An intelligent customer service system optimization method and system based on natural language processing

Abstract

The application discloses an intelligent customer service system optimization method and system based on natural language processing, relates to the technical field of natural language processing, and comprises the following steps: constructing a workflow arrangement engine, and connecting multiple intelligent customer service applications for different business scenarios to the workflow arrangement engine; receiving user consultation content through the workflow arrangement engine, performing intention recognition and business scenario classification on the user consultation content by using a natural language processing model, selecting corresponding intelligent customer service applications for response based on the classification result; and performing real-time monitoring and data analysis on the response result of the intelligent customer service applications, and adaptively adjusting the scheduling strategy of the workflow arrangement engine based on user feedback and response accuracy. The application constructs a dynamic scheduling mechanism based on a Markov decision process and a multi-dimensional service quality evaluation system, and significantly improves the service response speed, accuracy and stability of the intelligent customer service system in multiple scenarios.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, in particular to an intelligent customer service system optimization method and system based on natural language processing. BACKGROUND

[0002] As an important support for enterprise digital transformation, intelligent customer service systems achieve intelligent and automated customer service through natural language processing technology. Traditional intelligent customer service systems mainly use rule-based dialogue management and fixed task allocation mechanisms, which have obvious limitations in handling complex and variable business scenarios. With the development of deep learning technology, the new generation of intelligent customer service systems has begun to introduce neural network models for semantic understanding and intent recognition, significantly improving the interactive experience of the system. However, in actual application, due to the lack of effective service quality evaluation mechanisms and dynamic scheduling strategies, existing systems are difficult to cope with load fluctuations during business peak periods, resulting in service response delays and accuracy decreases. In multi-scene concurrent service, fixed scheduling strategies cannot optimize resource allocation according to real-time service capacity, affecting the overall service efficiency of the system.

[0003] In the prior art, the service quality evaluation of intelligent customer service systems usually relies on simple statistical indicators such as average response time and conversation completion rate, lacking dynamic monitoring and real-time optimization capabilities for the service process. In terms of task scheduling, a fixed weight-based scoring mechanism is mostly used, which cannot adapt to differentiated needs in different business scenarios. In addition, existing systems often use random allocation or polling methods when handling conversation migration, ignoring the impact of historical service quality on scheduling decisions, resulting in significant fluctuations in service quality. Therefore, there is an urgent need for a technical solution that can achieve dynamic service quality evaluation and intelligent scheduling optimization to improve the service stability and efficiency of intelligent customer service systems. The intelligent customer service system optimization method and system based on natural language processing proposed in the present application are aimed at the above technical problems, by introducing a scheduling algorithm based on Markov decision process and a multi-dimensional service quality evaluation mechanism, realizing the dynamic optimization and adaptive scheduling of intelligent customer service systems. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] Therefore, the present application provides an intelligent customer service system optimization method and system based on natural language processing, which can solve the problems mentioned in the background art.

[0006] To solve the above technical problems, the application provides the following technical scheme: an intelligent customer service system optimization method based on natural language processing, comprising: constructing a workflow orchestration engine, and connecting multiple intelligent customer service applications for different business scenarios to the workflow orchestration engine; the workflow orchestration engine comprises an application scheduling module and a context management module; receiving user consultation content through the workflow orchestration engine, performing intent recognition and business scenario classification on the user consultation content by using a natural language processing model, selecting a corresponding intelligent customer service application for response based on the classification result; and performing real-time monitoring and data analysis on the response result of the intelligent customer service application, and adaptively adjusting the scheduling strategy of the workflow orchestration engine based on user feedback and response accuracy.

[0007] As a preferred scheme of the intelligent customer service system optimization method based on natural language processing, each intelligent customer service application is provided with a corresponding knowledge base, model parameters and prompt word configuration.

[0008] The application scheduling module adopts a scheduling algorithm based on Markov decision process, comprising:

[0009] The performance indicators of the intelligent customer service application within a time window are collected, and the state space matrix comprising response accuracy, real-time load rate and service response score is constructed in combination with the session state information of the context management module;

[0010] The state transition probability is calculated based on the state space matrix, and the task scheduling priority score is calculated in combination with real-time service indicators;

[0011] The scheduling optimization is performed based on the context management module, specifically, for the intelligent customer service application whose task scheduling priority score is detected to continuously decrease, the context vector of the current session is obtained, the historical session similarity with other intelligent customer service applications is calculated, and the current session is migrated to the application with the highest task scheduling priority score and supporting the business scenario according to the historical session similarity result;

[0012] When the task scheduling priority score of the intelligent customer service application starts to recover but the service response score fluctuates, the task allocation proportion of the intelligent customer service application in the corresponding business scenario is adjusted according to the response accuracy in the state space matrix, and the requests of part of the business scenarios are redistributed to the application with the highest scheduling score.

