Customer service interaction method and system based on large language model

Abstract

The invention is suitable for the technical field of customer service interaction, and particularly relates to a customer service interaction method and system based on a large language model, and the method comprises the steps: constructing a general data database, building voice connection with a user, carrying out the information recording of a voice interaction process, and obtaining initial voice interaction data; acquiring user account information, performing data query based on the user account information, and constructing a user information database; processing the initial voice interaction data to obtain an initial voice processing result, and performing switching early warning detection to obtain an early warning detection result; and carrying out shunting processing, and synchronizing to corresponding service equipment to complete non-inductive switching of the real-time voice. According to the method, the voice interaction process is input into the large language model in a text mode, whether the user needs to switch the service specialist or not is predicted based on the interaction process with the user, and therefore when it is detected that switching is needed, preparation is started, and seamless switching of the service specialist is achieved when the demand changes.

Description

Technical Field

[0001] The present invention belongs to the technical field of customer service interaction, and in particular, relates to a customer service interaction method and system based on a large language model. Background Art

[0002] Customer service interactions refer to the process of communication between a business and its customers through various communication channels, with the aim of solving customer problems, providing product or service information, collecting feedback, or improving customer experience.

[0003] Whether it is enterprise customer service or e-commerce customer service, when the number of business types required by customer service is huge, enterprises or merchants need to set up different service specialists to serve callers with different needs. In order to efficiently respond to a large number of customer consultation needs, enterprises or e-commerce generally use customer service robots to respond. Customer service robots can respond to customers' simple consultation requests. Since customer service robots are prone to communication difficulties, many customers are unwilling to talk to customer service robots. After some customers interact with customer service robots, customer service robots find it difficult to understand customers' demands, which will cause customer complaints. Therefore, although customer service robots have been developed for a long time, manual customer service is still indispensable. Manual customer service is different from customer service robots. Although using manual customer service can provide a better service experience, its cost is relatively higher. In the process of manual customer service, manual customer service often needs to switch service specialists many times due to various reasons. Each time a service specialist is switched, the new service specialist needs to read and understand the historical conversation records, which will not only waste a lot of time, but also reduce the reception efficiency of manual customer service. Summary of the invention

[0004] The purpose of the present invention is to provide a customer service interaction method based on a large language model, aiming to solve the problem that in the process of manual customer service, due to various reasons, the manual customer service often needs to switch service specialists multiple times. Each time the service specialist is switched, the new service specialist needs to read and understand the historical conversation records, which not only wastes a lot of time, but also reduces the reception efficiency of the manual customer service.

[0005] The present invention is implemented as follows: a customer service interaction method based on a large language model, the method comprising:

[0006] Constructing a general information database, establishing a voice connection with the user, recording information about the voice interaction process, and obtaining initial voice interaction data, wherein the general information database records at least service scope data of each service specialist, and the service scope data is characterized by keywords;

[0007] Acquire user account information, perform data query based on the user account information, and build a user profile database according to the data query results, wherein the user profile database includes at least historical order information;

[0008] The initial voice interaction data is processed by a large language model to obtain an initial voice processing result, and a switching warning detection is performed based on the initial voice processing result to obtain a warning detection result;

[0009] Based on the early warning detection results, the real-time voice interaction records are diverted and synchronized to the corresponding service equipment. After receiving the switching instructions, the real-time voice is seamlessly switched. During the real-time voice conversation with the user, the user profile database is processed through the large language model to provide pre-loaded data to the service specialist.

[0010] Preferably, the steps of obtaining user account information, performing data query based on the user account information, and building a user profile database according to the data query results specifically include:

[0011] Perform data query based on the user's incoming call information to obtain the user's account information;

[0012] Searching the general data database based on the user account information to extract the user's historical orders, wherein the historical orders include completed orders and orders in progress;

[0013] The predicted demand of users for each historical order is determined based on the operations allowed to be performed on completed orders and orders in process, and the corresponding order information is retrieved based on the predicted demand to build a user information database.

[0014] Preferably, the step of processing the initial voice interaction data by using a large language model to obtain an initial voice processing result, performing switching warning detection according to the initial voice processing result, and obtaining a warning detection result specifically includes:

[0015] The initial voice interaction data is converted into initial text data through a voice conversion tool, and the initial text data is imported into a large language model to generate initial voice processing results;

[0016] The initial speech processing result is copied by a large language model to generate a plurality of copied speech texts, and keywords are extracted from the copied speech texts to obtain a plurality of keywords, wherein the keywords include event description keywords, event feature keywords, and event requirement keywords;

[0017] Match the overlap with each service range data, determine the service range data that best suits the current user, and generate early warning detection results.

