training correction module, which is used to monitor the response process of the customer service dispatched by the response scheduling module in real time through the AI-agent3 with a built-in semantic parsing model, and feed back the monitoring results to the corresponding customer service in real time through the AI-agent3 to assist in correcting the customer service response process;

[0046] The specific steps of strengthening the scheduling model construction include:

[0047] D1. Based on the obtained answer buffer pool and incoming call voice information, build the current input state of the enhanced scheduling model ,in, Indicates the current status of the nth customer service in the mth response buffer pool, that is, whether it is idle or in scheduling state; Indicates the hierarchical sorting status of the nth customer service in the mth response buffer pool at the current moment; Indicates the comprehensive score of the nth customer service in the mth response buffer pool at the current moment. Indicates the scheduling frequency of the nth customer service unit time in the mth response buffer pool at the current moment. represents the probability that the qth incoming call at the current moment belongs to the mth domain, Indicates the waiting time for the qth incoming call belonging to the mth area at the current moment, represents the probability that the qth incoming call in the mth domain is dispatched to the nth customer service in the mth answer buffer pool at the current moment, It indicates the probability that the nth customer service representative in the mth answer buffer pool to which the qth incoming call is dispatched at the current moment is the customer service representative with the highest comprehensive score in the mth answer buffer pool;

[0048] D2. Set the execution action trigger information according to the current input state, including:

[0049] D21. When the nth customer service is in the dispatching state, the first execution information is triggered, that is, the action of executing the corresponding customer service state conversion is triggered, and the action of changing the nth customer service state from "1" to "0" is executed; otherwise, the action of changing the nth customer service state from "0" to "1" is executed;

[0050] D22. When there are u customer services in the mth answer buffer pool that are currently idle, and the mth answer buffer pool currently corresponds to v incoming call schedulings in the mth field, the second execution information is triggered, that is, the action of scheduling the customer service with the largest customer service comprehensive score among the u customer services in the mth answer buffer pool that are idle to the incoming call with the longest waiting time is executed, and the above actions are repeated until all incoming calls and customer services are scheduled;

[0051] D23, when the remaining idle customer service in D22 cannot meet the call scheduling demand of the corresponding answer buffer pool, the third execution information is triggered, that is, the remaining call in the mth answer buffer pool is dispatched to the customer service with the largest comprehensive score in the remaining answer buffer pool for answering, and the answering process is assisted by AI-agent3, and the questions that cannot be assisted are recorded and fed back to the idle customer service in the corresponding answer buffer pool for a second call back reply;

[0052] D3, based on the D2 action trigger information, build the current execution action ,in, Indicates the action that triggers the execution of the first execution information. Indicates the action that triggers the execution of the corresponding second execution information. Indicates the action that triggers the execution of the corresponding third execution information.

[0053] Specifically, the specific steps of strengthening the scheduling model construction also include:

[0054] D4. Construct a reward function based on the corresponding execution action of D3 , Indicates the g-th execution of an action The corresponding reward is when the nth customer service status transition is correct for the gth execution. ,on the contrary, ; G represents the total number of times the nth customer service state changes; Indicates execution of an action The corresponding reward is the reward for dispatching all v incoming calls from the mth field to u idle customer service representatives in the mth answer buffer pool. It represents the reward for dispatching g calls out of v calls in the mth domain to u idle customer service representatives in the mth answer buffer pool. It represents the reward for dispatching h calls out of v calls in the mth domain to h idle customer service representatives in the non-mth answer buffer pool; , and Represents the gain coefficient of the corresponding reward function;

[0055] D5. Build a reinforcement scheduling model based on the reinforcement learning algorithm, and input the constructed current input state, execution action trigger information, current execution action and reward function into the reinforcement scheduling model for training. The trained reinforcement scheduling model is built into the answering scheduling module for scheduling between incoming calls and customer service.

[0056] Specifically, the built-in answer difficulty levels include easy, normal, difficult and hard levels; the easy level scoring interval is Corresponding to; general grade scoring interval and Corresponding to; more difficult level scoring interval and Corresponding to; Difficulty level score interval and Corresponding; the M domain response buffer pools correspond one-to-one to the M domains.

[0057] The method for building an intelligent platform system based on an AI model includes:

[0058] S1. Obtain customer service attribute data, historical training, work process data, micro-expression score data and training material data to build a secondary dynamic customer service portrait library and customer service training material library;

[0059] S2. A recommendation engine is built between the constructed secondary dynamic customer service portrait library and the customer service training material library, the built-in recommendation engine is connected to the interactive sub-platform, and AI-agent 1 and a sharing sub-platform are built in the interactive platform to construct a simulation training subsystem;

[0060] S3. Incoming call information is simulated through AI-agent2 with a built-in role-playing model. At the same time, a response buffer pool is constructed through the secondary evaluation scores obtained through the secondary dynamic customer service portrait library and the evaluation sharing module. The reinforcement scheduling model and AI-agent3 are built into the constructed response buffer interface to obtain a training simulation feedback subsystem.

