CN116843347A - Micro-grid intelligent customer service system and construction method thereof - Google Patents

Abstract

The invention relates to a micro-grid intelligent customer service system and a construction method thereof, wherein the construction method comprises the following specific steps: step S1, collecting data text and constructing a data set; step S2, training and fine tuning the data set constructed in the step S1 through a chatGLM artificial intelligent language model to generate a training chatGLM model; step S3, adding a ChatGLM center module in a service capability layer of the intelligent customer service system, integrating a trained ChatGLM model in the module, and creating an interactive robot dialogue interface and constructing task multi-round dialogue; and S4, counting and analyzing the interaction process with the client, upgrading the data set, and optimizing and updating the intelligent customer service system. Compared with the prior art, the invention provides knowledge management technology and statistical analysis information for enterprises in the electric related industry; a quick and effective technical means is established for the communication between enterprises and users; the method has the advantages of generating a solution scheme wanted by a customer more accurately, relieving the pressure of manual customer service and the like.

Description

Micro-grid intelligent customer service system and construction method thereof

Technical Field

The invention relates to an intelligent customer service technology, in particular to a micro-grid intelligent customer service system based on ChatGLM and a construction method thereof.

Background

The online customer service is a page communication technology for providing instant communication between Internet visitors and staff in the website by taking the website as a medium. With the development of internet technology, the demand of online customer service is gradually increased, and simultaneously, higher demands are also put on the quality of online customer service.

The traditional online customer service system is a generic term of WEB page version instant messaging software, and along with the development of the mobile internet, online communication needs to be performed across various clients such as WEB, APP, weChat public numbers, applets and the like. The traditional online customer service basically can only realize a one-to-one answer mode, can not infer the relation and logic between texts, and can not give the most proper answer according to the repeated asking and starting of the user. The time for communication between customer service and clients is prolonged, the service efficiency and effect are greatly reduced, the user experience of the clients is reduced, and the pressure of manual customer service is increased.

With the high-speed development of the artificial intelligence technology in recent two years, the AI artificial intelligence model is developed successively, the intelligent AI has large-scale knowledge processing technology, natural language understanding technology, knowledge management technology, automatic question-answering system, reasoning technology and the like, and has industry universality, and the intelligent customer service trained by the AI artificial intelligence can provide intelligent reception and auxiliary reception services by combining with the natural language processing (NLP, natural Language Processing) technology in the aspect of artificial intelligence and the like, so that the intelligent customer service system is widely applied to the industries such as finance, medical treatment and the like at present.

Therefore, a fast and effective technical means based on natural language is established for communication between enterprises and massive users, and statistical analysis information required by fine management can be provided for the enterprises, so that the technical problem to be solved is solved.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provide a micro-grid intelligent customer service system based on ChatGLM and a construction method thereof, which are used for generating the doubt solution and proposal scheme wanted by a customer more accurately for the information mentioned by the customer, saving the time for the customer to communicate with the customer service, greatly improving the working efficiency of the artificial customer service, relieving the pressure of the artificial customer service, accurately serving each customer and providing better customer service experience for the customer.

The aim of the invention can be achieved by the following technical scheme:

according to one aspect of the invention, a micro-grid intelligent customer service system is provided, the system is realized based on ChatGLM, the customer service system comprises a bottom layer operation environment, a data storage layer, a service capability layer, a business capability layer, an interface access layer, a security protection layer, a display layer and an access terminal layer, the service capability layer comprises a ChatGLM center module, when a user consults, the customer service system calls a model open interface API of the ChatGLM center module, and the ChatGLM center module identifies the user intention and carries out multi-round conversations.

As an optimal technical scheme, if the ChatGLM center module can not solve the problem of clients, the ChatGLM center module is accessed into manual customer service.

As an optimal technical scheme, the ChatGLM center module enters quality inspection after finishing client consultation every time and is used for optimizing and updating the system.

According to another aspect of the present invention, there is provided a construction method for the ChatGLM-based micro-grid intelligent customer service system, the method specifically including the steps of:

step S1, collecting data text and constructing a data set;

step S2, training and fine tuning the data set constructed in the step S1 through a chatGLM artificial intelligent language model to generate a training chatGLM model;

step S3, adding a ChatGLM center module in a service capability layer of the intelligent customer service system, integrating a trained ChatGLM model in the module, and creating an interactive robot dialogue interface and constructing task multi-round dialogue;

and S4, counting and analyzing the interaction process with the client, upgrading the data set, and optimizing and updating the intelligent customer service system.

As a preferable technical solution, the step S1 specifically includes the following steps:

step S1.1, collecting and sorting data from different sources into a data set;

s1.2, cleaning and preprocessing data;

s1.3, performing feature conversion and dimension reduction on the data;

and S1.4, merging the data sets into a training set, a verification set and a test set, wherein the training set is used for training a model, the verification set is used for adjusting model parameters and preventing overfitting, and the test set is used for evaluating the performance of the model.

As a preferable technical solution, the step S2 specifically includes the following steps:

s2.1, downloading and constructing a ChatGLM model frame and configuring an environment;

step S2.2, loading a data set;

s2.3, setting model parameters;

s2.4, training and fine tuning are carried out;

and S2.5, reasoning and evaluating the model, and adding an open interface API to generate a final training model.

As a preferred technical solution, the interface API in step S2.5 is used to retrieve information in real time.

As a preferable technical solution, the ChatGLM center module in step S3 needs to configure parameters and preset parameters.

As an preferable technical scheme, the intelligent customer service system in step S3 uses the problem of the customer as a parameter, calls the model open interface API, and renders the return value of the API interface to the chat page, thereby realizing multiple rounds of conversations.

