A data interaction method for intelligent customer service system with optimized response timeliness

Abstract

The present invention discloses a data interaction method for an intelligent customer service system for optimizing response timeliness, and specifically relates to the technical field of intelligent customer service systems. The specific steps include: collecting communication transfer efficiency information and concurrent scheduling management information of interactions between a large language model deployed on a cloud server and a local knowledge base deployed on a local server, obtaining a query return fluctuation coefficient and a request strategy priority coefficient, constructing an input floating model and generating a response fluctuation interference index, comparing the index with a preset response fluctuation interference threshold and performing signal classification, constructing a risk model to evaluate the fluctuation of response timeliness of intelligent customer service, and performing risk management according to the evaluation results. The present application effectively evaluates the communication transmission delay status between the cloud and the local, avoids the problem of large language model output deviation caused by excessive delay, and provides a method for identifying and judging large language model output faults for technical personnel.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent customer service systems, and more specifically, to an intelligent customer service system data interaction method for optimizing response timeliness. Background Art

[0002] The big language model generates an artificial intelligence model by using a large amount of data training to understand and generate natural language text. It is based on deep learning technology to capture the complexity and diversity of language. The local knowledge base is used to store knowledge sets on local devices or servers. The knowledge sets contain various types of information, such as documents, pictures, videos, audio, etc. By combining the big language model and the local knowledge base to build an intelligent customer service system, it can achieve 24-hour uninterrupted customer service, provide quick answers to inquiries from multiple customers at the same time in a short period of time, and learn user preferences and behavior patterns based on historical questions and interaction records to provide personalized services.

[0003] The advantage of an intelligent customer service system that integrates a large language model and a local knowledge base is that it provides a large amount of customer service in real time 24 hours a day. However, there is a difference in communication efficiency between the large language model deployed on the cloud and the local knowledge base deployed on the local server, which causes the large language model's understanding and reasoning ability of natural language to fluctuate, resulting in a deviation in the output content. There are many reasons for the deviation in the output of the large language model, but it is difficult for technical personnel to confirm whether the deviation occurs in the large language model during the iteration process, or whether the large language model has an understanding deviation fluctuation due to the data interaction between the local knowledge base and the large language model. It is difficult to confirm the effectiveness of the large language model itself, which causes technical personnel to waste time and energy on verifying the iterative evolution status of the large language model. On the other hand, processing a large number of concurrent requests during the peak period of customer service further hinders the communication efficiency between the large language model and the local knowledge base, resulting in delayed customer service management and untimely customer response.

[0004] In order to solve the above defects, a technical solution is now proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a data interaction method for an intelligent customer service system that optimizes response timeliness to address the deficiencies in the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a data interaction method for an intelligent customer service system that optimizes response timeliness, the specific steps comprising:

[0007] Collect the communication transfer efficiency information and concurrent scheduling management information of the interaction between the large language model deployed on the cloud server and the local knowledge base deployed on the local server, obtain the query return fluctuation coefficient based on the communication transfer efficiency information, and obtain the request strategy priority coefficient based on the concurrent scheduling management information;

[0008] An input floating model is constructed based on the query return fluctuation coefficient and the request strategy priority coefficient to evaluate the communication efficiency between the cloud server and the local server and the impact of concurrent requests on the intelligent customer service response, and generate a response fluctuation interference index.

[0009] Comparing the generated response fluctuation interference index with a preset response fluctuation interference threshold, and classifying the signal according to the comparison result;

[0010] A risk model is constructed based on the results of signal classification, the fluctuations in the response timeliness of intelligent customer service are evaluated, and risk management is performed based on the evaluation results.

[0011] Optionally, the method for acquiring the communication transfer efficiency information is:

[0012] Obtain the delay time for the cloud-based large language model to initiate interaction with the local knowledge base within the cycle time T, and mark the delay time for the cloud-based large language model to initiate interaction with the local knowledge base within the cycle time T as , where d is the number of the delay time for the large language model in the cloud to initiate interaction with the local knowledge base within the cycle time T, where ,and is a positive integer;

[0013] Get the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T, and mark the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T as , where b is the number of the delay time of the local knowledge base responding to the large language model in the cloud within the cycle time T, where ,and is a positive integer;

[0014] The communication transfer efficiency information is the discrete ratio of the delay time of the large language model on the cloud to initiate interaction with the local knowledge base and the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T. The calculation expression is: ,in, Transfer efficiency information for communication.