[0061] A computer-readable storage medium stores computer instructions, which, when executed, execute a method for building an intelligent platform system based on an AI model.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. The present invention combines the secondary dynamic customer service portrait library and the customer service training database through the recommendation engine module, and can intelligently recommend customized training courses based on the individual ability characteristics and knowledge gaps of customer service personnel, thereby significantly improving the pertinence and effectiveness of training; secondly, the training evaluation module performs real-time training effect evaluation through the built-in AI-agent1, which can not only timely feedback the customer service's proficiency in various response fields, but also dynamically update the customer service portrait to ensure that subsequent recommendations are more accurate; in addition, during the training process, the present invention uses the response scheduling module to construct a response buffer pool based on the customer service's ability evaluation score, and uses the enhanced scheduling model to achieve efficient and accurate scheduling decisions, ensuring that each customer service can be assigned to the response task that best suits their ability level. At the same time, the training evaluation module further improves the pertinence of the training evaluation, ensuring that the training suggestions fed back to the customer service are more in line with actual needs, thereby effectively making up for the shortcomings of the existing technology in response scheduling and training evaluation, and comprehensively improving the customer service quality and response efficiency.

[0064] 2. The present invention can simulate various answering scenarios according to real call information through the role-playing model in the training simulation feedback subsystem, thereby enhancing the customer service's ability to deal with complex situations. The answer scheduling module builds an answer buffer pool and optimizes the scheduling strategy using a reinforcement learning algorithm to ensure that each scheduling can match the most suitable candidate to handle calls in a specific field, thereby improving the service quality and efficiency of the customer service. The training evaluation module and the training correction module monitor the customer service answering process in real time and provide timely feedback and auxiliary corrections, which helps the customer service to quickly improve their shortcomings. This closed-loop training and evaluation mechanism greatly improves the professional quality and service level of the customer service team, while also reducing the error rate in the customer service process and significantly improving customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a module architecture diagram of an intelligent platform system constructed based on an AI model in Example 1 of the present invention;

[0066] Figure 2 This is a diagram of the role-playing model architecture in Example 1 of the present invention;

[0067] Figure 3 This is a flow chart of the method for building an intelligent platform system based on an AI model in Example 2 of the present invention. DETAILED DESCRIPTION

[0068] Example 1

[0069] See also Figure 1 , an embodiment provided by the present invention: an intelligent platform system constructed based on an AI model, comprising: an interactive sub-platform, a simulation training sub-system, a practical training simulation feedback sub-system and an embedded device;

[0070] The simulation training subsystem is used for intelligent simulation training and evaluation of customer service; the simulation training subsystem includes customer service portrait module, scenario simulation module, intelligent course library module, recommendation engine module, training evaluation module and evaluation sharing module;

[0071] The customer service portrait module is used to obtain customer attribute data, historical training or work process data, and build a dynamic customer service portrait library through a multimodal algorithm; further, the multimodal algorithm in this embodiment includes a Bert-based text algorithm, a speech feature extraction algorithm, and a knowledge graph construction algorithm;

[0072] The scenario simulation module is used to simulate the answering scenario using the AI-agent built into the simulation training subsystem, and collect the customer service micro-expression data during the simulated answering process, and perform intelligent detection and evaluation through the micro-expression algorithm based on the acquired micro-expression data to obtain the micro-expression state score, and feed the acquired micro-expression state score back to the dynamic customer service portrait library for secondary portrait, and obtain the secondary dynamic customer service portrait library; further, the micro-expression algorithm in this embodiment includes optical flow method, geometric feature method and facial action coding system, etc.;

[0073] The intelligent course library module builds a customer service training library through the acquired training materials and natural language algorithms, and embeds a deep reasoning model in the interface of the built customer service training library; further, the deep reasoning model in this embodiment includes a rule-based reasoning model, GPT, and a pre-trained Bert model;

[0074] The recommendation engine module recommends matching training courses to each customer service representative through a built-in recommendation algorithm based on the acquired secondary dynamic customer service portrait library and customer service training database; further, the recommendation algorithm in this embodiment includes a collaborative filtering algorithm and a knowledge graph recommendation algorithm;

[0075] The training evaluation module is used to evaluate the results of customer service training according to the recommended training courses in real time through the built-in AI-agent1, obtain the corresponding customer service training score, and output the response areas that the customer service is good at and not good at and the degree of proficiency in the corresponding areas, and feed back the response areas that the customer service is good at and not good at and the corresponding degree of proficiency to the corresponding customer service node in the secondary dynamic customer service portrait library for storage.

[0076] Furthermore, in this embodiment, the specific steps of the training evaluation module outputting the corresponding customer service's expertise and weaknesses in response areas include:

[0077] A1. Obtain the training process and result data of the M field training courses recommended by N customer service personnel according to the recommendation engine module, and pre-process the obtained data; the training process and result data include: the completion rate of each field course, the correct rate of answering questions in each field, the corresponding rate of answering questions in each field, and the micro-expression state score of the customer service personnel answering questions in each field in the scenario simulation; further, the corresponding fields in this embodiment are specifically set by those skilled in the art in the specific practice process, such as life field, work field, technical field, etc.;

[0078] A2. Based on the preprocessed training process and result data, Q response evaluation indicators corresponding to each field are constructed. According to the constructed response evaluation indicators, the customer service training evaluation set is obtained. ; Further, the Q response evaluation indicators corresponding to each field are set by those skilled in the art according to the specific determined fields;

[0079] A3. Build an evaluation model based on the fuzzy evaluation algorithm and embed the built model into AI-agent1. Input this into AI-agent1 to obtain the customer service training score of the nth customer service representative in the mth field. ;

[0080] A4. Set the three-level assessment threshold and and , set the fourth-level proficiency level assessment interval according to the third-level assessment threshold, specifically: When When When , it means that the nth customer service is more proficient in the mth field. When , it means that the nth customer service representative is not good at the mth field;

[0081] A5: The four-level proficiency evaluation interval set in A4 is built into AI-agent1. Based on the customer service training score obtained by real-time evaluation, AI-agent1 outputs the proficiency of the nth customer service in the mth field, and obtains the corresponding proficient and non-proficient response fields of the nth customer service based on the proficiency and customer service training score. When the corresponding customer service training scores in M ​​fields meet the requirements, If the conditions are not met, the corresponding customer service staff will not be allowed to enter the practical training stage.