[0018] Preferably, the steps of performing shunting processing on the real-time voice interaction record based on the early warning detection result, synchronizing to the corresponding service device, receiving the switching instruction, and completing the step of seamless switching of the real-time voice specifically include:

[0019] Determine the corresponding service scope based on the early warning detection results, randomly select an idle service device, and synchronously import a copy of the real-time voice interaction record to the corresponding service device;

[0020] receiving readiness feedback information sent from the service device, and sending a switching instruction to the service device based on the readiness feedback information;

[0021] After the switching instruction is sent, the idle service device performs voice interaction with the user according to the switching instruction, and disconnects the voice connection between the current service device and the user.

[0022] Preferably, in the process of real-time voice communication with the user, the user profile database is processed by a large language model to provide the service specialist with the step of pre-loading data, including: recording real-time interactive voice, converting the real-time interactive voice into real-time interactive text, analyzing the real-time interactive text by a large language model, obtaining text features mentioned in the real-time interactive text, searching the user profile database based on the text features, extracting corresponding prompt information, completing pre-loading of data based on the prompt information, obtaining pre-loaded data, and displaying the pre-loaded data to the service specialist.

[0023] Another object of the present invention is to provide a customer service interaction system based on a large language model, the system comprising:

[0024] A voice connection module is used to build a general information database, establish a voice connection with the user, record information about the voice interaction process, and obtain initial voice interaction data. The general information database records at least service scope data of each service specialist, and the service scope data is characterized by keywords.

[0025] A database construction module, used to obtain user account information, perform data query based on the user account information, and construct a user profile database according to the data query results, wherein the user profile database includes at least historical order information;

[0026] A switching prediction module is used to process the initial voice interaction data through a large language model to obtain an initial voice processing result, and to perform switching warning detection based on the initial voice processing result to obtain a warning detection result;

[0027] The seamless switching module is used to divert the real-time voice interaction records based on the early warning detection results, synchronize them to the corresponding service equipment, receive switching instructions, and complete the seamless switching of real-time voice. During the real-time voice conversation with the user, the user profile database is processed through the large language model to provide pre-loaded data to the service specialist.

[0028] Preferably, the database construction module includes:

[0029] An account information query unit, used to query data based on the user's incoming call information to obtain the user's account information;

[0030] An order extraction unit, used to search the general data database based on the user account information and extract the user's historical orders, wherein the historical orders include completed orders and orders in progress;

[0031] The data retrieval unit is used to determine the user's predicted demand for each historical order based on the operations allowed to be executed by the completed orders and the orders in process, retrieve the corresponding order data based on the predicted demand, and build a user data database.

[0032] Preferably, the switching prediction module includes:

[0033] A speech processing unit, used to convert the initial speech interaction data into initial text data through a speech conversion tool, import the initial text data into a large language model, and generate an initial speech processing result;

[0034] A text duplication unit, used to duplicate the initial speech processing result through a large language model to generate multiple duplicate speech texts, extract keywords from the duplicate speech texts to obtain multiple keywords, wherein the keywords include event description keywords, event feature keywords, and event requirement keywords;

[0035] The switching warning unit is used to match the overlap with each service range data, determine the service range data that best suits the current user, and generate a warning detection result.

[0036] Preferably, the senseless switching module comprises:

[0037] A device selection unit, used to determine the corresponding service range based on the early warning detection result, randomly select an idle service device, and synchronously import a copy of the real-time voice interaction record to the corresponding service device;

[0038] An instruction transceiver unit, configured to receive readiness feedback information sent from a service device, and send a switching instruction to the service device based on the readiness feedback information;

[0039] The voice switching unit is used to, after the switching instruction is sent, enable the idle service device to perform voice interaction with the user according to the switching instruction and disconnect the voice connection between the current service device and the user.

[0040] Preferably, in the process of real-time voice communication with the user, the user profile database is processed by a large language model to provide the service specialist with the step of pre-loading data, including: recording real-time interactive voice, converting the real-time interactive voice into real-time interactive text, analyzing the real-time interactive text by a large language model, obtaining text features mentioned in the real-time interactive text, searching the user profile database based on the text features, extracting corresponding prompt information, completing pre-loading of data based on the prompt information, obtaining pre-loaded data, and displaying the pre-loaded data to the service specialist.

[0041] The present invention provides a customer service interaction method based on a large language model. By recording the user's voice process and inputting the voice interaction process into the large language model in the form of text, the user's needs are analyzed, and whether the user needs to switch service specialists is predicted based on the interaction process with the user. When the need for switching is detected, preparations are started, and when the needs change, seamless switching of service specialists is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of a customer service interaction method based on a large language model provided by an embodiment of the present invention;

[0043] Figure 2 A flowchart of the steps of obtaining user account information, performing data query based on the user account information, and building a user profile database according to the data query results provided by an embodiment of the present invention;

[0044] Figure 3 A flowchart of the steps of processing initial voice interaction data through a large language model to obtain an initial voice processing result, performing switching warning detection according to the initial voice processing result, and obtaining a warning detection result provided in an embodiment of the present invention;

[0045] Figure 4 A flowchart of the steps of performing shunting processing on real-time voice interaction records based on early warning detection results, synchronizing to corresponding service devices, receiving switching instructions, and completing seamless switching of real-time voice provided by an embodiment of the present invention;