[0015] Optionally, the method for obtaining the query return fluctuation coefficient according to the communication transfer efficiency information is:

[0016] Collect the communication transfer efficiency information of multiple time periods T and integrate them into a data set, and use e as the number of the communication transfer efficiency information in the data set, and ,in, is a positive integer;

[0017] The default communication transfer efficiency limit is , the communication transfer efficiency information in the data set and the communication transfer efficiency limit Compare and mark the communication transfer efficiency information in the data set in the order of the data set. If the communication transfer efficiency information in the data set is lower than the communication transfer efficiency limit , then it is marked as 1 if the communication transfer efficiency information in the data set is higher than or equal to the communication transfer efficiency limit , then it is marked as 0, and a communication sequence consisting of 0 and 1 is obtained;

[0018] The number of consecutive 1s between any two 0s in the marked communication sequence is , where k is the number of consecutive 1s between any two 0s in the communication sequence, and , then the query returns the calculation expression of the fluctuation coefficient as , where Returns the volatility factor for the query.

[0019] Optionally, the method for obtaining concurrent scheduling management information is:

[0020] In the time period T, the number of concurrent requests between the cloud server and the local server is collected, and the number of concurrent requests between the cloud server and the local server in n time periods T is integrated into a concurrent data set in chronological order, and the number of concurrent requests between the cloud server and the local server in different time periods T is marked as , where r is the number of concurrent requests in different time periods T, and ,in is a positive integer;

[0021] In the time period T, the number of requests terminated between the cloud server and the local server is collected, and the number of requests terminated between the cloud server and the local server in n time periods T is integrated into the termination data set in chronological order, and the number of requests terminated between the cloud server and the local server in different time periods T is marked as , where r is the number of concurrent requests in different time periods T, and ,in is a positive integer;

[0022] According to the arrangement order of time period T, the data in the terminal data set of n time periods T are summed up with the data in the concurrent data set to obtain the concurrent scheduling management information. The calculation expression of the concurrent scheduling management information is: , where It is concurrent scheduling management information, where: represents the sequence number of n time periods T, and ,in, Is a positive integer.

[0023] Optionally, the method for obtaining the request strategy priority coefficient according to the concurrent scheduling management information is:

[0024] The preset resource capacity and concurrent request quantity is , the data in the concurrent data set generated in n time periods T are subtracted from the number of concurrent requests of resource capacity to obtain the excess concurrent number. The calculation expression of the excess concurrent number is: , according to the arrangement order of time periods T, generate a termination sequence according to the concurrent scheduling management information obtained within n time periods T, according to the arrangement order of time periods T, generate an excess sequence according to the excess concurrency obtained within n time periods T, bring each data in the termination sequence and each data in the corresponding excess sequence into the preset formula, obtain each fluctuation data, calculate the average value of the fluctuation data of n time periods T to obtain the balanced fluctuation data, calculate the difference between the balanced fluctuation data and the preset fluctuation threshold to obtain the request strategy priority coefficient.

[0025] Optionally, the query return fluctuation coefficient and the request strategy priority coefficient are used to construct an input floating model, and a method for generating a response fluctuation interference index is as follows:

[0026] After normalizing the query return fluctuation coefficient and the request strategy priority coefficient, the normalized query return fluctuation coefficient and the request strategy priority coefficient are brought into the input floating model. The expression of the input floating model is: , where The response fluctuation disturbance index generated for the input floating model, is the request strategy priority coefficient, are the ratio coefficients of the request strategy priority coefficient and the query return fluctuation coefficient, respectively, and All are positive numbers.

[0027] Optionally, the logic of performing signal classification based on the comparison between the generated response fluctuation interference index and the preset response fluctuation interference threshold is:

[0028] The preset response fluctuation interference threshold is , the generated response fluctuation interference index Interference threshold with response fluctuation For comparison, if the generated response fluctuation interference index Less than or equal to the response fluctuation interference threshold , a green signal is generated;

[0029] If the generated response fluctuation interference index Greater than the response fluctuation interference threshold , a yellow signal is generated.