[0013] As a preferred scheme of the intelligent customer service system optimization method based on natural language processing, the task scheduling priority score is calculated by the following formula:

[0014] ;

[0015] wherein, is a state transition probability, represents an application service state at time t, represents a current task allocation decision, is a service response score calculated based on a session completion rate and a knowledge base matching degree, is a real-time load rate, and are a standard deviation and a mean value of the service response score, respectively.

[0016] As a preferred solution of the intelligent customer service system optimization method based on natural language processing provided by the application, wherein: the context management module adopts a hierarchical data structure to store the session state information, including:

[0017] a key-value pair database is adopted to establish a basic data layer to store session identification, time stamp, intelligent customer service application identification, and response accuracy rate;

[0018] a vector database is adopted to construct a vector data layer to store the context vector; the vector database establishes a vector index structure based on a local sensitive hashing algorithm;

[0019] a time series database is adopted to establish a state data layer to record time series data of the service response score, and if it is detected that the storage amount of the time series data exceeds a first preset threshold, historical data exceeding a preset time range is migrated to a cold storage area, and the vector index structure is retained in the cold storage area.

[0020] As a preferred solution of the intelligent customer service system optimization method based on natural language processing provided by the application, wherein: a natural language processing model is used to perform intent recognition and business scenario classification on the user consultation content, and a corresponding intelligent customer service application is selected for response based on the classification result, including the following steps:

[0021] a deep semantic understanding network is used to extract feature representation of the user consultation content, generate a query vector, and retrieve similar session records from the vector data layer;

[0022] a scenario feature vector is obtained according to the query vector and the service response score in the time series data layer;

[0023] intent recognition and scenario classification are performed based on the scenario feature vector, if the highest classification score corresponds to multiple intelligent customer service applications, task scheduling priority scores of each intelligent customer service application are obtained, and the application with the highest score is selected for response; if only a single intelligent customer service application is corresponded, it is judged whether the task scheduling priority score of the intelligent customer service application is higher than a second preset threshold, if yes, the intelligent customer service application is selected for response, otherwise, the request is allocated to an application in a standby application pool with the highest task scheduling priority score.

[0024] As a preferred scheme of the intelligent customer service system optimization method based on natural language processing, wherein: the scene feature vector is represented by the following formula:

[0025] ;

[0026] Wherein, is the scene feature vector; is the current query vector; is the vector representation of the i-th similar historical conversation; is the service response score of the i-th similar historical conversation at time t; k is the number of selected most similar conversations; is the weight coefficient for balancing the current query and historical information;

[0027] The classification score is represented by the following formula:

[0028] ;

[0029] Wherein, W is the weight matrix of the classification layer, and b is the bias term of the classification layer.

[0030] As a preferred scheme of the intelligent customer service system optimization method based on natural language processing, wherein: the response result of the intelligent customer service application is monitored and analyzed in real time, and based on user feedback and response accuracy, the scheduling strategy of the workflow orchestration engine is adaptively adjusted, including the following steps:

[0031] Obtain the service response score, user feedback score and response accuracy in the state space matrix in the time series database, and generate a service quality evaluation result;

[0032] Based on the service quality evaluation result, the scheduling strategy of the workflow orchestration engine is optimized, specifically: if the service quality evaluation result is continuously lower than the quality threshold and the task scheduling priority score of the intelligent customer service application is in a downward trend, the intelligent customer service application is temporarily removed from the application calling sequence of the current business scenario; if the service quality evaluation result returns to normal and the task scheduling priority score is higher than the second preset threshold, the intelligent customer service application is added to the end of the application calling sequence of the business scenario; if the service quality evaluation result continuously exceeds the quality threshold and the task scheduling priority score continues to rise, the priority of the intelligent customer service application in the application calling sequence is gradually increased.