[0046] Figure 5 An architecture diagram of a customer service interaction system based on a large language model provided in an embodiment of the present invention;

[0047] Figure 6 An architectural diagram of a database construction module provided by an embodiment of the present invention;

[0048] Figure 7 An architecture diagram of a switching prediction module provided by an embodiment of the present invention;

[0049] Figure 8 An architectural diagram of a senseless switching module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0052] like Figure 1 As shown, it is a flowchart of a customer service interaction method based on a large language model provided by an embodiment of the present invention, and the method includes:

[0053] S100, constructing a general information database, establishing a voice connection with the user, recording information of the voice interaction process, and obtaining initial voice interaction data, wherein the general information database records at least service scope data of each service specialist, and the service scope data is characterized by keywords.

[0054] In this step, a general information database is constructed. Whether it is an enterprise or a merchant, a large amount of information related to the enterprise or merchant will be prepared. Taking network operators as an example, network operators have many service items and types, so the purposes of calls from different users vary greatly, and it is necessary to set up service specialists of various business types, such as voice package modules, network broadband modules, product sales modules, and procurement and installation modules. Different service specialists correspond to different service types. After the user accesses the phone, the business segment corresponding to the service specialist connected to the current user may not match the user's needs. At this time, switching is required. When there is a change in demand, such as user A needs to complete content consultation on multiple business segments during a call, it is necessary to switch service specialists multiple times. In the existing service process, during the personnel switching process, the user needs to wait in order to access a new service specialist. The new service specialist needs to re-understand the previous conversation information, so it may be necessary to repeatedly ask the user the same question. The problem will affect the user experience. In the present invention, all the required information is stored in a general information database, and a large language model is constructed. The large language model can analyze the information input by the user, so as to extract the keywords in the voice record between the user and the service specialist. Then, when the user establishes a voice connection with the first service specialist, the voice interaction process between the two is recorded to obtain initial voice interaction data. The general information database records at least the service scope data of each service specialist, and the service scope data is characterized by keywords. For example, a type A service specialist is used to serve customers in the network broadband sector, and its keywords may include: network speed, broadband, upstream bandwidth, downstream bandwidth, freeze, access equipment, delay, downtime and other keywords. When the chat record between the user and the service specialist involves the above keywords, it means that the user may need to access the service specialist of the corresponding service sector, that is, each service scope data is used to determine the service scope of each service specialist.

[0055] S200, obtaining user account information, performing data query based on the user account information, and constructing a user profile database according to the data query result, wherein the user profile database at least includes historical order information.

[0056] In this step, the user account information is obtained. When the user calls, the user's caller number is obtained. The user data can be retrieved based on the caller number. Each number corresponds to a user. After the user is determined, the orders generated by the user in the historical process can be retrieved, and the number of orders related to the user can be stored independently to create a temporary user information database. Then, when performing data query, you only need to search the temporary user information database, which reduces the time of data query and improves the response speed.

[0057] S300, processing the initial voice interaction data through a large language model to obtain an initial voice processing result, performing a switching warning detection based on the initial voice processing result to obtain a warning detection result.

[0058] In this step, the initial voice interaction data is processed by the large language model. The initial voice interaction data stores the voice interaction records between the user and the service specialist. The voice interaction records are converted into a text conversation format by a voice conversion tool, and the converted text is imported into the large language model. The large language model is used to analyze the conversation records between the two, so as to extract keywords and obtain the initial voice processing results. According to the matching relationship between the initial voice processing results and the service scope data corresponding to each service specialist, it is determined whether there is a trend of change in the user's needs, thereby generating an early warning detection result.

[0059] S400, based on the early warning detection results, diverts the real-time voice interaction records, synchronizes them to the corresponding service equipment, receives switching instructions, and completes the seamless switching of real-time voice. During the real-time voice conversation with the user, the user information database is processed through a large language model to provide pre-loaded data to the service specialist.

[0060] In this step, the real-time voice interaction record is diverted based on the early warning detection result. When the early warning detection result shows that the current user may need to switch services, a copy of the conversation record between the current service specialist and the user is imported into the service device of the new service specialist, and the previous conversation is processed through a large language model to generate a conversation summary, which is displayed to the new service specialist. The new service specialist can understand the interaction information between the user and the previous service specialist based on the conversation summary. After the new service specialist is ready, feedback information is sent to the service device of the previous service specialist. At this time, the previous service specialist can complete the switching of the service device at any time. When switching, the user's voice information is directly forwarded to the service device where the new service specialist is located, and the connection between the previous service specialist and the user is disconnected, realizing a seamless switching, improving the user experience, and improving the reception efficiency.

[0061] like Figure 2 As shown, as a preferred embodiment of the present invention, the steps of obtaining user account information, performing data query based on the user account information, and building a user profile database according to the data query results specifically include:

[0062] S201, performing data query based on the user's incoming call information to obtain the user's account information.