[0030] Optionally, the logic for constructing a risk model based on the result of signal classification is:

[0031] When a yellow signal is generated, a data set is generated based on the integration of several consecutive response fluctuation interference index data of the intelligent customer service system within T time after the yellow signal is generated, and the response fluctuation interference index in the data set is calibrated as ,in is the response fluctuation interference index number, that is , where x is a positive integer;

[0032] Calculate the standard deviation of several response fluctuation interference indices in the data set, calibrate the standard deviation of the response fluctuation interference index as St, and compare the standard deviation of the response fluctuation interference index St with the preset response fluctuation interference index standard deviation threshold Of, and perform risk management based on the comparison results.

[0033] Optionally, the logic of risk management is as follows:

[0034] If St is greater than or equal to Of, the intelligent customer service system is marked as output fluctuation, and the staff is prompted to verify the server network.

[0035] If St is less than Of, the intelligent customer service system is marked as having stable output, prompting the staff that there is no high latency risk in the intelligent customer service system, and the server transmission data is monitored and recorded.

[0036] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0037] The present application collects the interaction information between the large language model deployed on the cloud server and the local knowledge base deployed on the local server, obtains the communication transfer efficiency information and the concurrent scheduling management information, obtains the query return fluctuation coefficient and the request strategy priority coefficient according to the communication transfer efficiency information and the concurrent scheduling management information, builds an input floating model according to the query return fluctuation coefficient and the request strategy priority coefficient, evaluates the communication efficiency between the cloud server and the local server and the impact of concurrent requests on the response of the intelligent customer service, effectively evaluates the communication transmission delay status between the cloud and the local, avoids the problem that the large language model cannot understand the context and respond in time due to excessive delay, and effectively improves the response timeliness under high load conditions through the evaluation of concurrent logic, further improves the data transmission efficiency between the large language model on the cloud and the local server, and ensures the stability of the large language model's understanding and reasoning ability at the time of output. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0039] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Example 1: Please refer to Figure 1 As shown, the present invention is a data interaction method for an intelligent customer service system that optimizes response timeliness, and the specific steps include:

[0042] Collect the communication transfer efficiency information and concurrent scheduling management information of the interaction between the large language model deployed on the cloud server and the local knowledge base deployed on the local server, obtain the query return fluctuation coefficient based on the communication transfer efficiency information, and obtain the request strategy priority coefficient based on the concurrent scheduling management information;

[0043] An input floating model is constructed based on the query return fluctuation coefficient and the request strategy priority coefficient to evaluate the communication efficiency between the cloud server and the local server and the impact of concurrent requests on the intelligent customer service response, and generate a response fluctuation interference index.

[0044] Comparing the generated response fluctuation interference index with a preset response fluctuation interference threshold, and classifying the signal according to the comparison result;

[0045] A risk model is constructed based on the results of signal classification, the fluctuations in the response timeliness of intelligent customer service are evaluated, and risk management is performed based on the evaluation results.

[0046] If the interaction response delay between the large language model deployed on the cloud server and the local knowledge base deployed on the local server is too high, the following adverse effects will occur:

[0047] The large language model interacts with the local knowledge base to provide accurate answers based on the user's query. When the interaction delay between the cloud server and the local server is too high, each query and data exchange takes longer, which directly increases the total time from the user's request to the response, resulting in a decrease in the response speed of the overall system.

[0048] The intelligent customer service system needs to handle multiple rounds of conversations, understand the context, and respond based on the content of the previous round of conversations. If each round of conversation requires interaction between the cloud and the local server, and the interaction delay is too high, the intelligent customer service system may not be able to process the context information in real time, making the multiple rounds of conversations unsmooth, affecting the intelligence and continuity of the system.

[0049] High latency may cause the system to fail to obtain necessary information from the previous round of conversation in a timely manner when processing user questions, resulting in confused answers, out of context, or even loss of key information, further affecting user experience;

[0050] In scenarios where the intelligent customer service system provides real-time feedback to customers, such as inquiries about bank accounts and payment status, the high latency between the cloud and local servers will reduce real-time performance and affect customers' needs for immediate response;

[0051] High latency affects the interactivity of the intelligent customer service system. The time interval between each user's question and the system's response is too long, resulting in a poor user interaction experience, which in turn causes users to give up interacting with the intelligent customer service system.