[0082] The evaluation sharing module shares the problems of the corresponding customer service obtained by the training evaluation module through the shared sub-platform built into the simulation training subsystem, and allows the customer service who is good at the problem to reply to the shared problem with a solution strategy through the sharing sub-platform, and feeds back the reply strategy to the corresponding customer service. At the same time, the customer service who answers the solution strategy is rewarded with points through AI-agent1, and the reward points are added to the training score of the corresponding customer service to obtain a secondary evaluation score; further, in the evaluation sharing module of this embodiment, one solution strategy reply will add 1 point to the customer service training score of the customer service who makes the reply.

[0083] The training simulation feedback subsystem is used for customer service response process simulation, training process evaluation and evaluation result feedback; the training simulation feedback subsystem includes training simulation module, response preprocessing module, response scheduling module, training evaluation module, training correction module and evaluation feedback module;

[0084] The training simulation module is used to simulate the incoming call voice information of different answering scenarios, emotional states and fields through the AI-agent2 with built-in role-playing model based on the collected historical incoming call voice information; for further information, please refer to Figure 2 , the specific steps of constructing the role-playing model in this embodiment include:

[0085] B1. Acquire the incoming call text information by using a voice-to-text extraction algorithm based on the acquired historical incoming call voice information; further, the voice-to-text extraction algorithm in this embodiment includes DeepSpeech;

[0086] B2. Automatically label the acquired incoming call text information with domain labels according to the association rules, and construct an incoming call text library using the labeled incoming call text information;

[0087] B3. Input the acquired historical incoming call voice information into the pre-trained voice extraction sub-model, acquire different types of basic voice features in the corresponding incoming call voice information, and construct an incoming call voice basic feature library using the acquired basic voice features; further, the voice extraction sub-model in this embodiment includes Librosa, OpenSMILE, and Wav2Vec 2.0, etc.;

[0088] B4. Input the acquired historical incoming call voice information into the pre-trained emotion extraction sub-model, acquire different types of voice emotion features and emotion types in the corresponding incoming call voice information, and construct an incoming call emotion feature library using the acquired voice emotion features and emotion types; in this embodiment, the emotion extraction sub-model includes SEResNet / SEResNeXt, attention network, etc.;

[0089] B5. Use a random algorithm to build and pre-train a random generation sub-model, and build a built-in answering difficulty level into the random generation sub-model. Input the features in the incoming call information text library, incoming call voice basic feature library, and incoming call emotional feature library into the random generation sub-model to simulate the generation of incoming call voice information at different answering difficulty levels.

[0090] Furthermore, in this embodiment, the built-in answer difficulty levels include easy, general, difficult and difficult levels; the easy level scoring interval is Corresponding to; general grade rating interval and Corresponding to; more difficult level scoring interval and Corresponding to; Difficulty level score interval and correspond.

[0091] This process builds a caller information text library, a caller voice basic feature library, and a caller emotional feature library. The training simulation module can generate a variety of simulated calls for different fields, emotional states, and answering difficulties. This not only enhances the customer service staff's ability to handle complex situations, but also helps them accumulate valuable experience. In addition, by setting simulation training at different difficulty levels, the customer service staff's skill level can be gradually improved to ensure that they can calmly deal with various challenges that may arise in actual work. This data-driven training method not only reduces the cost of traditional training, but also significantly improves the training effect and the overall quality of the customer service team.

[0092] The answer pre-processing module uses the AI-agent2 built into the training simulation feedback subsystem to pre-reply to incoming call information and obtain the corresponding fields and questions of the incoming call information, thereby reducing the occupancy rate of customer service for abnormal calls;

[0093] The answer scheduling module builds an answer buffer pool through the secondary dynamic customer service portrait library and the secondary evaluation score obtained by the evaluation sharing module, and schedules the customer service in the answer buffer pool through the enhanced scheduling model according to the constructed buffer pool and the fields and problems in the call information obtained by the answer preprocessing module, answers the corresponding field call information, and improves the efficiency of customer service; further, the specific steps of the answer scheduling module in this embodiment to build the answer buffer pool include:

[0094] C1. Build M domain response buffer pools based on the customer service response domains stored in the secondary dynamic customer service portrait library, and store the corresponding response domain customer service training scores that meet The customer service information of the conditions is stored in the constructed corresponding domain response buffer pool; further, in this embodiment, the M domain response buffer pools correspond to the M domains one by one;

[0095] C2. Sort the customer services in the mth answer buffer pool by the proficiency level evaluation interval, and use the obtained secondary evaluation score to update the hierarchical sorting of all the answer buffer pools after the hierarchical sorting, and perform scheduling and answer evaluation settings for all the answer buffer pools after the hierarchical sorting. In this embodiment, the hierarchical sorting is specifically as follows: the mth answer buffer pool is divided into 4 different levels according to the first three evaluation intervals of the four-level proficiency level evaluation intervals set in A4, and the customer services in each level are sorted in descending order according to the secondary evaluation scores. When the comprehensive score of the customer service corresponding to the current level customer service is greater than the upper limit threshold of this level, the corresponding customer service enters a higher level; otherwise, it enters a lower level; for example, the nth customer service in the mth answer buffer pool is currently in the second level, that is, At the corresponding level, when the current comprehensive score of the nth customer service is greater than , then the nth customer service enters At the corresponding level, when the current customer service comprehensive score of the nth customer service is less than or equal to and greater than or equal to , then the nth customer service enters Corresponding level.