[0063] In this step, data query is performed based on the user's incoming call information, and the incoming call information includes the user's calling number. By querying based on the calling number, the user's identity, including the user's name and gender, etc., can be further determined.

[0064] S202, searching the general information database based on the user account information to extract the user's historical orders, wherein the historical orders include completed orders and orders in progress.

[0065] In this step, a search is performed in the general information database based on the user account information, and all data related to the user stored in the general information database is retrieved based on the user account information, including historical orders. Historical orders are divided into completed orders and processing orders. For completed orders, the main body of the order has been completed, such as a user applying for broadband installation, and the installation service has been completed, and the user is in the process of normal use. Processing orders refer to orders that the user has submitted an order application but have not yet been executed.

[0066] S203, determining the predicted demand of the user for each historical order according to the operations allowed to be executed by the completed orders and the orders in process, retrieving the corresponding order information based on the predicted demand, and constructing a user information database.

[0067] In this step, the user's predicted demand for each historical order is determined based on the operations allowed for completed orders and orders in progress. Different types of orders have different allowed operations. For example, user A purchases product B, the order has been completed, and the return and exchange period has expired. At this time, the order only allows the user to issue invoices or perform after-sales operations for quality issues. For ongoing orders, exchange operations, return operations, and product consultation operations can be performed. The above-mentioned allowed operations are stored in the user profile database as the predicted demand for each historical order of the user. When the user mentions the corresponding order, the service scope can be narrowed.

[0068] like Figure 3 As shown, as a preferred embodiment of the present invention, the steps of processing the initial voice interaction data by a large language model to obtain an initial voice processing result, performing switching warning detection according to the initial voice processing result, and obtaining a warning detection result specifically include:

[0069] S301, converting the initial voice interaction data into initial text data through a voice conversion tool, importing the initial text data into a large language model, and generating an initial voice processing result.

[0070] In this step, the initial voice interaction data is converted into initial text data through a voice conversion tool. The voice conversion tool is a voice recognition engine. The conversation voice between the user and the service specialist is input into it, and the corresponding conversation text can be output, which is the initial voice processing result.

[0071] S302, the initial speech processing result is copied through a large language model to generate multiple copied speech texts, and keywords are extracted from the copied speech texts to obtain multiple keywords, wherein the keywords include event description keywords, event feature keywords, and event requirement keywords.

[0072] In this step, the initial speech processing result is copied by the large language model. The large language model has the ability to copy and can convert the same text into multiple communication methods with the same topic, such as "I have a broadband in your house. It was originally said to have a bandwidth of 300M, but it has been very unstable in the past two days. It was not stuck before, but now it often freezes." After copying, it can be "I have a network in your house. Recently, the network has fluctuated greatly, and the network speed will drop from time to time. The bandwidth is 300M, but now it is often not reached." After multiple copies, the number of texts is increased, and multiple copied voice texts are obtained. The keywords recorded in the copied voice text are extracted by the large language model, and the keywords are classified, including event description keywords, event feature keywords, and event demand keywords. Among them, the event description keywords are used to determine the nature of the event, such as broadband, mobile network, bandwidth, etc. The event feature keywords are specific features about the event, such as network freeze, high latency, slow network speed, etc. The event demand keywords are the needs put forward by the user for the event, such as on-site repair, increase bandwidth, unsubscribe, etc.

[0073] S303, performing overlap matching with each service range data, determining the service range data most suitable for the current user, and generating an early warning detection result.

[0074] In this step, the overlap is matched with each service scope data. A service specialist corresponds to one service scope data. The number of times each keyword appears in each service scope data is determined by extracting the event description keywords, event feature keywords and event demand keywords. Specifically, a weight is assigned to each keyword to calculate the number of keywords falling into each service scope data. For example, the weight of the event description keyword is 0.2, the weight of the event feature keyword is 0.3, and the weight of the event demand keyword is 0.5. If an event description keyword falls into the corresponding service scope data, it is regarded as the keyword appearing 0.2 times in the service scope data. After the statistics are completed, the number of times the keyword corresponding to each service scope data appears is determined, and the number ratio is calculated. If there are three service scope data, they are A, B, C, and D respectively. , B and C, in the current conversation process, the number of keywords that fall into A's service scope data is P1, the number of keywords that fall into B's service scope data is P2, and the number of keywords that fall into C's service scope data is P3. Then the corresponding number ratios of A, B and C are P1 / (P1+P2+P3), P2 / (P1+P2+P3), and P3 / (P1+P2+P3). When the number ratio is greater than the preset value, it is determined that there is a service demand that falls into the corresponding service scope data. For example, the preset value is 0.7, when P3 / (P1+P2+P3), it means that the user's current demand is within the C service scope. If the user is currently communicating with the service specialist corresponding to the C service scope data, there is no switching demand. If the user is not currently communicating with the service specialist corresponding to the C service scope data, it is determined that there is a switching demand.