[0033] To further solve the above technical problems, the present invention provides the following technical solutions: an intelligent customer service system optimization system based on natural language processing, comprising: an orchestration module for building a workflow orchestration engine and connecting multiple intelligent customer service applications for different business scenarios to the workflow orchestration engine;

[0034] A processing module is configured to receive user consultation content through the workflow orchestration engine, perform intent recognition and business scenario classification on the user consultation content using a natural language processing model, and select a corresponding intelligent customer service application for response based on the classification results;

[0035] The optimization module is used to monitor and analyze the response results of the intelligent customer service application in real time, and adaptively adjust the scheduling strategy of the workflow orchestration engine based on user feedback and response accuracy.

[0036] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, it implements the steps of the intelligent customer service system optimization method based on natural language processing as described above.

[0037] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the intelligent customer service system optimization method based on natural language processing as described above are implemented.

[0038] Beneficial effects of the present invention: The present invention realizes unified scheduling and dynamic optimization of multi-scenario intelligent customer service applications by constructing a workflow orchestration engine and a context management module of a hierarchical data structure. The scheduling algorithm based on the Markov decision process, combined with real-time service indicators and historical session data, enables the system to adaptively adjust the task allocation strategy, significantly reducing service response time and quality fluctuations. During the session migration process, accurate migration is performed by calculating the similarity of historical sessions, maintaining a high level of service satisfaction. In addition, the multi-dimensional service quality evaluation mechanism designed by the present invention realizes comprehensive monitoring and dynamic optimization of service quality by integrating user feedback scores, response accuracy and other indicators, effectively solving the problem that the fixed weight scoring method in the prior art is difficult to adapt to dynamic service scenarios, so that the system can maintain stable service performance in different business scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 A whole flow schematic diagram of an intelligent customer service system optimization method based on natural language processing is provided for the present application.

[0041] Figure 2 A logic schematic diagram of an application scheduling module of an intelligent customer service system optimization method based on natural language processing is provided for the present application.

[0042] Figure 3 An interaction schematic diagram of an intelligent customer service system optimization system based on natural language processing is provided for the present application.

[0043] Figure 4 A computer device diagram in an intelligent customer service system optimization method based on natural language processing is provided for the present application. DETAILED DESCRIPTION

[0044] In order to make the above objectives, characteristics and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0045] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the scope of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0046] Embodiment 1, with reference to Figures 1-3 An intelligent customer service system optimization method based on natural language processing is provided for an embodiment of the present application.

[0047] In the related art,

[0048] The present application can effectively solve the above-mentioned problems. Next, how to implement the intelligent customer service system optimization method based on natural language processing will be described in detail in combination with multiple embodiments.

[0049] Figure 1 A whole flowchart of an intelligent customer service system optimization method based on natural language processing is shown, which includes:

[0050] S1: Construct a workflow orchestration engine, and access multiple intelligent customer service applications for different business scenarios to the workflow orchestration engine.

[0051] Specifically, the workflow orchestration engine is a micro-service scheduling platform built on a distributed architecture, including an application scheduling module and a context management module. The application scheduling module adopts a scheduling algorithm based on Markov decision process, as shown in the following formula: Figure 2 The method comprises the following steps:

[0052] First, the performance indicators of the intelligent customer service application within a specified time window T are collected, and combined with the session state information of the context management module, a state space matrix S containing response accuracy, real-time load rate and service response score is constructed.

[0053] It should be noted that the business scenarios include at least one of product consultation, order inquiry, after-sales service and complaint handling. The performance indicators include average response time, session completion rate, knowledge base matching degree and system resource utilization rate.

[0054] Secondly, the state transition probability is calculated based on the state space matrix, and the task scheduling priority score is calculated combined with the real-time service indicators. The calculation of state transition probability adopts the standard method in Markov decision process, and the calculation formula of task scheduling priority score is as follows:

[0055]

[0056] wherein, is the state transition probability, represents the application service state at time t, represents the current task allocation decision, is the service response score calculated based on the session completion rate and the knowledge base matching degree, is the real-time load rate, and are the standard deviation and the mean value of the service response score, respectively.