[0096] Furthermore, in this embodiment, the specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting include:

[0097] C21. Determine the number of response buffer pools of the nth customer service according to the acquired field of response expertise of the nth customer service. If the nth customer service has at least one response buffer pool, the specific settings include:

[0098] C211. Set the initial answer score and dispatch frequency of the nth customer service in the mth answer buffer pool to 0. When the nth customer service is not selected for dispatch, its current status in all corresponding answer buffer pools is "1", that is, the nth customer service is currently in an idle state;

[0099] C212. When the nth customer service is extracted from an existing answer buffer pool for scheduling, its current status in all corresponding answer buffer pools is represented by "0", that is, the nth customer service is in the scheduling state. When the scheduling response is completed, its current status in all corresponding answer buffer pools is converted from "0" to "1", and the scheduling frequency and initial response score of the nth customer service in the corresponding answer buffer pool are increased by 1;

[0100] C213, use AI-agent3 to evaluate the dispatch response process in C212 in real time and obtain the training evaluation score of the corresponding response process ;

[0101] C214, according to the obtained ,like , then the initial response score of the nth customer service in the mth response buffer pool remains unchanged. , then the initial response score of the nth customer service in the mth response buffer pool is increased by 1. , then the initial response score of the nth customer service in the mth response buffer pool is increased by 2, and when the nth customer service completes the response and obtains a "good review", the initial response score is increased by 1;

[0102] C215. Add the dispatch bonus points, response evaluation bonus points, and favorable comment bonus points obtained by the nth customer service in the process of C212-C214 to the customer service training score corresponding to the mth field, and obtain the customer service comprehensive score of the nth customer service in the mth field;

[0103] C216. According to the obtained customer service comprehensive score and the four-level proficiency level evaluation interval, the proficiency level of the nth customer service in the mth field at the current moment is obtained, and the hierarchical ranking of the nth customer service in the mth response buffer pool is updated twice according to the proficiency level at the current moment;

[0104] C218, if , then the initial answer score of the nth customer service in the mth answer buffer pool is reduced by 1, and when the nth customer service completes the answer and gets a "bad review", the initial answer score is directly reduced by 1;

[0105] C219, repeat the process of C215 to add the answer score obtained in C218 to the customer service training score corresponding to the m-th field, obtain the customer service comprehensive score of the n-th customer service in the m-th field, and repeat the process of C216 according to the obtained customer service comprehensive score to perform a second update on the hierarchical sorting;

[0106] C220, when the customer service comprehensive score corresponding to the proficiency level of the nth customer service in the mth response buffer pool is less than When the nth customer service is removed from the mth answer buffer pool, AI-agent3 monitors the answer problems obtained during the process of the nth customer service being extracted from the mth answer buffer pool, and feeds them back to the intelligent course library module to generate customer service training materials in the corresponding field, and conducts secondary training in the corresponding field. When the nth customer service's secondary customer service training score in the mth field meets the requirements, When the condition is met, the nth customer service will be automatically added to the mth response buffer pool;

[0107] C221, when the comprehensive scores of the customer service corresponding to the expertise level of the nth customer service in all the response buffer pools are less than When the nth customer service representative is directly excluded from the training process, the answer questions obtained by AI-agent3 monitoring are fed back to the intelligent course library module to generate customer service training materials in the corresponding field for retraining.

[0108] C222. When the nth customer service is currently selected from the p answer buffer pools, the corresponding proficiency level of the nth customer service in the p answer buffer pools is obtained and compared, the answer buffer pool corresponding to the maximum proficiency level of the nth customer service is obtained, and the call information corresponding to the field of the nth customer service answer is selected from the answer buffer pool;

[0109] C223. When two or more of the nth customer service have the same expertise level and the highest expertise level in the p answer buffer pools, the customer service comprehensive scores of the nth customer service in the answer buffer pools with the same expertise level are compared, and the nth customer service is scheduled to answer the incoming call information in the corresponding field from the answer buffer pool with the highest customer service comprehensive score, and the current state of the nth customer service in the remaining answer buffer pool is changed to "0";

[0110] C224. Repeat C213-C221 to evaluate the response process of C222 and C223, and conduct secondary training based on the evaluation results.

[0111] This process quantifies the evaluation and dynamic adjustment of customer service staff's response capabilities in different areas. The system can assign them to the most suitable work tasks based on their actual performance and expertise, thereby maximizing the overall effectiveness of the customer service team. This mechanism not only helps to improve customer service response speed and problem-solving efficiency, but also effectively reduces service quality problems caused by mismatched allocation, thereby improving customer satisfaction. In addition, by real-time monitoring and evaluation of customer service performance, the system can promptly identify customer service's shortcomings in specific areas, and quickly take targeted training measures to make up for these shortcomings, thereby ensuring the professionalism and competitiveness of the customer service team.