[0052] When the data of the local knowledge base is synchronized or updated from the cloud, the delay between the cloud and the local server will directly lead to the delay of data synchronization. The intelligent customer service system cannot obtain the latest knowledge base content in time, which affects the accuracy and timeliness of the answers provided;

[0053] For real-time data that requires rapid update and provision, such as account balances and order status, high latency will cause the intelligent customer service system to be unable to respond quickly to customer needs, resulting in information lag or inaccuracy, affecting the overall service quality of the intelligent customer service system;

[0054] The intelligent customer service system has the ability to dynamically adjust service strategies based on user feedback and interactions, such as adjusting the way of answering, switching question processing models, etc. If the interaction delay between the cloud and the local server is too high, the system may not be able to quickly adjust based on real-time data or user behavior, resulting in the adaptability and flexibility of intelligent customer service being limited.

[0055] In order to avoid the above adverse effects, the communication transfer efficiency information and concurrent scheduling management information of the interaction between the large language model deployed on the cloud server and the local knowledge base deployed on the local server are collected. The method for obtaining the communication transfer efficiency information is as follows:

[0056] Obtain the delay time for the cloud-based large language model to initiate interaction with the local knowledge base within the cycle time T, and mark the delay time for the cloud-based large language model to initiate interaction with the local knowledge base within the cycle time T as , where d is the number of the delay time for the large language model in the cloud to initiate interaction with the local knowledge base within the cycle time T, where ,and is a positive integer;

[0057] Get the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T, and mark the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T as , where b is the number of the delay time of the local knowledge base responding to the large language model in the cloud within the cycle time T, where ,and is a positive integer;

[0058] The communication transfer efficiency information is the discrete ratio of the delay time of the large language model on the cloud to initiate interaction with the local knowledge base and the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T. The calculation expression is: ,in, Transfer efficiency information for communication.

[0059] The method for obtaining the query return fluctuation coefficient based on the communication transfer efficiency information is:

[0060] Collect the communication transfer efficiency information of multiple time periods T and integrate them into a data set, and use e as the number of the communication transfer efficiency information in the data set, and ,in, is a positive integer;

[0061] The default communication transfer efficiency limit is , the communication transfer efficiency information in the data set and the communication transfer efficiency limit Compare and mark the communication transfer efficiency information in the data set in the order of the data set. If the communication transfer efficiency information in the data set is lower than the communication transfer efficiency limit , then it is marked as 1 if the communication transfer efficiency information in the data set is higher than or equal to the communication transfer efficiency limit , then it is marked as 0, and a communication sequence consisting of 0 and 1 is obtained;

[0062] The number of consecutive 1s between any two 0s in the marked communication sequence is , where k is the number of consecutive 1s between any two 0s in the communication sequence, and , then the query returns the calculation expression of the fluctuation coefficient as , where Returns the volatility factor for the query.

[0063] It should be noted that the delay time for the large language model on the cloud to initiate interaction with the local knowledge base within the cycle time T, and the delay time for the local knowledge base to respond to the large language model on the cloud within the cycle time T, are both obtained by recording the timestamps of key events through the logging tool in the intelligent customer service system. The start time is recorded when the large language model starts to process user input, and the query initiation time is recorded when the query request is initiated. The difference between the start time and the initiation time is calculated to obtain the time delay for the large language model to initiate the query request. The query time is obtained through the database query log of the local knowledge base, and the timing tool is used in the interface of the local knowledge base to record the time from the start of the query to the return of the result as the delay time for the local knowledge base to respond to the large language model on the cloud. The communication transfer efficiency limit is set by professional and technical personnel in this field according to the ability of the large language model to understand the input, which will not be elaborated here.

[0064] If the intelligent customer service system based on a large language model and a local knowledge base is inefficient in managing the queues of a large number of concurrent requests, the following adverse effects will occur:

[0065] When multiple concurrent requests enter the system, the intelligent customer service system cannot process each request quickly, causing the request to wait in the queue for too long, directly increasing the total response time of each request, causing users to wait longer to get a response;

[0066] If there is no effective load balancing or queuing strategy, some requests will over-occupy the resources of the intelligent customer service system, resulting in increased response time for other requests and forming a bottleneck.