[0057] Based on the context management module for scheduling optimization: for the intelligent customer service application whose task scheduling priority score is continuously decreasing, the context vector of its current session is obtained, and the historical session similarity with other intelligent customer service applications is calculated, and according to the historical session similarity result, the current session is migrated to the application with the highest task scheduling priority score and supporting the business scenario; when the task scheduling priority score of the intelligent customer service application starts to rise but the service response score fluctuates, the task allocation proportion of the intelligent customer service application in the corresponding business scenario is adjusted according to the response accuracy in the state space matrix, and part of the requests of the business scenario are redistributed to the application with the highest scheduling score.

[0058] Table 1 Comparison of performance indicators under different scenarios

[0059] Scenario type Average response time reduction ratio Service accuracy rate improvement Session migration satisfaction maintenance rate Service quality fluctuation reduction ratio Regular scenario 28.3% 15.7% 93% 46.2% Promotion peak 35.1% 12.9% 91% 42.8% Emergency 31.2% 14.3% 90% 44.5% ​

[0060] By analyzing Table 1, the present application focuses on the problems existing in the current intelligent customer service system in multi-scene task scheduling, and proposes an intelligent scheduling scheme based on Markov decision process. The scheme quantitatively characterizes the service capability of the customer service application by constructing a state space matrix containing response accuracy, real-time load rate and service response score. In addition, the present application designs a task scheduling priority scoring mechanism that considers historical service performance, current service efficiency and service quality stability, solving the problem that the traditional fixed weight scoring method cannot adapt to dynamic service scenarios. In different scenarios, for the regular scenario, the scoring mechanism reduces the average response time of task allocation by 28.3%, and improves the service accuracy by 15.7%; during the promotion peak, the average response time is reduced by 35.1%, and the service accuracy is improved by 12.9%; in the emergency processing, the average response time is reduced by 31.2%, and the service accuracy is improved by 14.3%. In addition, the present application focuses on optimizing the conversation migration strategy, and through calculating the historical conversation similarity for accurate migration, the service satisfaction after conversation migration can be maintained above 90% in various scenarios. It is worth noting that this dynamic scheduling mechanism shows good adaptive ability, and in the regular scenario, promotion peak and emergency processing, the service quality fluctuation of the system is reduced by 46.2%, 42.8% and 44.5% respectively.

[0061] Further, the context management module stores the conversation state information using a hierarchical data structure, including:

[0062] The key-value pair database is used to establish a basic data layer to store the conversation identifier, timestamp, intelligent customer service application identifier, and response accuracy;

[0063] The vector database is used to construct a vector data layer to store the context vector; the vector database establishes a vector index structure based on the local sensitive hashing algorithm;

[0064] The time series database is used to establish a state data layer to record the time series data of the service response score. If it is detected that the storage amount of the time series data exceeds a first preset threshold, the historical data exceeding the preset time range is migrated to a cold storage area, and the vector index structure is retained in the cold storage area.

[0065] It should be noted that the first preset threshold refers to the upper limit of the capacity of the conversation state data in the time series database. In the intelligent customer service system, a large number of service response score records are generated for each conversation, and these time series data are used to calculate the standard deviation and mean, which in turn affects the task scheduling priority score. Based on actual business requirements and system performance considerations, the number of concurrent conversations for a single intelligent customer service application is usually used as a benchmark, combined with the sampling frequency of the service response score to determine the threshold, so as to ensure the calculation real-time while avoiding data storage pressure.

[0066] The preset time range defines the retention period of the time series data in the active storage area. Considering the service characteristics of the intelligent customer service system, the recent service response score is more valuable for calculating the task scheduling priority. The time range is usually determined based on the timeliness characteristics of the business scenario, such as product consultation sessions may focus on data within the last few hours, while after-sales service sessions may need to refer to historical data of a longer period. The system optimizes the utilization efficiency of storage resources through this differentiated time range setting.

[0067] S2: Receive user consultation content through the workflow orchestration engine, use natural language processing model to perform intent recognition and business scenario classification on the user consultation content, and select the corresponding intelligent customer service application for response based on the classification result.

[0068] Specifically, the feature representation of the user consultation content is extracted based on the deep semantic understanding network, a query vector is generated, and similar session records are retrieved from the vector data layer.

[0069] According to the query vector and the service response score in the time series data layer, a scenario feature vector is obtained:

[0070] ;

[0071] wherein, is the scenario feature vector; is the current query vector; is the vector representation of the i-th similar historical session; is the service response score of the i-th similar historical session at time t; k is the number of selected most similar sessions; is the weight coefficient for balancing the current query and historical information.