[0112] Furthermore, the specific steps of building the enhanced scheduling model in this embodiment include:

[0113] D1. Based on the obtained answer buffer pool and incoming call voice information, build the current input state of the enhanced scheduling model ,in, Indicates the current status of the nth customer service in the mth response buffer pool, that is, whether it is idle or in scheduling state; Indicates the hierarchical sorting status of the nth customer service in the mth response buffer pool at the current moment; Indicates the comprehensive score of the nth customer service in the mth response buffer pool at the current moment. Indicates the scheduling frequency of the nth customer service unit time in the mth response buffer pool at the current moment. represents the probability that the qth incoming call at the current moment belongs to the mth domain, Indicates the waiting time for the qth incoming call belonging to the mth area at the current moment, represents the probability that the qth incoming call in the mth domain is dispatched to the nth customer service in the mth answer buffer pool at the current moment, represents the probability that the nth customer service representative in the mth answer buffer pool to which the qth incoming call is dispatched at the current moment is the customer service representative with the highest comprehensive score in the mth answer buffer pool; further, in this embodiment, and It is obtained through the historical customer service response process scheduling data and matching algorithm;

[0114] D2. Set the execution action trigger information according to the current input state, including:

[0115] D21. When the nth customer service is in the dispatching state, the first execution information is triggered, that is, the action of executing the corresponding customer service state conversion is triggered, and the action of changing the nth customer service state from "1" to "0" is executed; otherwise, the action of changing the nth customer service state from "0" to "1" is executed;

[0116] D22. When there are u customer services in the mth answer buffer pool that are currently idle, and the mth answer buffer pool currently corresponds to v incoming call schedulings in the mth field, the second execution information is triggered, that is, the action of scheduling the customer service with the largest customer service comprehensive score among the u customer services in the mth answer buffer pool that are idle to the incoming call with the longest waiting time is executed, and the above actions are repeated until all incoming calls and customer services are scheduled;

[0117] D23, when the remaining idle customer service in D22 cannot meet the call scheduling demand of the corresponding answer buffer pool, the third execution information is triggered, that is, the remaining call in the mth answer buffer pool is dispatched to the customer service with the largest comprehensive score in the remaining answer buffer pool for answering, and the answering process is assisted by AI-agent3, and the questions that cannot be assisted are recorded and fed back to the idle customer service in the corresponding answer buffer pool for a second call back reply;

[0118] D3, based on the D2 action trigger information, build the current execution action ,in, Indicates the action that triggers the execution of the first execution information. Indicates the action that triggers the execution of the corresponding second execution information. Indicates the action that triggers the execution corresponding to the third execution information;

[0119] D4. Construct a reward function based on the corresponding execution action of D3 , Indicates the g-th execution of an action The corresponding reward is when the nth customer service status transition is correct for the gth execution. ,on the contrary, ; G represents the total number of times the nth customer service state changes; Indicates execution of an action The corresponding reward is the reward for dispatching all v incoming calls from the mth field to u idle customer service representatives in the mth answer buffer pool. It represents the reward for dispatching g calls out of v calls in the mth domain to u idle customer service representatives in the mth answer buffer pool. It represents the reward for dispatching h calls out of v calls in the mth domain to h idle customer service representatives in the non-mth answer buffer pool; , and Represents the gain coefficient of the corresponding reward function;

[0120] D5. Construct a reinforcement scheduling model based on the reinforcement learning algorithm, and input the constructed current input state, execution action trigger information, current execution action and reward function into the reinforcement scheduling model for training. The trained reinforcement scheduling model is built into the answer scheduling module for scheduling between incoming calls and customer service.

[0121] The training and correction module is used to monitor the response process of the customer service scheduled by the response scheduling module in real time through AI-agent3 with a built-in semantic parsing model, and to feed back the monitoring results to the corresponding customer service in real time through AI-agent3 to assist in correcting the customer service's response process; further, the semantic parsing model in this embodiment is constructed and obtained by the Transformer algorithm or the BERT algorithm.

[0122] The training evaluation module uses the built-in AI-agent3 to conduct real-time evaluation of the response process of the dispatched customer service, obtain the training evaluation score, and conduct secondary training and judgment on the customer service based on the training evaluation score;

[0123] The evaluation feedback module is used to feed back the results of the secondary training judgment of the corresponding customer service and the corresponding existing problems to the intelligent course library module to generate a secondary training course and conduct secondary training for the corresponding customer service; further, the secondary training of the corresponding customer service in this embodiment is the same as the process of C220 and C221 and will not be repeated here.

[0124] The interactive sub-platform is used for customer service registration and login, as well as interactive feedback between customer service and the simulation training subsystem and the practical training simulation feedback subsystem;

[0125] Embedded devices are used for the deployment and operation of the interactive sub-platform, simulation training subsystem, and practical training simulation feedback subsystem.