[0075] like Figure 4 As shown, as a preferred embodiment of the present invention, the steps of shunting the real-time voice interaction record based on the early warning detection result, synchronizing it to the corresponding service device, receiving the switching instruction, and completing the seamless switching of the real-time voice specifically include:

[0076] S401, determine the corresponding service scope based on the early warning detection result, randomly select an idle service device, and synchronously import a copy of the real-time voice interaction record to the corresponding service device.

[0077] In this step, the corresponding service scope is determined based on the early warning detection result. If the early warning detection result determines that the current user's demand falls from the M service scope to the N service scope, then the idle service personnel within the N service scope are queried, and an idle service personnel is randomly selected. A copy of the real-time voice interaction record is synchronously imported into the corresponding service device. At this time, the service personnel corresponding to the service device can directly hear the conversation between the current service personnel and the user, and process the previous conversation through the large language model to generate a conversation summary, which is displayed to the new service specialist.

[0078] S402: Receive readiness feedback information sent from the service device, and send a switching instruction to the service device based on the readiness feedback information.

[0079] S403: After the switching instruction is sent, the idle service device performs voice interaction with the user according to the switching instruction, and disconnects the voice connection between the current service device and the user.

[0080] In this step, the readiness feedback information sent from the service device is received. When the new service specialist fully understands the previous conversation information, the new service specialist sends the readiness feedback information to the service device corresponding to the current service specialist. At this time, the service specialist currently connected to the user can switch the service at any time. When it is determined to switch the service, the idle service device performs voice interaction with the user according to the switching instruction, and disconnects the voice connection between the current service device and the user; the above switching process can be repeated many times. During the service of the service specialist, the real-time interactive voice is converted into real-time interactive text, the real-time interactive text is analyzed by a large language model, the text features mentioned in the real-time interactive text are obtained, the user information database is searched based on the text features, the corresponding prompt information is extracted, the data is preloaded based on the prompt information, the preloaded data is obtained, and the preloaded data is displayed to the service specialist.

[0081] like Figure 5 As shown, a customer service interaction system based on a large language model is provided in an embodiment of the present invention, and the system includes:

[0082] The voice connection module 100 is used to build a general information database, establish a voice connection with the user, record information on the voice interaction process, and obtain initial voice interaction data. The general information database records at least the service scope data of each service specialist, and the service scope data is characterized by keywords.

[0083] In this system, the voice connection module 100 constructs a general information database. Whether it is an enterprise or a merchant, a large amount of information related to the enterprise or merchant will be prepared. Taking network operators as an example, network operators have many service items and types. Therefore, the purposes of calls from different users vary greatly, and it is necessary to set up service specialists of various business types, such as voice package modules, network broadband modules, product sales modules, and procurement and installation modules. Different service specialists correspond to different service types. After the user accesses the phone, the business sector corresponding to the service specialist connected to the current user may not match the user's needs. At this time, switching is required. When there is a change in demand, such as user A needs to complete content consultation on multiple business sectors during a call, it is necessary to switch service specialists multiple times. In the existing service process, during the personnel switching process, the user needs to wait to access a new service specialist. The new service specialist needs to re-understand the previous conversation information, so it may be necessary to repeat the user. Asking the same question will affect the user experience. In the present invention, all the required information is stored in a general information database, and a large language model is constructed. The large language model can analyze the information input by the user, so as to extract keywords in the voice record between the user and the service specialist. Then, when the user establishes a voice connection with the first service specialist, the voice interaction process between the two is recorded to obtain initial voice interaction data. The general information database records at least the service scope data of each service specialist, and the service scope data is characterized by keywords. For example, a type A service specialist is used to serve customers in the network broadband sector, and its keywords may include: network speed, broadband, upstream bandwidth, downstream bandwidth, freeze, access equipment, delay, downtime and other keywords. When the chat record between the user and the service specialist involves the above keywords, it means that the user may need to access the service specialist of the corresponding service sector, that is, each service scope data is used to determine the service scope of each service specialist.

[0084] The database construction module 200 is used to obtain user account information, perform data query based on the user account information, and construct a user profile database according to the data query results. The user profile database at least includes historical order information.

[0085] In this system, the database construction module 200 obtains user account information. When a user calls, the user's caller number is obtained. The user data can be retrieved based on the caller number. Each number corresponds to a user. After the user is determined, the orders generated by the user in the historical process can be retrieved, and the number of orders related to the user can be stored independently to create a temporary user information database. Then, when performing data query, only the temporary user information database needs to be retrieved, which reduces the time for data query and improves the response speed.

[0086] The switching prediction module 300 is used to process the initial voice interaction data through a large language model to obtain an initial voice processing result, and perform switching warning detection based on the initial voice processing result to obtain a warning detection result.

[0087] In this system, the switching prediction module 300 processes the initial voice interaction data through the large language model. The initial voice interaction data stores the voice interaction records between the user and the service specialist. The voice interaction records are converted into a text conversation format through a voice conversion tool, and the converted text is imported into the large language model. The large language model is used to analyze the conversation records between the two, so as to extract keywords and obtain the initial voice processing results. According to the matching relationship between the initial voice processing results and the service scope data corresponding to each service specialist, it is determined whether there is a trend of change in the user's needs, thereby generating an early warning detection result.