[0072] Based on the scenario feature vector, intent recognition and scenario classification are performed. If the highest-scoring scenario category corresponds to multiple intelligent customer service applications, the task scheduling priority score of each intelligent customer service application is obtained, the highest-scoring application is selected for response, and if only a single intelligent customer service application is involved, it is determined whether the task scheduling priority score of the intelligent customer service application is higher than a second preset threshold. If yes, the intelligent customer service application is selected for response, otherwise the request is distributed to the application with the highest task scheduling priority score in the standby application pool.

[0073] It should be noted that in the dispatching decision of the intelligent customer service system, the second preset threshold is a task scheduling access judgment value of a single intelligent customer service application. The threshold is set based on the calculation formula of the task scheduling priority score, and is used to judge whether the application that monopolizes a certain business scenario has sufficient service capability. Considering the quality of service requirements of the intelligent customer service system, the threshold is usually set to between 0.6-0.7, which not only guarantees the basic requirements of service quality, but also reserves a reasonable optimization space for system scheduling. When the score is lower than this threshold, the system will divert the request to the standby application pool, thereby ensuring the stability of the service quality.

[0074] The classification score is represented by the following formula:

[0075] ;

[0076] Wherein, W is the weight matrix of the classification layer, and b is the bias term of the classification layer.

[0077] Preferably, the traditional method often only relies on the semantic features of the current query for classification, ignoring the influence of historical service quality on classification decision, resulting in that the classification result cannot adapt to the dynamically changing service capability. In view of this problem, the application proposes a feature representation method that fuses historical service response scores. By introducing a service response score weighting mechanism (Wscore) into the calculation formula of the scene feature vector, ), the historical conversation with better service performance obtains higher reference value, thereby improving the accuracy and reliability of classification.

[0078] S3: Real-time monitoring and data analysis on the response result of the intelligent customer service application, and self-adaptive adjustment of the scheduling strategy of the workflow orchestration engine based on user feedback and response accuracy.

[0079] Specifically, first, the service response score, user feedback score and response accuracy in the state space matrix in the time series database are obtained to generate a service quality evaluation result.

[0080] Secondly, the scheduling strategy of the workflow orchestration engine is optimized based on the service quality evaluation result, specifically, if the service quality evaluation result is continuously lower than the quality threshold and the task scheduling priority score of the intelligent customer service application is in a downward trend, the intelligent customer service application is temporarily removed from the application calling sequence of the current business scenario; if the service quality evaluation result returns to normal and the task scheduling priority score is higher than the second preset threshold, the intelligent customer service application is added to the end of the application calling sequence of the business scenario; if the service quality evaluation result continuously exceeds the quality threshold and the task scheduling priority score continues to rise, the priority of the intelligent customer service application in the application calling sequence is gradually improved.

[0081] It should be noted that the user feedback score refers to the subjective evaluation of the user on the service quality after the end of the conversation in the intelligent customer service system, including the scores of three dimensions of problem solving, service attitude and response speed, the comprehensive score value obtained by weighted calculation, and is used to quantify the satisfaction degree of the user to the service.

[0082] The service quality evaluation result refers to the comprehensive evaluation index of service quality obtained by weighting and fusing the user feedback score, response accuracy rate and service response score, wherein the weight coefficients of each index are dynamically optimized based on historical service data by a machine learning method, and are used to comprehensively reflect the service capability of the intelligent customer service application.

[0083] The quality threshold refers to the reference value for evaluating the service quality in the intelligent customer service system, which is obtained by statistical analysis on historical service data, and the calculation method is the historical mean value of the quality evaluation result minus a times the standard deviation, wherein a is an adjustable coefficient parameter. Preferably, the value range of the quality threshold is [0.6, 0.8], when the service scene is regular consultation, the value of k is 1.5, and the quality threshold is usually set to 0.7; when the service scene is emergency handling, the value of k is 2.0, and the quality threshold can be appropriately reduced to 0.65; when the service scene is the peak period of promotion activities, the value of k is 1.8, and the quality threshold is set to 0.75, so as to ensure that the system can maintain a reasonable service quality standard under different business scenes. Based on the experimental data, the dynamic adjustment mechanism of the quality threshold can effectively balance the service quality and the system efficiency, so that the service evaluation of the intelligent customer service application is more adaptive.