[0067] In the case of poor queue management, high load or large concurrency, the intelligent customer service system will time out or lose some requests, which not only causes customers to fail to get service in time, but also requires repeated requests, further increasing the user's waiting time;

[0068] Inefficient queue management leads to excessive backlog of requests on the server, which over-consumes the resources of the intelligent customer service system, including CPU, memory, bandwidth, etc. As the number of requests increases, the responsiveness of the intelligent customer service system decreases, causing more requests to enter the queue, forming a vicious cycle;

[0069] Serious request backlogs may cause the context of multiple rounds of conversations to be lost. When the intelligent customer service system is waiting in line for request processing, it cannot obtain context information in time, resulting in the inability to accurately understand and process subsequent user input, affecting the coherence and intelligence of multiple rounds of conversations.

[0070] Long waiting times may cause the conversation to be interrupted or timed out, which will affect the intelligent customer service's ability to interact continuously and force the user to restart the conversation, seriously affecting the interactive experience.

[0071] In the case of low queue management efficiency, when the local knowledge base and the large language model need to be frequently synchronized, inefficient queue management will lead to data synchronization delays, so that the answers obtained by users are not the latest or most accurate, which will significantly affect the reliability of the intelligent customer service system in scenarios where real-time data needs to be queried;

[0072] Losses or errors may occur during request processing, resulting in some requests not being returned normally or inconsistent data being returned, affecting data accuracy and system stability;

[0073] Inefficient queue management leads to a backlog of requests, which makes monitoring data, including response time, processing time, queue length, etc., inaccurate. It is difficult for managers to distinguish whether performance problems are caused by excessive system load or improper queuing mechanism, which increases the complexity of system diagnosis and troubleshooting.

[0074] Due to request backlogs and delays, the health status of the intelligent customer service system may not be reflected in a timely manner, resulting in delayed discovery and resolution of problems. Inefficient queue management may make it difficult to identify system bottlenecks and problems in a timely manner under high load, affecting maintenance and optimization efficiency.

[0075] In order to avoid the above adverse effects, the communication transfer efficiency information and concurrent scheduling management information between the large language model deployed on the cloud server and the local knowledge base deployed on the local server are collected. The method for obtaining the concurrent scheduling management information is as follows:

[0076] In the time period T, the number of concurrent requests between the cloud server and the local server is collected, and the number of concurrent requests between the cloud server and the local server in n time periods T is integrated into a concurrent data set in chronological order, and the number of concurrent requests between the cloud server and the local server in different time periods T is marked as , where r is the number of concurrent requests in different time periods T, and ,in is a positive integer;

[0077] In the time period T, the number of requests terminated between the cloud server and the local server is collected, and the number of requests terminated between the cloud server and the local server in n time periods T is integrated into the termination data set in chronological order, and the number of requests terminated between the cloud server and the local server in different time periods T is marked as , where r is the number of concurrent requests in different time periods T, and ,in is a positive integer;

[0078] According to the arrangement order of time period T, the data in the terminal data set of n time periods T are summed up with the data in the concurrent data set to obtain the concurrent scheduling management information. The calculation expression of the concurrent scheduling management information is: , where It is concurrent scheduling management information, where: represents the sequence number of n time periods T, and ,in, Is a positive integer.

[0079] The method for obtaining the request strategy priority coefficient based on concurrent scheduling management information is:

[0080] The preset resource capacity and concurrent request quantity is , the data in the concurrent data set generated in n time periods T are subtracted from the number of concurrent requests of resource capacity to obtain the excess concurrent number. The calculation expression of the excess concurrent number is: , according to the arrangement order of time periods T, generate a termination sequence according to the concurrent scheduling management information obtained within n time periods T, according to the arrangement order of time periods T, generate an excess sequence according to the excess concurrency obtained within n time periods T, bring each data in the termination sequence and each data in the corresponding excess sequence into the preset formula, obtain each fluctuation data, calculate the average value of the fluctuation data of n time periods T to obtain the balanced fluctuation data, calculate the difference between the balanced fluctuation data and the preset fluctuation threshold to obtain the request strategy priority coefficient.