[0126] Example 2

[0127] See also Figure 3 Another embodiment provided by the present invention is a method for constructing an intelligent platform system based on an AI model, the steps comprising:

[0128] S1. Acquire customer service attribute data, historical training, work process data, micro-expression scoring data and training material data to build a secondary dynamic customer service portrait library and a customer service training material library; further, in this embodiment, the secondary dynamic customer service portrait library and the customer service training material library are constructed by the customer service portrait module, the scenario simulation module and the intelligent course library module in sequence;

[0129] S2. A recommendation engine is built between the constructed secondary dynamic customer service portrait library and the customer service training material library, the built-in recommendation engine is connected to the interactive sub-platform, and AI-agent 1 and a sharing sub-platform are built in the interactive platform to construct a simulation training subsystem; further, the sharing sub-platform in this embodiment is realized by an evaluation sharing module;

[0130] S3, simulate incoming call information through AI-agent2 with built-in role-playing model, and build a response buffer pool through the secondary evaluation score obtained by the secondary dynamic customer service portrait library and the evaluation sharing module, and build a reinforcement scheduling model and AI-agent3 into the constructed response buffer interface to obtain the training simulation feedback subsystem;

[0131] S4. Connect the constructed simulation training subsystem and practical training simulation feedback subsystem through the evaluation feedback module, and deploy the connected simulation training subsystem and practical training simulation feedback subsystem and the interactive sub-platform into the embedded device to construct an intelligent platform system.

[0132] Example 3

[0133] A computer-readable storage medium stores computer instructions, which, when executed, execute a method for building an intelligent platform system based on an AI model.

[0134] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements a method for constructing an intelligent platform system based on an AI model when executing the computer program.

[0135] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are within the protection of the present invention.

[0136] If the disclosed technical solution involves personal information, the product using the disclosed technical solution has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the disclosed technical solution involves sensitive personal information, the product using the disclosed technical solution has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. An intelligent platform system built based on AI model, characterized by: include: Simulation training subsystem and practical training simulation feedback subsystem; the simulation training subsystem includes a scenario simulation module, an intelligent course library module, a recommendation engine module, and a training evaluation module; The recommendation engine module recommends matching training courses to each customer service through a recommendation algorithm based on the secondary dynamic customer service portrait library obtained by the scenario simulation module and the customer service training database obtained by the intelligent course library module; the training evaluation module evaluates the training results of the customer service according to the recommended training courses in real time through the built-in AI-agent 1, obtains the corresponding customer service training score, and outputs the response areas that the customer service is good at and not good at and the degree of proficiency in the corresponding areas, and feeds back the response areas that the customer service is good at and not good at and the corresponding degree of proficiency to the corresponding customer service node in the secondary dynamic customer service portrait library for storage; The training simulation feedback subsystem includes a training simulation module, an answer preprocessing module, and an answer scheduling module; the training simulation module is used to simulate the incoming call information of different answering scenarios, emotional states, and fields through the AI-agent2 with a built-in role-playing model based on the collected historical incoming call voice information; the answer scheduling module constructs an answer buffer pool through the acquired secondary dynamic customer service portrait library and the secondary evaluation score obtained by the evaluation sharing module, and according to the constructed answer buffer pool and the fields and problems in the incoming call information obtained by the answer preprocessing module, the customer service in the answer buffer pool is scheduled through the enhanced scheduling model to answer the incoming call information of the corresponding field; The specific steps of the response scheduling module to construct the response buffer pool include: C1. Build M domain response buffer pools based on the customer service response domains stored in the secondary dynamic customer service portrait library, and store the corresponding response domain customer service training scores that meet The customer service information of the conditions is stored in the corresponding domain response buffer pool constructed; Among them, represents the customer service training score of the nth customer service representative in the mth field; Indicates the evaluation threshold set; C2. hierarchically sort the customer services in the mth answer buffer pool according to the proficiency level evaluation interval, and update the hierarchical sorting of all answer buffer pools after hierarchical sorting using the obtained secondary evaluation scores, and perform scheduling and answer evaluation settings for all answer buffer pools after hierarchical sorting; The training simulation feedback subsystem also includes a training evaluation module, which uses the built-in AI-agent 3 to perform real-time evaluation on the response process of the dispatched customer service, obtain the training evaluation score, and perform secondary training on the customer service according to the training evaluation score; The specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting include: The number of response buffer pools of the nth customer service is determined according to the acquired answering field corresponding to the nth customer service. If the nth customer service has at least one response buffer pool, the specific settings include: C211. Set the initial answer score and scheduling frequency of the nth customer service in the mth answer buffer pool to 0. When the nth customer service is not selected for scheduling, its current status in all corresponding answer buffer pools is "1", that is, the nth customer service is currently in an idle state; C212. When the nth customer service is extracted from an existing response buffer pool for scheduling, its current state in all corresponding existing response buffer pools is represented by "0", that is, the nth customer service is in the scheduling state. When the scheduling response is completed, its current state in all corresponding existing response buffer pools is converted from "0" to "1", and the scheduling frequency and initial response score of the nth customer service in the corresponding response buffer pool are each increased by 1.

