CN-122022825-A - Customer service data interaction method, system and equipment based on enterprise WeChat and large model

CN122022825ACN 122022825 ACN122022825 ACN 122022825ACN-122022825-A

Abstract

The invention discloses a customer service data interaction method, system and equipment based on enterprise WeChat and a large model, wherein an enterprise WeChat customer service access module acquires current enterprise WeChat information to be replied and sends the current enterprise WeChat information to an intelligent customer service center, the intelligent customer service center determines that the current enterprise WeChat information does not meet the condition of manual seat access, a current consultation type identification result is acquired based on an intention identification model and sent to a large model dialogue engine, the large model dialogue engine inputs the current enterprise WeChat information to be replied to the large language model according to the determination result, a corresponding current reply result is acquired from the target business system and sent to a user terminal, and the intelligent customer service center determines that the intelligent customer service center meets the condition of manual seat access and determines that a current target manual seat terminal is communicated. The embodiment of the invention can determine whether to continue intelligent conversation by using a large model conversation engine or access an artificial seat for timely information reply in real time after accurately identifying the communication intention of the current user in the intelligent customer service system.

Inventors

ZHANG XING
ZHONG LEI
FANG YUCHAO
LI XIONGCHENG
CHEN JIANHONG
ZHANG GE
ZHANG SHAOMIN

Assignees

云南花伍科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260213

Claims (10)