[0081] It should be noted that the preset formula is obtained by analyzing known historical data, based on the resource carrying capacity of the intelligent customer service system, the understanding and output capabilities of the large language model, and the data transmission efficiency between the large language model and the local knowledge base. It is obtained through time series analysis, deep learning and other methods. For example, the preset formula is expressed as , where For fluctuation data, are the weight coefficients of excess concurrency and concurrent scheduling management information, respectively, and are all positive numbers; the specific preset formula is set by professional and technical personnel in this field; the number of concurrent requests for resource capacity is the maximum number of concurrent requests that the resource carrying capacity of the intelligent customer service system can support, which is set by professional and technical personnel in this field according to the resource volume of the intelligent customer service system and is not elaborated here.

[0082] An input floating model is constructed based on the query return fluctuation coefficient and the request strategy priority coefficient to evaluate the communication efficiency between the cloud server and the local server and the impact of concurrent requests on the intelligent customer service response, and generate a response fluctuation interference index.

[0083] The method of constructing the input floating model with the query return fluctuation coefficient and the request strategy priority coefficient and generating the response fluctuation interference index is as follows:

[0084] After normalizing the query return fluctuation coefficient and the request strategy priority coefficient, the normalized query return fluctuation coefficient and the request strategy priority coefficient are brought into the input floating model. The expression of the input floating model is: , where The response fluctuation disturbance index generated for the input floating model, is the request strategy priority coefficient, are the ratio coefficients of the request strategy priority coefficient and the query return fluctuation coefficient, respectively, and All are positive numbers.

[0085] Comparing the generated response fluctuation interference index with a preset response fluctuation interference threshold, and classifying the signal according to the comparison result;

[0086] The logic of signal classification based on the comparison between the generated response fluctuation interference index and the preset response fluctuation interference threshold is:

[0087] The preset response fluctuation interference threshold is , the generated response fluctuation interference index Interference threshold with response fluctuation For comparison, if the generated response fluctuation interference index Less than or equal to the response fluctuation interference threshold , a green signal is generated;

[0088] If the generated response fluctuation interference index Greater than the response fluctuation interference threshold , a yellow signal is generated.

[0089] Build a risk model based on the results of signal classification, evaluate the fluctuation of the response timeliness of intelligent customer service, and perform risk management based on the evaluation results;

[0090] The logic of building a risk model based on the results of signal classification is:

[0091] When a yellow signal is generated, a data set is generated based on the integration of several consecutive response fluctuation interference index data of the intelligent customer service system within T time after the yellow signal is generated, and the response fluctuation interference index in the data set is calibrated as ,in is the response fluctuation interference index number, that is , where x is a positive integer;

[0092] Calculate the standard deviation of several response fluctuation interference indexes in the data set, and calibrate the standard deviation of the response fluctuation interference index as St, and compare the standard deviation of the response fluctuation interference index St with the preset response fluctuation interference index standard deviation threshold Of, and perform risk management based on the comparison results. The logic of risk management is as follows:

[0093] If St is greater than or equal to Of, the intelligent customer service system is marked as output fluctuation, and the staff is prompted to verify the server network.

[0094] If St is less than Of, the intelligent customer service system is marked as having stable output, prompting the staff that there is no high latency risk in the intelligent customer service system, and the server transmission data is monitored and recorded.

[0095] The present application collects the interaction information between the large language model deployed on the cloud server and the local knowledge base deployed on the local server, obtains the communication transfer efficiency information and the concurrent scheduling management information, obtains the query return fluctuation coefficient and the request strategy priority coefficient according to the communication transfer efficiency information and the concurrent scheduling management information, builds the input floating model according to the query return fluctuation coefficient and the request strategy priority coefficient, evaluates the communication efficiency between the cloud server and the local server and the impact of concurrent requests on the intelligent customer service response, effectively evaluates the communication transmission delay status between the cloud and the local, avoids the problem of large language model output deviation caused by excessive delay, provides technical personnel with an output monitoring method for the large language model, effectively identifies the state and connection between the output deviation and the delayed response, and effectively improves the response timeliness under high load conditions through the evaluation of concurrent logic, further improves the data transmission efficiency between the large language model on the cloud and the local server, and ensures the stability of the large language model's understanding and reasoning ability at the time of output.