2. The intelligent platform system based on AI model construction as claimed in claim 1, characterized in that: The specific steps of the training evaluation module outputting the corresponding customer service's strong and weak response areas include: A1. Obtain the training process and result data of M field training courses recommended by N customer service personnel according to the recommendation engine module, and pre-process the obtained data; the training process and result data include: the completion rate of each field course, the correct rate of answering questions in each field, the corresponding rate of answering questions in each field, and the micro-expression state score of the customer service personnel answering questions in each field in the scenario simulation; A2. Based on the preprocessed training process and result data, Q response evaluation indicators corresponding to each field are constructed. According to the constructed response evaluation indicators, the customer service training evaluation set is obtained. ; A3. Build an evaluation model based on the fuzzy evaluation algorithm and embed the built model into AI-agent1. Input this into AI-agent1 to obtain the customer service training score of the nth customer service representative in the mth field. ; A4. Set the three-level assessment threshold and and , set the fourth-level proficiency level assessment interval according to the third-level assessment threshold, specifically: When When When , it means that the nth customer service is more proficient in the mth field. When , it means that the nth customer service representative is not good at the mth field; A5: The four-level proficiency evaluation interval set in A4 is built into AI-agent1. Based on the customer service training score obtained by real-time evaluation, AI-agent1 outputs the proficiency of the nth customer service in the mth field, and obtains the corresponding proficient and non-proficient response fields of the nth customer service based on the proficiency and customer service training score. When the corresponding customer service training scores in M ​​fields meet the requirements, If the conditions are not met, the corresponding customer service staff will not be allowed to enter the practical training stage.

3. The intelligent platform system based on AI model construction as claimed in claim 2, characterized in that: The specific steps of constructing the role-playing model include: B1. Obtaining the incoming call text information by using the voice-to-text extraction algorithm based on the historical incoming call voice information; B2. Automatically label the acquired incoming call text information with domain labels according to the association rules, and construct an incoming call text library using the labeled incoming call text information; B3, input the acquired historical incoming call voice information into the pre-trained voice extraction sub-model, acquire different types of basic voice features in the corresponding incoming call voice information, and construct an incoming call voice basic feature library using the acquired basic voice features; B4, input the acquired historical incoming call voice information into the pre-trained emotion extraction sub-model, obtain different types of voice emotion features and emotion types in the corresponding incoming call voice information, and use the acquired voice emotion features and emotion types to build an incoming call emotion feature library; B5. Use a random algorithm to build and pre-train a random generation sub-model, and build a built-in answering difficulty level into the random generation sub-model. Input the features in the incoming call information text library, incoming call voice basic feature library, and incoming call emotional feature library into the random generation sub-model to simulate the generation of incoming call voice information at different answering difficulty levels.

4. The intelligent platform system based on AI model construction as claimed in claim 3, characterized in that: The specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting also include: C213, through the AI-agent3, the dispatch response process in C212 is evaluated in real time to obtain the training evaluation score of the corresponding response process ; C214, according to the obtained ,like , then the initial response score of the nth customer service in the mth response buffer pool remains unchanged. , then the initial response score of the nth customer service in the mth response buffer pool is increased by 1. , then the initial response score of the nth customer service in the mth response buffer pool is increased by 2, and when the nth customer service completes the response and obtains a "good review", the initial response score is increased by 1; C215. Add the dispatch bonus points, response evaluation bonus points, and favorable comment bonus points obtained by the nth customer service in the process of C212-C214 to the customer service training score corresponding to the mth field, to obtain the customer service comprehensive score of the nth customer service in the mth field; C216. Based on the obtained customer service comprehensive score and the four-level expertise level evaluation range, obtain the expertise level of the nth customer service in the mth field at the current moment, and perform a second update on the hierarchical ranking of the nth customer service in the mth response buffer pool based on the expertise level at the current moment.

5. The intelligent platform system based on AI model construction as claimed in claim 4, characterized in that: The specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting also include: C218, if , then the initial response score of the nth customer service in the mth response buffer pool is reduced by 1, and when the nth customer service completes the response and obtains a "bad review", the initial response score is directly reduced by 1; C219, repeating the process of C215 to add the answer score obtained in C218 to the customer service training score corresponding to the m-th field, obtaining the customer service comprehensive score of the n-th customer service in the m-th field, and repeating the process of C216 to update the hierarchical sorting for the second time according to the obtained customer service comprehensive score; C220, when the customer service comprehensive score corresponding to the proficiency level of the nth customer service in the mth response buffer pool is less than When the nth customer service is removed from the mth answer buffer pool, AI-agent3 monitors the answer problems obtained during the process of the nth customer service being extracted from the mth answer buffer pool, and feeds them back to the intelligent course library module to generate customer service training materials in the corresponding field, and conducts secondary training in the corresponding field. When the nth customer service's secondary customer service training score in the mth field meets the requirements, When the condition is met, the nth customer service will be automatically added to the mth response buffer pool; C221, when the comprehensive scores of the customer service corresponding to the expertise level of the nth customer service in all the response buffer pools are less than When the nth customer service representative is directly excluded from the training process, the answer questions obtained by AI-agent3 monitoring are fed back to the intelligent course library module to generate customer service training materials in the corresponding field for retraining.

6. The intelligent platform system based on AI model construction as claimed in claim 5, characterized in that: The specific steps of scheduling and setting response evaluation for all response buffer pools after hierarchical sorting also include: C222. When the nth customer service is currently selected from p answer buffer pools, the corresponding proficiency level of the nth customer service in the p answer buffer pools is obtained and compared, the answer buffer pool corresponding to the maximum proficiency level of the nth customer service is obtained, and the call information of the field corresponding to the answer of the nth customer service is selected from the answer buffer pool; C223. When two or more of the nth customer service have the same expertise level and the highest expertise level in the p answer buffer pools, the customer service comprehensive scores of the nth customer service in the answer buffer pools with the same expertise level are compared, and the nth customer service is scheduled from the answer buffer pool with the highest customer service comprehensive score to answer the incoming call information in the corresponding field, and the current state of the nth customer service in the remaining answer buffer pools is changed to "0"; C224. Repeat C213-C221 to evaluate the response process of C222 and C223, and conduct secondary training based on the evaluation results.