[0084] In summary, the application realizes the unified scheduling and dynamic optimization of the multi-scene intelligent customer service application by constructing the workflow orchestration engine and the context management module of the hierarchical data structure. Based on the scheduling algorithm of Markov decision process, combined with real-time service indicators and historical conversation data, the system can adaptively adjust the task allocation strategy, significantly reducing the service response time and quality fluctuation. In the conversation migration process, the historical conversation similarity is calculated for accurate migration, and a high service satisfaction degree is maintained. In addition, the multi-dimensional service quality evaluation mechanism designed by the application fuses the user feedback score, response accuracy rate and other indicators to realize the comprehensive monitoring and dynamic optimization of the service quality, effectively solving the problem that the fixed weight scoring method in the prior art cannot adapt to the dynamic service scene, so that the system can maintain stable service performance under different business scenes.

[0085] Embodiment 2, refer to Figure 3 An intelligent customer service system optimization system based on natural language processing is provided for an embodiment of the application.

[0086] Figure 2 The module interaction diagram of the system includes:

[0087] The orchestration module is configured to construct a workflow orchestration engine, and access multiple intelligent customer service applications for different business scenarios to the workflow orchestration engine.

[0088] The processing module is configured to receive user consultation content through the workflow orchestration engine, perform intent recognition and business scenario classification on the user consultation content by using a natural language processing model, and select a corresponding intelligent customer service application for response based on the classification result.

[0089] The optimization module is configured to perform real-time monitoring and data analysis on the response result of the intelligent customer service application, and adaptively adjust the scheduling strategy of the workflow orchestration engine based on user feedback and response accuracy.

[0090] Embodiment 3, refer to Figure 4 As an embodiment of the present application, which is different from the previous embodiment, the function, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application or the part that contributes to the prior art or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0091] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions, or in conjunction with these instruction execution systems, devices or apparatus. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, device or apparatus, or in conjunction with these instruction execution systems, devices or apparatus.

[0092] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer readable medium can be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, using a suitable medium, into a computer readable medium.

[0093] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0094] Example 4, which is an embodiment of the present application, provides a natural language processing-based intelligent customer service system optimization method. In order to verify the beneficial effects of the present application, economic benefit calculation and simulation experiments are used for scientific demonstration.

[0095] In order to fully verify the technical effects of the present application, this embodiment constructs a distributed test environment containing 15 service nodes, each node is configured with Intel Xeon E5-2680 v4 processor, 32GB DDR4 memory and 1TB SSD storage. The test data set contains 2 million real customer service dialogue records from a large e-commerce platform, covering product consultation, order inquiry, after-sales service and other business scenarios. The control group adopts a fixed weight allocation scheme based on rules, and the experimental group adopts the Markov decision process dynamic scheduling scheme of the present application. The experimental period is 45 days, and different time periods simulate daily operation, holiday promotion, system failure and other scenarios, and collect system performance indicators. In order to ensure data reliability, a stratified sampling method is used, 2000 conversation samples are randomly selected for in-depth analysis for each type of scene, and a Bootstrap method is used for confidence interval estimation.

[0096] The embodiment focuses on the performance of the system under high concurrency and load fluctuation conditions. By injecting different intensities of request traffic (from 100 QPS to 1000 QPS), the load bearing capacity and service quality stability of the system are tested. At the same time, evaluation indicators such as session state migration success rate, knowledge base matching accuracy, and service quality consistency are introduced. In the data collection process, distributed tracking technology is used to record the complete processing link of each request, and real-time monitoring and data visualization are realized through Prometheus and Grafana. To ensure the objectivity of the experiment, the embodiment invites 50 professional testers to conduct function verification and user experience evaluation, and adopts a double-blind test method to collect user feedback data.