[0096] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0097] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of computer program goods. The computer program goods include one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0098] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0099] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0100] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0101] If the functions are implemented in the form of software functional units and sold or used as independent goods, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of software goods, which are stored in a storage medium and include several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0102] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A data interaction method for an intelligent customer service system for optimizing response timeliness, characterized in that: The specific steps include: Collect the communication transfer efficiency information and concurrent scheduling management information of the interaction between the large language model deployed on the cloud server and the local knowledge base deployed on the local server, obtain the query return fluctuation coefficient based on the communication transfer efficiency information, and obtain the request strategy priority coefficient based on the concurrent scheduling management information; An input floating model is constructed based on the query return fluctuation coefficient and the request strategy priority coefficient to evaluate the communication efficiency between the cloud server and the local server and the impact of concurrent requests on the intelligent customer service response, and generate a response fluctuation interference index. The method of constructing the input floating model with the query return fluctuation coefficient and the request strategy priority coefficient and generating the response fluctuation interference index is as follows: After normalizing the query return fluctuation coefficient and the request strategy priority coefficient, the normalized query return fluctuation coefficient and the request strategy priority coefficient are brought into the input floating model. The expression of the input floating model is: , where Re s Response fluctuation disturbance index, Rs, generated for the input floating model pc is the request strategy priority coefficient, Qr fc is the query return fluctuation coefficient, β1 and β2 are the proportional coefficients of the request strategy priority coefficient and the query return fluctuation coefficient, and both β1 and β2 are positive numbers; Comparing the generated response fluctuation interference index with a preset response fluctuation interference threshold, and classifying the signal according to the comparison result; A risk model is constructed based on the results of signal classification, the fluctuations in the response timeliness of intelligent customer service are evaluated, and risk management is performed based on the evaluation results.

2. According to claim 1, a data interaction method for an intelligent customer service system for optimizing response timeliness is characterized in that: The method for obtaining communication transfer efficiency information is: Obtain the delay time for the cloud-based large language model to initiate interaction with the local knowledge base within the cycle time T, and mark the delay time for the cloud-based large language model to initiate interaction with the local knowledge base within the cycle time T as In d , where d is the number of the delay time for the large language model in the cloud to initiate interaction with the local knowledge base within the cycle time T, where d = {1, 2, 3…f}, and f is a positive integer; Get the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T, and mark the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T as Re b , where b is the number of the delay time of the local knowledge base responding to the large language model in the cloud within the cycle time T, where b = {1, 2, 3…v}, and v is a positive integer; The communication transfer efficiency information is the discrete ratio of the delay time of the large language model on the cloud to initiate interaction with the local knowledge base and the delay time of the local knowledge base responding to the large language model on the cloud within the cycle time T. The calculation expression is: , where Ct e Transfer efficiency information for communication.

3. According to claim 2, a data interaction method for an intelligent customer service system for optimizing response timeliness is characterized in that: The method for obtaining the query return fluctuation coefficient based on the communication transfer efficiency information is: Collect communication transfer efficiency information of multiple time periods T and integrate them into a data set, with e being the serial number of the communication transfer efficiency information in the data set, and e={1,2,3…w}, where w is a positive integer; The default communication transfer efficiency limit is Ct th , the communication transfer efficiency information in the data set and the communication transfer efficiency limit Ct th Compare and mark the communication transfer efficiency information in the data set in the order of the data set. If the communication transfer efficiency information in the data set is lower than the communication transfer efficiency limit Ct th , then it is marked as 1 if the communication transfer efficiency information in the data set is higher than or equal to the communication transfer efficiency limit Ct th , then it is marked as 0, and a communication sequence consisting of 0 and 1 is obtained; The number of consecutive 1s between any two 0s in the communication sequence is Pe k , where k is the number of consecutive 1s between any two 0s in the communication sequence, and k = {1, 2, 3…l}, then the calculation expression of the fluctuation coefficient returned by the query is , where Qr fc Returns the volatility factor for the query.