7. The intelligent platform system based on AI model construction as claimed in claim 6, characterized in that: The training simulation feedback subsystem also includes a training correction module, which is used to monitor the response process of the customer service dispatched by the response scheduling module in real time through the AI-agent3 with a built-in semantic parsing model, and feed back the monitoring results to the corresponding customer service in real time through the AI-agent3 to assist in correcting the customer service's response process; The specific steps of building the enhanced scheduling model include: D1. Based on the obtained answer buffer pool and incoming call voice information, build the current input state of the enhanced scheduling model ,in, Indicates the current status of the nth customer service in the mth response buffer pool, that is, whether it is idle or in scheduling state; Indicates the hierarchical sorting status of the nth customer service in the mth response buffer pool at the current moment; Indicates the comprehensive score of the nth customer service in the mth response buffer pool at the current moment. Indicates the scheduling frequency of the nth customer service unit time in the mth response buffer pool at the current moment. represents the probability that the qth incoming call at the current moment belongs to the mth domain, Indicates the waiting time for the qth incoming call belonging to the mth area at the current moment, represents the probability that the qth incoming call in the mth domain is dispatched to the nth customer service in the mth answer buffer pool at the current moment, It indicates the probability that the nth customer service representative in the mth answer buffer pool to which the qth incoming call is dispatched at the current moment is the customer service representative with the highest comprehensive score in the mth answer buffer pool; D2. Setting execution action trigger information according to the current input state, specifically including: D21. When the nth customer service is in the dispatching state, the first execution information is triggered, that is, the action of executing the corresponding customer service state conversion is triggered, and the action of changing the nth customer service state from "1" to "0" is executed; otherwise, the action of changing the nth customer service state from "0" to "1" is executed; D22. When there are u customer services in the mth answer buffer pool that are currently idle, and the mth answer buffer pool currently corresponds to v incoming call schedulings in the mth field, the second execution information is triggered, that is, the action of scheduling the customer service with the largest customer service comprehensive score among the u customer services in the mth answer buffer pool that are idle to the incoming call with the longest waiting time is executed, and the above actions are repeated until all incoming calls and customer services are scheduled; D23, when the remaining idle customer service in D22 cannot meet the call scheduling demand of the corresponding answer buffer pool, the third execution information is triggered, that is, the remaining call in the mth answer buffer pool is dispatched to the customer service with the largest comprehensive score in the remaining answer buffer pool for answering, and the answering process is assisted by AI-agent3, and the questions that cannot be assisted are recorded and fed back to the idle customer service in the corresponding answer buffer pool for a second call back reply; D3: Construct the current execution action based on the execution action trigger information described in D2 ,in, Indicates the action that triggers the execution of the first execution information. Indicates the action that triggers the execution of the corresponding second execution information. Indicates the action that triggers the execution of the corresponding third execution information.

8. The intelligent platform system based on AI model construction as claimed in claim 7, characterized in that: The specific steps of building the enhanced scheduling model also include: D4. Construct a reward function based on the corresponding execution action of D3 , Indicates the g-th execution of an action The corresponding reward is when the nth customer service state transition is correct for the gth execution. ,on the contrary, ; G represents the total number of times the nth customer service state changes; Indicates execution of an action The corresponding reward is the reward for dispatching all v incoming calls from the mth field to u idle customer service representatives in the mth answer buffer pool. It represents the reward for dispatching g calls out of v calls in the mth domain to u idle customer service representatives in the mth answer buffer pool. It represents the reward for dispatching h calls out of v calls in the mth domain to h idle customer service representatives in the non-mth answer buffer pool; , and Represents the gain coefficient of the corresponding reward function; D5. Construct a reinforcement scheduling model based on the reinforcement learning algorithm, and input the constructed current input state, execution action trigger information, current execution action and reward function into the reinforcement scheduling model for training. The trained reinforcement scheduling model is built into the answer scheduling module for scheduling between incoming calls and customer service.

9. The intelligent platform system based on AI model construction as claimed in claim 8, characterized in that: The built-in answer difficulty levels include easy, normal, difficult and hard levels; the easy level scoring interval is Corresponding to; the general grade scoring interval is Corresponding to; the more difficult level scoring interval and Corresponding to; the difficulty level scoring interval is Corresponding; the M domain response buffer pools correspond one-to-one to the M domains.

10. A method for constructing an intelligent platform system based on an AI model, which is implemented based on the intelligent platform system constructed based on an AI model according to any one of claims 1 to 9, characterized in that the steps include: S1. Obtain customer service attribute data, historical training, work process data, micro-expression score data and training material data to build a secondary dynamic customer service portrait library and customer service training material library; S2. A recommendation engine is built between the constructed secondary dynamic customer service portrait library and the customer service training material library, the built-in recommendation engine is connected to the interactive sub-platform, and AI-agent 1 and a sharing sub-platform are built in the interactive platform to construct a simulation training subsystem; S3. Incoming call information is simulated through AI-agent2 with a built-in role-playing model. At the same time, a response buffer pool is constructed through the secondary evaluation scores obtained through the secondary dynamic customer service portrait library and the evaluation sharing module. The reinforcement scheduling model and AI-agent3 are built into the constructed response buffer interface to obtain a training simulation feedback subsystem.

Images