[0097] Table 2: System performance comparison data under different load conditions

[0098] Concurrent quantity (QPS) Scheme type Average response time (ms) CPU utilization rate (%) Memory occupancy rate (%) Service accuracy rate (%) Session completion rate (%) Knowledge base matching degree (%) State migration success rate (%) 100 Traditional scheme 245 35.6 42.3 87.5 89.2 85.6 88.4 100 The present application 182 28.4 35.7 93.8 95.6 92.3 94.7 500 Traditional scheme 678 72.5 78.4 82.3 84.7 80.2 82.6 500 The present application 385 58.9 65.2 90.5 92.8 88.9 91.5 1000 Traditional scheme 1256 94.8 92.6 75.4 78.9 72.8 74.5 1000 The present application 524 75.3 78.9 87.2 89.5 85.4 88.2

[0099] As shown in Table 2, through analysis of system performance data under different load conditions, the present application shows significant advantages in various core indicators. Under low load (100 QPS), the average response time is reduced by 25.7%, and CPU and memory occupancy are reduced by 20.2% and 15.6%, respectively; under medium load (500 QPS), the performance advantage is further expanded, with a response time reduction of 43.2% and more efficient system resource utilization; especially in the high load (1000 QPS) scenario, the present application shows excellent performance optimization effect, with a response time of only 41.7% of the traditional scheme, while the service accuracy and session completion rate can still remain at a high level (87.2% and 89.5%, respectively). Through the data of knowledge base matching degree and state migration success rate, it can be seen that the present application has obvious advantages in service quality guarantee, and even under high load conditions, the decline of these two indicators is much smaller than that of the traditional scheme.

[0100] Table 3: Multi-dimensional service quality evaluation index comparison data

[0101] Evaluation dimension Time window (min) Traditional scheme fluctuation rate (%) The present application fluctuation rate (%) Stability improvement rate (%) Abnormal recovery time (s) Service degradation trigger rate (%) Resource scheduling efficiency (%) Response time consistency 15 32.6 13.8 57.7 48 8.5 91.2 Service accuracy rate stability 30 28.9 12.4 57.1 42 7.2 93.5 Resource utilization balance degree 45 35.7 15.2 57.4 45 8.9 92.8 Session state continuity 60 30.5 13.1 57.0 38 6.8 94.1 System throughput fluctuation 90 33.8 14.5 57.1 44 8.1 92.9

[0102] As shown in Table 3, in all evaluation dimensions, the volatility of the present invention is significantly lower than that of the traditional solution, and the average stability improvement rate reaches 57.3%. In particular, in terms of response time consistency, the volatility dropped from 32.6% to 13.8%, which is the most significant improvement. By analyzing the data of different time windows, it can be found that the present invention maintains a high stability during long-term operation, and the service degradation trigger rate is less than 8% on average, which is much better than the traditional solution. The resource scheduling efficiency is always maintained above 90%, indicating that the dynamic scheduling strategy of the present invention can effectively balance the system load and service quality. It is worth noting that in terms of abnormal recovery time, the present invention also shows a stronger fault recovery capability, and the average recovery time is controlled within 45 seconds.

[0103] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for optimizing an intelligent customer service system based on natural language processing, characterized in that: include: Build a workflow orchestration engine and connect multiple intelligent customer service applications for different business scenarios to the workflow orchestration engine; The workflow orchestration engine includes an application scheduling module and a context management module; Receive user consultation content through the workflow orchestration engine, use the natural language processing model to identify the intent and classify the business scenarios of the user consultation content, and select the corresponding intelligent customer service application to respond based on the classification results; Perform real-time monitoring and data analysis on the response results of the intelligent customer service application, and adaptively adjust the scheduling strategy of the workflow orchestration engine based on user feedback and response accuracy; Each of the intelligent customer service applications is provided with a corresponding knowledge base, model parameters and prompt word configuration; The application scheduling module adopts a scheduling algorithm based on Markov decision process, including: Collecting performance indicators of the intelligent customer service application within a time window and, in combination with the session state information of the context management module, constructing a state space matrix including response accuracy, real-time load rate, and service response score; Calculating the state transition probability based on the state space matrix and calculating the task scheduling priority score in combination with the real-time service indicator; Scheduling optimization is performed based on the context management module. Specifically, for an intelligent customer service application whose task scheduling priority score is detected to be continuously decreasing, the context vector of the current session is obtained, and the similarity between the current session and the historical session of other intelligent customer service applications is calculated. Based on the historical session similarity results, the current session is migrated to the application with the highest task scheduling priority score and that supports the business scenario. When the task scheduling priority scheduling score of the intelligent customer service application begins to recover but the service response score fluctuates, the task allocation ratio of the intelligent customer service application in the corresponding business scenario is adjusted according to the response accuracy in the state space matrix, and the requests of some business scenarios are reallocated to the application with the highest scheduling score; The task scheduling priority score is calculated using the following formula: ； in, is the state transition probability, Indicates the application service status at time t, represents the current task allocation decision, The service response score is calculated based on the conversation completion rate and knowledge base matching degree. is the real-time load rate, and are the standard deviation and mean of the service response scores, respectively; The natural language processing model is used to identify the intent and classify the business scenarios of the user's consultation content, and the corresponding intelligent customer service application is selected to respond based on the classification results, including the following steps: Extracting feature representations of the user consultation content based on a deep semantic understanding network, generating a query vector, and retrieving similar conversation records from the vector data layer; Obtaining a scenario feature vector based on the query vector and the service response score in the time series data layer; Performing intent recognition and scene classification based on the scene feature vector; The scene feature vector is expressed by the following formula: ； in, is the scene feature vector; is the current query vector; is the vector representation of the i-th similar historical session; is the service response score of the i-th similar historical session at time t; k is the number of the most similar sessions selected; The weight coefficient for balancing the current query and historical information.