4. According to claim 1, a data interaction method for an intelligent customer service system for optimizing response timeliness is characterized in that: The method for obtaining concurrent scheduling management information is: In the time period T, the number of concurrent requests between the cloud server and the local server is collected, and the number of concurrent requests between the cloud server and the local server in n time periods T is integrated into a concurrent data set in chronological order, and the number of concurrent requests between the cloud server and the local server in different time periods T is marked as Co r , where r is the number of concurrent requests in different time periods T, and r = {1, 2, 3…t}, where t is a positive integer; In the time period T, the number of requests terminated between the cloud server and the local server is collected, and the number of requests terminated between the cloud server and the local server in n time periods T is integrated into the termination data set in chronological order, and the number of requests terminated between the cloud server and the local server in different time periods T is marked as Te r , where r is the number of concurrent requests in different time periods T, and r = {1, 2, 3…t}, where t is a positive integer; According to the arrangement order of time period T, the data in the terminal data set of n time periods T are summed up with the data in the concurrent data set to obtain the concurrent scheduling management information. The calculation expression of the concurrent scheduling management information is: , where Cs m is concurrent scheduling management information, where m represents the sequence number of n time periods T, and m={1,2,3…n}, where n is a positive integer.

5. According to claim 4, a data interaction method for an intelligent customer service system for optimizing response timeliness is characterized in that: The method for obtaining the request strategy priority coefficient based on concurrent scheduling management information is: The preset resource capacity concurrent request number is Rc cr , the data in the concurrent data set generated in n time periods T are subtracted from the number of concurrent requests of resource capacity to obtain the excess concurrent number. The calculation expression of the excess concurrent number is: , according to the arrangement order of time periods T, generate a termination sequence according to the concurrent scheduling management information obtained in n time periods T, according to the arrangement order of time periods T, generate an excess sequence according to the excess concurrency obtained in n time periods T, bring each data in the termination sequence and each data in the corresponding excess sequence into a preset formula, obtain each fluctuation data, calculate the average value of the fluctuation data of n time periods T to obtain the balanced fluctuation data, calculate the difference between the balanced fluctuation data and the preset fluctuation threshold to obtain the request strategy priority coefficient; The default formula is expressed as Fl u =α1*Ex c +α2*Cs m , where Fl u is the fluctuating data, α1 and α2 are the weight coefficients of excess concurrency and concurrent scheduling management information, respectively, and both α1 and α2 are positive numbers.

6. The method for data interaction of an intelligent customer service system for optimizing response timeliness according to claim 5, characterized in that: The logic of signal classification based on the comparison between the generated response fluctuation interference index and the preset response fluctuation interference threshold is: The preset response fluctuation interference threshold is Re th , the generated response fluctuation interference index Re s Interference threshold Re with response fluctuation th For comparison, if the generated response fluctuation interference index Re s Less than or equal to the response fluctuation interference threshold Re th , a green signal is generated; If the generated response fluctuation interference index Re s Greater than the response fluctuation interference threshold Re th , a yellow signal is generated.

7. The method for data interaction of an intelligent customer service system for optimizing response timeliness according to claim 6, characterized in that: The logic of building a risk model based on the results of signal classification is: When a yellow signal is generated, a data set is generated based on the integration of several consecutive response fluctuation interference index data of the intelligent customer service system within T time after the yellow signal is generated, and the response fluctuation interference index in the data set is calibrated as Rf i , where i is the response fluctuation interference index number, i = {1, 2, 3…x}, where x is a positive integer; Calculate the standard deviation of several response fluctuation interference indices in the data set, calibrate the standard deviation of the response fluctuation interference index as St, and compare the standard deviation of the response fluctuation interference index St with the preset response fluctuation interference index standard deviation threshold Of, and perform risk management based on the comparison results.

8. The method for data interaction of an intelligent customer service system for optimizing response timeliness according to claim 7, characterized in that: The logic of risk management is as follows: If St is greater than or equal to Of, the intelligent customer service system is marked as output fluctuation, and the staff is prompted to verify the server network. If St is less than Of, the intelligent customer service system is marked as having stable output, prompting the staff that there is no high latency risk in the intelligent customer service system, and the server transmission data is monitored and recorded.

Images