[0088] The seamless switching module 400 is used to divert the real-time voice interaction records based on the early warning detection results, synchronize them to the corresponding service equipment, receive switching instructions, and complete the seamless switching of real-time voice. During the real-time voice conversation with the user, the user information database is processed through the large language model to provide pre-loaded data to the service specialist.

[0089] In this system, the seamless switching module 400 performs diversion processing on the real-time voice interaction record based on the early warning detection result. When the early warning detection result shows that the current user may have the need to switch services, a copy of the conversation record between the current service specialist and the user is imported into the service device of the new service specialist, and the previous conversation is processed by the large language model to generate a conversation summary, which is displayed to the new service specialist. The new service specialist can understand the interaction information between the user and the previous service specialist based on the conversation summary. After the new service specialist is ready, feedback information is sent to the service device of the previous service specialist. At this time, the previous service specialist can complete the switching of the service device at any time. When switching, the user's voice information is directly forwarded to the service device where the new service specialist is located, and the connection between the previous service specialist and the user is disconnected, thereby realizing seamless switching, improving user experience, and improving reception efficiency.

[0090] like Figure 6 As shown, as a preferred embodiment of the present invention, the database construction module 200 includes:

[0091] The account information query unit 201 is used to perform data query based on the user's incoming call information to obtain the user's account information.

[0092] In this module, the account information query unit 201 performs data query based on the user's incoming call information, which includes the user's calling number. By querying based on the calling number, the user's identity, including the user's name and gender, etc., can be further determined.

[0093] The order extraction unit 202 is used to search the general information database based on the user account information and extract the historical orders of the user, wherein the historical orders include completed orders and orders in progress.

[0094] In this module, the order extraction unit 202 searches the general information database based on the user account information, and retrieves all data related to the user stored in the general information database based on the user account information, including historical orders. Historical orders are divided into completed orders and processing orders. For completed orders, the main body of the order has been completed, such as a user applying for broadband installation, and the installation service has been completed, and the user is in normal use. A processing order refers to an order for which the user has submitted an order application but has not yet been executed.

[0095] The data retrieval unit 203 is used to determine the user's predicted demand for each historical order based on the operations allowed to be executed by the completed orders and the orders in process, retrieve the corresponding order data based on the predicted demand, and build a user data database.

[0096] In this module, the data retrieval unit 203 determines the user's predicted demand for each historical order based on the operations allowed for completed orders and orders in progress. Different types of orders have different allowed operations. For example, if user A purchases product B, the order has been completed and the return and exchange period has expired, the order only allows the user to issue invoices or perform after-sales operations for quality issues. For ongoing orders, exchange operations, return operations, and product consultation operations can be performed. The above-mentioned allowed operations are stored in the user data database as the predicted demand for each historical order of the user. When the user mentions the corresponding order, the service scope can be narrowed.

[0097] like Figure 7 As shown, as a preferred embodiment of the present invention, the switching prediction module 300 includes:

[0098] The speech processing unit 301 is used to convert the initial speech interaction data into initial text data through a speech conversion tool, import the initial text data into a large language model, and generate an initial speech processing result.

[0099] In this module, the voice processing unit 301 converts the initial voice interaction data into initial text data through a voice conversion tool. The voice conversion tool is a voice recognition engine. By inputting the conversation voice between the user and the service specialist into it, the corresponding conversation text can be output, which is the initial voice processing result.

[0100] The text replication unit 302 is used to replicate the initial speech processing result through a large language model to generate multiple replicated speech texts, extract keywords from the replicated speech texts, and obtain multiple keywords, including event description keywords, event feature keywords, and event requirement keywords.

[0101] In this module, the text replication unit 302 replicates the initial speech processing result through the large language model. The large language model has replication capability and can convert the same text into multiple communication methods with the same theme, such as "I have a broadband in your house. It was originally said to have a bandwidth of 300M, but it has been very unstable in the past two days. It was not stuck before, but now it often freezes." After replication, it can be "I have a network in your house. Recently, the network has fluctuated greatly, and the network speed has dropped from time to time. The bandwidth is 300M, but now it is often not reached." After multiple replications, the number of texts is increased, and multiple replicated voice texts are obtained. The keywords recorded in the replicated voice texts are extracted through the large language model, and the keywords are classified, including event description keywords, event feature keywords, and event demand keywords. Among them, the event description keywords are used to determine the nature of the event, such as broadband, mobile network, bandwidth, etc. The event feature keywords are specific features about the event, such as network freeze, high latency, slow network speed, etc. The event demand keywords are the needs put forward by the user for the event, such as on-site maintenance, increased bandwidth, unsubscribe, etc.