1. The customer service data interaction method based on the enterprise WeChat and the large model is applied to a customer service data interaction system based on the enterprise WeChat and the large model, and is characterized in that the customer service data interaction system based on the enterprise WeChat and the large model comprises an intelligent customer service center, an enterprise WeChat customer service access module, a large model dialogue engine and a service system access module, wherein the enterprise WeChat customer service access module and the large model dialogue engine are both in communication connection with the intelligent customer service center, the large model dialogue engine is also in communication connection with the service system access module, a pretrained large language model is deployed in the large model dialogue engine, and the service system access module is in communication connection with a plurality of external service systems based on a preset communication protocol: The enterprise WeChat customer service access module responds to an intelligent interaction request sent by a user terminal, acquires current enterprise WeChat information to be replied corresponding to the intelligent interaction request, and sends the information to the intelligent customer service center; If the intelligent customer service center determines that the current enterprise WeChat information to be replied does not meet the artificial seat access condition based on a pre-trained intention recognition model, acquiring a current consultation type recognition result corresponding to the current enterprise WeChat information to be replied based on the intention recognition model, and sending the current consultation type recognition result to the large model dialogue engine; the large model dialogue engine determines a target service system to be called currently according to the current consultation type identification result, correspondingly calls the target service system in a plurality of external service systems based on the service system access module, and takes the target service system as a current background knowledge base of the large model dialogue engine; The large model dialogue engine inputs the current enterprise WeChat information to be replied to the large language model to acquire current reply target service data from the target service system, performs semantic integration processing on the current reply target service data to acquire a current reply result, and sends the current reply result to the enterprise WeChat customer service access module; The enterprise WeChat customer service access module sends the current reply result to the user terminal; And if the intelligent customer service center determines that the current enterprise WeChat information to be replied meets the artificial seat access condition based on the pre-trained intention recognition model, acquiring a preset artificial seat allocation strategy and current artificial seat terminal state information, and determining a current target artificial seat terminal according to the current user information corresponding to the current enterprise WeChat information to be replied, the current artificial seat terminal state information and the artificial seat allocation strategy so as to enable the current target artificial seat terminal to establish communication connection with the user terminal.
2. The method according to claim 1, wherein, if the intelligent customer service center determines that the current to-be-replied enterprise WeChat information does not meet a manual agent access condition based on a pre-trained intent recognition model, acquiring a current consultation type recognition result corresponding to the current to-be-replied enterprise WeChat information based on the intent recognition model, and transmitting the current consultation type recognition result to the large model dialogue engine, further comprises: the intelligent service center determines a current communication emotion recognition result corresponding to the current enterprise WeChat information to be replied based on an emotion recognition sub-model in the intention recognition model; the intelligent service center station acquires preamble dialogue information belonging to the same current session with the current enterprise WeChat information to be replied and a preamble emotion recognition result set corresponding to the preamble dialogue information; If the intelligent service center determines that the current communication emotion recognition result belongs to any one emotion recognition result in a preset emotion recognition result set based on the intention recognition model, and each preceding emotion recognition result in the preceding emotion recognition result set does not belong to the preset emotion recognition result set, judging that the current enterprise WeChat information to be replied meets the artificial seat access condition; And if the intelligent service center determines that the current communication emotion recognition result does not belong to the preset emotion recognition result set based on the intention recognition model, and each of the preceding emotion recognition results in the preceding emotion recognition result set does not belong to the preset emotion recognition result set, judging that the current enterprise WeChat information to be replied does not meet the artificial seat access condition.
3. The method of claim 1, wherein the intent recognition sub-model in the intent recognition model comprises an input layer, an embedded layer, an encoding layer, a pooling layer, and a classification header connected in sequence; The obtaining, based on the intent recognition model, a current consultation type recognition result corresponding to the current enterprise WeChat information to be replied includes: Acquiring a current initial semantic vector corresponding to the current enterprise WeChat information to be replied based on an input layer, an embedding layer and an encoding layer of the intention recognition sub-model; acquiring a current semantic vector corresponding to the current initial semantic vector based on a pooling layer of the intention recognition sub-model; and acquiring the current consultation type recognition result corresponding to the current semantic vector based on the classification head of the intention recognition sub-model.
4. The method of claim 1, wherein the business data in each of the plurality of external business systems corresponds to business data of a cut-flower vending scenario and comprises at least a logistics business system, an order business system, and a commodity inventory business system; The large model dialogue engine determines a target service system to be called currently according to the current consultation type identification result, correspondingly calls the target service system in a plurality of external service systems based on the service system access module, and comprises the following steps: If the large model dialogue engine is based on the current consultation type identification result corresponding to the logistics information consultation identification result of the cut flowers, taking a logistics service system in a plurality of external service systems as the target service system, and correspondingly calling the logistics service system by a model context protocol based on the service system access module to establish communication connection; if the large model dialogue engine is based on the current consultation type recognition result corresponding to the fresh cut flower order information consultation recognition result, taking an order service system in a plurality of external service systems as the target service system, and correspondingly calling the order service system by a model context protocol based on the service system access module to establish communication connection; And if the large model dialogue engine is based on the current consultation type identification result corresponding to the commodity inventory information consultation identification result of the cut flowers, taking a commodity inventory service system in a plurality of external service systems as the target service system, and correspondingly calling the commodity inventory service system according to a model context protocol based on the service system access module to establish communication connection.
5. The method of claim 1, wherein the inputting the current enterprise WeChat information to be replied to the large language model to obtain current reply target service data from the target service system, and performing semantic integration processing on the current reply target service data to obtain a current reply result comprises: inputting the current enterprise WeChat information to be replied to the big language model to obtain a current retrieval condition; Acquiring corresponding current reply target service data from the target service system according to the current retrieval condition; And inputting the current reply target business data into the large language model for semantic integration and natural language generation to obtain the current reply result.
6. The method of claim 1, wherein the determining the current target artificial agent terminal according to the current user information corresponding to the current to-be-replied enterprise WeChat information, the current artificial agent terminal status information, and the artificial agent allocation policy comprises: Acquiring a current consultation type identification result corresponding to the current enterprise WeChat information to be replied, manual seat information of current user history docking communication in the current user information and cut flower type information purchased by the current user history, and forming an input text; acquiring a decision tree model corresponding to the artificial agent allocation strategy, and taking the artificial agent terminal belonging to the idle unoccupied state in the current artificial agent terminal state information as a plurality of selectable items of an output layer of the decision tree model; And inputting the input text into the decision tree model to obtain a current target artificial seat terminal from a plurality of selectable items of an output layer.
7. The method of claim 1, wherein the determining the current target artificial agent terminal according to the current user information corresponding to the current to-be-replied enterprise WeChat information, the current artificial agent terminal status information, and the artificial agent allocation policy comprises: acquiring current user information corresponding to WeChat information of a current enterprise to be replied, and acquiring a current agent busyness value corresponding to each agent terminal and a last-round artificial agent call end time point corresponding to the current user information in the current artificial agent terminal state information, wherein the current agent busyness value corresponding to the agent terminal = current user number corresponding to the total number of users to be treated, the current agent busyness value is normalized to a value, if the last-round artificial agent call end time point of the current user is not null, the last-round artificial agent call end time point corresponding to the current user information is obtained by each agent terminal = current user call time point corresponding to the current user information-last-round artificial agent call end time point, and if the last-round artificial agent call end time point of the current user is null, the last-round artificial agent call end time point corresponding to the current user information is obtained by each agent terminal = first preset interval time; The method comprises the steps of inputting a current agent busyness value corresponding to each agent terminal in current artificial agent terminal state information and a last-round artificial agent conversation interval duration value corresponding to the current user information of each agent terminal as input parameters into a weighted summation formula corresponding to an artificial agent allocation strategy, and determining an agent terminal access priority value corresponding to each agent terminal in the current artificial agent terminal state information, wherein the weighted summation formula corresponding to the artificial agent allocation strategy is an agent terminal access priority value = current agent busyness value, a first weight value + an agent terminal conversation interval duration value corresponding to the current user information of the last-round artificial agent conversation interval duration value, and a second weight value/a second preset interval duration value; And determining a minimum agent terminal access priority value from the agent terminal access priority values corresponding to the agent terminals in the current artificial agent terminal state information, and taking the agent terminal corresponding to the minimum agent terminal access priority value as the current target artificial agent terminal.
8. The customer service data interaction system based on the enterprise WeChat and the large model is characterized by comprising an intelligent customer service center, an enterprise WeChat customer service access module, a large model dialogue engine and a service system access module, wherein the enterprise WeChat customer service access module and the large model dialogue engine are both in communication connection with the intelligent customer service center; the enterprise WeChat customer service access module is used for responding to an intelligent interaction request sent by a user terminal, acquiring current enterprise WeChat information to be replied corresponding to the intelligent interaction request, and sending the information to the intelligent customer service center; The intelligent customer service center is used for acquiring a current consultation type identification result corresponding to the current enterprise WeChat information to be replied based on the intention identification model and sending the current consultation type identification result to the large model dialogue engine if the current enterprise WeChat information to be replied is determined to not meet the artificial seat access condition based on the pre-trained intention identification model; the large model dialogue engine is used for determining a target service system to be called currently according to the current consultation type identification result, correspondingly calling the target service system in a plurality of external service systems based on the service system access module, and taking the target service system as a current background knowledge base of the large model dialogue engine; The large model dialogue engine is further used for inputting the current enterprise WeChat information to be replied to the large language model so as to acquire current reply target service data from the target service system, carrying out semantic integration processing on the current reply target service data to acquire a current reply result, and sending the current reply result to the enterprise WeChat customer service access module; the enterprise WeChat customer service access module is also used for sending the current reply result to the user terminal; The intelligent customer service center is further configured to acquire a preset artificial agent allocation policy and current artificial agent terminal state information if the current enterprise WeChat information to be replied meets an artificial agent access condition based on a pre-trained intention recognition model, and determine a current target artificial agent terminal according to current user information corresponding to the current enterprise WeChat information to be replied, the current artificial agent terminal state information and the artificial agent allocation policy, so that the current target artificial agent terminal and the user terminal establish communication connection.
9. A computer device, characterized in that the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the customer service data interaction method based on enterprise WeChat and large model according to any one of claims 1-7 when executing the computer program.
10. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program comprising program instructions which, when executed by a processor, can implement the enterprise WeChat and Large model based customer service data interaction method according to any of claims 1-7.