2. The method for optimizing an intelligent customer service system based on natural language processing according to claim 1, wherein: The context management module stores the session state information in a hierarchical data structure, including: A key-value database is used to establish a basic data layer to store session IDs, timestamps, smart customer service application IDs, and response accuracy. A vector database is used to construct a vector data layer to store the context vector; the vector database establishes a vector index structure based on a locality sensitive hashing algorithm; A time series database is used to establish a state data layer to record the time series data of the service response score. If it is detected that the storage volume of the time series data exceeds a first preset threshold, the historical data beyond the preset time range is migrated to a cold storage area, and the vector index structure is retained in the cold storage area.

3. The method for optimizing an intelligent customer service system based on natural language processing according to claim 2, wherein: If the scenario category with the highest classification score corresponds to multiple intelligent customer service applications, the task scheduling priority scores of each intelligent customer service application are obtained, and the application with the highest score is selected to respond; if it corresponds to only a single intelligent customer service application, it is determined whether the task scheduling priority score of the intelligent customer service application is higher than the second preset threshold. If so, the intelligent customer service application is selected to respond, otherwise the request is assigned to the application with the highest task scheduling priority score in the backup application pool.

4. The method for optimizing an intelligent customer service system based on natural language processing according to claim 3, wherein: The classification score is expressed by the following formula: ； Among them, W is the weight matrix of the classification layer, and b is the bias term of the classification layer.

5. The method for optimizing an intelligent customer service system based on natural language processing according to claim 4, wherein: Performing real-time monitoring and data analysis on the response results of the intelligent customer service application, and adaptively adjusting the scheduling strategy of the workflow orchestration engine based on user feedback and response accuracy, including the following steps: Obtaining the service response score, user feedback score and response accuracy in the state space matrix in the time series database to generate a service quality evaluation result; The scheduling strategy of the workflow orchestration engine is optimized based on the service quality evaluation result. Specifically, if the service quality evaluation result is continuously lower than the quality threshold and the task scheduling priority score of the intelligent customer service application is on a downward trend, the intelligent customer service application is temporarily removed from the application call sequence of the current business scenario; if the service quality evaluation result returns to normal and the task scheduling priority score is higher than the second preset threshold, the intelligent customer service application is added to the end of the application call sequence of the business scenario; if the service quality evaluation result continuously exceeds the quality threshold and the task scheduling priority score continues to rise, the priority of the intelligent customer service application in the application call sequence is gradually increased.

6. A natural language processing-based intelligent customer service system optimization system, based on the natural language processing-based intelligent customer service system optimization method according to any one of claims 1 to 5, characterized in that: include, An orchestration module, used to build a workflow orchestration engine and connect multiple intelligent customer service applications for different business scenarios to the workflow orchestration engine; A processing module is configured to receive user consultation content through the workflow orchestration engine, perform intent recognition and business scenario classification on the user consultation content using a natural language processing model, and select a corresponding intelligent customer service application for response based on the classification results; The optimization module is used to monitor and analyze the response results of the intelligent customer service application in real time, and adaptively adjust the scheduling strategy of the workflow orchestration engine based on user feedback and response accuracy.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent customer service system optimization method based on natural language processing described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent customer service system optimization method based on natural language processing according to any one of claims 1 to 5 are implemented.

Images