[0102] The switching warning unit 303 is used to perform overlap matching with various service range data, determine the service range data most suitable for the current user, and generate a warning detection result.

[0103] In this module, the switching warning unit 303 performs overlap matching with each service scope data. Each service specialist corresponds to one service scope data. The number of times each keyword appears in each service scope data is determined by extracting the event description keywords, event feature keywords and event demand keywords. Specifically, a weight is assigned to each keyword to calculate the number of keywords falling into each service scope data. For example, the weight of the event description keyword is 0.2, the weight of the event feature keyword is 0.3, and the weight of the event demand keyword is 0.5. If an event description keyword falls into the corresponding service scope data, it is regarded as the keyword appearing 0.2 times in the service scope data. After the statistics are completed, the number of times the keyword corresponding to each service scope data appears is determined, and the number ratio is calculated. If there are three service scope data, They are A, B and C respectively. In the current conversation, the number of keywords that fall into the service scope data of A is P1, the number of keywords that fall into the service scope data of B is P2, and the number of keywords that fall into the service scope data of C is P3. Then the corresponding number ratios of A, B and C are P1 / (P1+P2+P3), P2 / (P1+P2+P3), and P3 / (P1+P2+P3). When the number ratio is greater than the preset value, it is determined that there is a service demand that falls into the corresponding service scope data. For example, the preset value is 0.7, when P3 / (P1+P2+P3), it means that the user's current demand is in the service scope of C. If the user is currently communicating with the service specialist corresponding to the service scope data of C, there is no switching demand. If the user is not currently communicating with the service specialist corresponding to the service scope data of C, it is determined that there is a switching demand.

[0104] like Figure 8 As shown, as a preferred embodiment of the present invention, the senseless switching module 400 includes:

[0105] The device selection unit 401 is used to determine the corresponding service range based on the early warning detection result, randomly select an idle service device, and synchronously import a copy of the real-time voice interaction record to the corresponding service device.

[0106] In this module, the device selection unit 401 determines the corresponding service scope based on the early warning detection result. If the early warning detection result determines that the current user's needs fall from the M service scope to the N service scope, then the idle service personnel within the N service scope are queried, and an idle service personnel is randomly selected. A copy of the real-time voice interaction record is synchronously imported into the corresponding service device. At this time, the service personnel corresponding to the service device can directly hear the conversation between the current service personnel and the user, and process the previous conversation through the large language model to generate a conversation summary, which is displayed to the new service specialist.

[0107] The instruction transceiver unit 402 is used to receive the readiness feedback information sent from the service device, and send a switching instruction to the service device based on the readiness feedback information.

[0108] The voice switching unit 403 is used to, after the switching instruction is sent, enable the idle service device to perform voice interaction with the user according to the switching instruction and disconnect the voice connection between the current service device and the user.

[0109] In this module, the readiness feedback information sent from the service device is received. When the new service specialist fully understands the previous conversation information, the new service specialist sends the readiness feedback information to the service device corresponding to the current service specialist. At this time, the service specialist currently connected to the user can switch the service at any time. When it is determined to switch the service, the idle service device interacts with the user by voice according to the switching instruction, and disconnects the voice connection between the current service device and the user. The above switching process can be repeated many times. During the service of the service specialist, the real-time interactive voice is converted into real-time interactive text, and the real-time interactive text is analyzed by a large language model to obtain the text features mentioned in the real-time interactive text. The user information database is searched based on the text features to extract the corresponding prompt information. The data is preloaded based on the prompt information to obtain the preloaded data, and the preloaded data is displayed to the service specialist.

[0110] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0111] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0112] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A customer service interaction method based on a large language model, characterized in that: The method comprises: Constructing a general information database, establishing a voice connection with the user, recording information about the voice interaction process, and obtaining initial voice interaction data, wherein the general information database records at least service scope data of each service specialist, and the service scope data is characterized by keywords; Acquire user account information, perform data query based on the user account information, and build a user profile database according to the data query results, wherein the user profile database includes at least historical order information; The initial voice interaction data is processed by a large language model to obtain an initial voice processing result, and a switching warning detection is performed based on the initial voice processing result to obtain a warning detection result; Based on the early warning detection results, the real-time voice interaction records are diverted and synchronized to the corresponding service equipment. After receiving the switching instructions, the real-time voice is seamlessly switched. During the real-time voice conversation with the user, the user profile database is processed through the large language model to provide pre-loaded data to the service specialist.

2. The customer service interaction method based on a large language model according to claim 1, characterized in that: The steps of obtaining user account information, performing data query based on the user account information, and building a user profile database according to the data query results specifically include: Perform data query based on the user's incoming call information to obtain the user's account information; Searching the general data database based on the user account information to extract the user's historical orders, wherein the historical orders include completed orders and orders in progress; The predicted demand of users for each historical order is determined based on the operations allowed to be performed on completed orders and orders in process, and the corresponding order information is retrieved based on the predicted demand to build a user information database.