Description

Customer service data interaction method, system and equipment based on enterprise WeChat and large model Technical Field The invention relates to the technical field of intelligent customer service, in particular to a customer service data interaction method, system and equipment based on enterprise WeChat and large model. Background With the development of electronic commerce and online services, enterprises generally provide consultation, ordering, logistics inquiry and after-sale services for users through an online customer service system. At present, the common customer service technical scheme mainly comprises an artificial seat customer service system, a rule type intelligent customer service system and a single question-answering robot system, wherein the artificial seat customer service system receives user information through enterprise micro (enterprise WeChat) and carries out manual answer, the rule type intelligent customer service system realizes automatic answer based on keyword matching or a fixed rule base and lacks context understanding capability, and the single question-answering robot system carries out question-answering based on a natural language processing model, but is generally limited to a static knowledge base and does not have real-time service data access capability. When the customer service system adopts the system of the prior art, the following disadvantages exist: 1) The intelligent customer service adopting the simple artificial intelligence technology has limited understanding capability, is difficult to accurately identify the real intention of the user, can not be timely determined to continue to automatically reply according to the real intention of the user, or can not be timely switched to the artificial customer service, and can not ensure the instantaneity of the user for acquiring the required accurate information; 2) The service system and the model have high coupling degree, the model needs to be redeveloped or trained every time one service capability is added, the expansion cost is high, for example, a logistics system, an order system, a commodity inventory system and the like are connected into a customer service system and are matched with an artificial intelligent model in the customer service system, and the artificial intelligent model in the customer service system needs to be redeveloped and trained every time the artificial intelligent model is matched with one service system; 3) Service system data such as regional logistics aging, order status, order failure reasons and the like cannot be obtained in real time, and only templated replies can be returned. Disclosure of Invention The embodiment of the invention provides a customer service data interaction method, system and equipment based on enterprise WeChat and large model, which aim to solve the problems that in the prior art, the intelligent customer service understanding capability adopting a simple artificial intelligent technology is limited, the real intention of a user is difficult to accurately identify, automatic reply cannot be continuously and timely determined according to the real intention of the user, or the customer service is timely switched to the artificial customer service, and the instantaneity of acquiring required accurate information cannot be ensured. In a first aspect, the embodiment of the invention provides a customer service data interaction method based on enterprise WeChat and large model, which is applied to a customer service data interaction system based on enterprise WeChat and large model, wherein the customer service data interaction system based on enterprise WeChat and large model comprises an intelligent customer service center, an enterprise WeChat customer service access module, a large model dialogue engine and a service system access module, the enterprise WeChat customer service access module and the large model dialogue engine are both in communication connection with the intelligent customer service center, the large model dialogue engine is also in communication connection with the service system access module, and a pre-trained large language model is deployed in the large model dialogue engine, and the service system access module is in communication connection with a plurality of external service systems based on a preset communication protocol, the method comprises the following steps: The enterprise WeChat customer service access module responds to an intelligent interaction request sent by a user terminal, acquires current enterprise WeChat information to be replied corresponding to the intelligent interaction request, and sends the information to the intelligent customer service center; If the intelligent customer service center determines that the current enterprise WeChat information to be replied does not meet the artificial seat access condition based on a pre-trained intention recognition model, acquiring a current consultation type rec