3. The customer service interaction method based on a large language model according to claim 1, characterized in that: The steps of processing the initial voice interaction data by using the large language model to obtain the initial voice processing result, performing switching warning detection according to the initial voice processing result, and obtaining the warning detection result specifically include: The initial voice interaction data is converted into initial text data through a voice conversion tool, and the initial text data is imported into a large language model to generate initial voice processing results; The initial speech processing result is copied by a large language model to generate a plurality of copied speech texts, and keywords are extracted from the copied speech texts to obtain a plurality of keywords, wherein the keywords include event description keywords, event feature keywords, and event requirement keywords; Match the overlap with each service range data, determine the service range data that best suits the current user, and generate early warning detection results.

4. The customer service interaction method based on a large language model according to claim 1, characterized in that: The steps of performing shunting processing on the real-time voice interaction record based on the early warning detection result, synchronizing to the corresponding service device, receiving the switching instruction, and completing the seamless switching of the real-time voice specifically include: Determine the corresponding service scope based on the early warning detection results, randomly select an idle service device, and synchronously import a copy of the real-time voice interaction record to the corresponding service device; receiving readiness feedback information sent from the service device, and sending a switching instruction to the service device based on the readiness feedback information; After the switching instruction is sent, the idle service device performs voice interaction with the user according to the switching instruction, and disconnects the voice connection between the current service device and the user.

5. The customer service interaction method based on a large language model according to claim 1, characterized in that: The step of processing the user profile database through the large language model during the real-time voice communication with the user and providing the service specialist with preloaded data includes: recording the real-time interactive voice, converting the real-time interactive voice into real-time interactive text, analyzing the real-time interactive text through the large language model, obtaining text features mentioned in the real-time interactive text, searching the user profile database based on the text features, extracting corresponding prompt information, completing preloading of data based on the prompt information, obtaining preloaded data, and displaying the preloaded data to the service specialist.

6. A customer service interaction system based on a large language model, characterized in that: The system comprises: A voice connection module is used to build a general information database, establish a voice connection with the user, record information about the voice interaction process, and obtain initial voice interaction data. The general information database records at least service scope data of each service specialist, and the service scope data is characterized by keywords. A database construction module, used to obtain user account information, perform data query based on the user account information, and construct a user profile database according to the data query results, wherein the user profile database includes at least historical order information; A switching prediction module is used to process the initial voice interaction data through a large language model to obtain an initial voice processing result, and to perform switching warning detection based on the initial voice processing result to obtain a warning detection result; The seamless switching module is used to divert the real-time voice interaction records based on the early warning detection results, synchronize them to the corresponding service equipment, receive switching instructions, and complete the seamless switching of real-time voice. During the real-time voice conversation with the user, the user profile database is processed through the large language model to provide pre-loaded data to the service specialist.

7. The customer service interaction system based on a large language model according to claim 6, characterized in that: The database construction module includes: An account information query unit, used to query data based on the user's incoming call information to obtain the user's account information; An order extraction unit, used to search the general data database based on the user account information and extract the user's historical orders, wherein the historical orders include completed orders and orders in progress; The data retrieval unit is used to determine the user's predicted demand for each historical order based on the operations allowed to be executed by the completed orders and the orders in process, retrieve the corresponding order data based on the predicted demand, and build a user data database.

8. The customer service interaction system based on a large language model according to claim 6, characterized in that: The switching prediction module comprises: A speech processing unit, used to convert the initial speech interaction data into initial text data through a speech conversion tool, import the initial text data into a large language model, and generate an initial speech processing result; A text duplication unit, used to duplicate the initial speech processing result through a large language model to generate multiple duplicate speech texts, extract keywords from the duplicate speech texts to obtain multiple keywords, wherein the keywords include event description keywords, event feature keywords, and event requirement keywords; The switching warning unit is used to match the overlap with each service range data, determine the service range data that best suits the current user, and generate a warning detection result.

9. The customer service interaction system based on a large language model according to claim 6, characterized in that: The senseless switching module comprises: A device selection unit, used to determine the corresponding service range based on the early warning detection result, randomly select an idle service device, and synchronously import a copy of the real-time voice interaction record to the corresponding service device; An instruction transceiver unit, configured to receive readiness feedback information sent from a service device, and send a switching instruction to the service device based on the readiness feedback information; The voice switching unit is used to, after the switching instruction is sent, enable the idle service device to perform voice interaction with the user according to the switching instruction and disconnect the voice connection between the current service device and the user.

10. The customer service interaction system based on a large language model according to claim 6, characterized in that: In the process of real-time voice communication with the user, the user data database is processed by the large language model to provide the service specialist with pre-loaded data, including: recording the real-time interactive voice, converting the real-time interactive voice into real-time interactive text, analyzing the real-time interactive text by the large language model, and obtaining Take the text features mentioned in the real-time interactive text and search the user information database based on the text features. Extract the corresponding prompt information, complete the preloading of data based on the prompt information, and obtain the preloaded data. Display the preloaded data to the service specialist.

Images