As a preferable technical solution, the step S4 specifically includes the following steps:

step S4.1, adding a user interaction record collection function in the customer service system;

s4.2, counting and analyzing a customer interaction process, collecting new data, adjusting a data set and a model, and retraining;

and S4.3, re-executing the steps S1-S3 to upgrade the intelligent customer service system.

Compared with the prior art, the invention has the following advantages:

1) The invention establishes an enterprise complete knowledge base, and provides fine-grained knowledge management technology and statistical analysis information required by fine management for enterprises in the electric related industry;

2) The assistance micro-grid customer service is more intelligent, and a quick and effective technical means based on natural language is established for communication between enterprises and mass users;

3) According to the invention, for the information mentioned by the clients, the doubt solution and proposal scheme wanted by the clients is generated more accurately, so that the time for the clients to communicate with the customer service is saved, the working efficiency of the manual customer service is greatly improved, the pressure of the manual customer service is reduced, each client is accurately served, and better customer service experience is provided for the clients;

4) The ChatGLM adopted by the invention unifies the pre-training framework of the current main stream, is a universal language model, and enables a system applying the ChatGLM to have a wide future development space;

5) The ChatGLM adopted by the invention has only 62 hundred million parameters, the deployment requirement is low, and a user can locally deploy on a consumer-level display card, so that the reasoning cost is greatly reduced, and the efficiency is improved.

Drawings

Fig. 1 is a flowchart of a method for constructing a smart customer service system of a micro grid based on ChatGLM according to the present invention;

FIG. 2 is a flow chart of the present invention for constructing a dataset;

FIG. 3 is a schematic diagram of the invention through a ChatGLM artificial intelligence language training model;

fig. 4 is a flowchart of a chartglm model call by the smart customer service system for a micro-grid according to the present invention;

fig. 5 is a schematic diagram of a smart customer service system of a micro grid according to ChatGLM of the present invention;

fig. 6 is a flowchart of a micro grid smart customer service system of ChatGLM according to the present invention;

the reference numerals in fig. 5 indicate:

1. the cloud system comprises a bottom layer operation environment 10, a cloud host, an independent server 11, a third-party virtual host 12, a dock container 13;

2. a data storage layer 20, mysql,21, redis caches, 22, oss object storage, 23 and ftp storage;

3. a service capability layer, 30, a dubbo service framework, 300, an IM message module, 301, a session module, 302, a customer service center, 303, a platform management module, 304, a user center, 305, a chatGLM center module, 306, a data center, 307, a data synchronization module, 31, a basic service middleware, 310, a zk registry, 311, a rock mq,312, a search engine, 313, a data synchronization Maxwell,314 and a log component;

4. the system comprises a business capability layer 40, a department personnel management module 41, a channel management module 42, a strategy management module 43, an agent workbench 44, a common term management module 45, a session management module 46, a client management module 47, a data statistics module 48, an agent configuration management module 49 and a platform configuration management module;

5. interface access layer, 50, unifying api gateway;

6. security protection layer 60, firewall 61, DDOS high IP,62, intrusion detection;

7. display layers, 70, vue/act, 71, android/ios,72, html;

8. access terminal layer, 80, pc seat side/administrator portal, 81, chatui mobile side: H5/APP,82, chatuiPC terminal.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.

The invention provides a method for constructing a micro-grid intelligent customer service system based on ChatGLM, which is characterized in that according to data corpus of related electric industry and micro-grid question-answering network, a data set is constructed, training and fine tuning are carried out through a ChatGLM artificial intelligent language model, an interactive robot dialogue interface is created, a neural network algorithm is adopted to construct task type multi-round dialogue, and customer interaction process is counted and analyzed to optimize and update intelligent customer service.

The specific process implemented by the method refers to fig. 2, and comprises the following steps:

and S1, collecting data texts related to the electric industry and the micro-grid question-answering network, and constructing a data set.

The method specifically comprises the following steps:

step S1.1, combining the electric industry characteristics and the related data of the micro-grid, collecting and summarizing in several modes: searching or purchasing a data set shared by other people through the Internet; collecting data from a database manually or automatically; automatically collecting data from the internet by a crawler program; data is collected by manually entering summaries. Finally, data from different sources are collected and arranged into a dialogue data set.

It is necessary to supplement that, in order to better achieve the alignment with the intention of the customer, the communication scene between the customer service and the customer is reproduced, the language style of ChatGLM is changed to the mouth kiss of the enterprise and the unit, and the effective multi-round dialogue communication is completed. The method includes the steps that historical chat records of clients, such as WeChat chat records and the like, are acquired as far as possible, files are exported and made into dialogue formats and stored in json, and chat data sets are made;

step S1.2, performing data cleaning and preprocessing through Pandas, removing or rewriting blank spaces, line-feed symbols, punctuation marks and the like in the text, deleting or modifying stop words, misspelling, repeated words and the like in the text, filling in missing values, and finally finishing into unified typesetting and format.

And S1.3, performing content adjustment, such as splitting a plurality of long sentences into short sentences of one line. And selecting and using a Pandas library to extract the characteristics of the data according to experience and domain knowledge, extracting required characteristics, and carrying out some characteristic conversion and dimension reduction through NumPy and other tools.

Step S1.4, dividing the data set into a training set, a verification set and a test set. The training set is used to train the model, the validation set is used to adjust model parameters and prevent overfitting, and the test set is used to evaluate the performance of the model.

Step S2, training and fine tuning are carried out through a ChatGLM artificial intelligence language model to generate a training model, and a specific process realized by the method is shown in FIG. 3 and comprises the following steps:

and S2.1, downloading and constructing a ChatGLM model framework, configuring Transformers, pyTorch-Nightly and other related dependencies and environments, importing required libraries, and defining a transducer model.

And S2.2, replacing and loading the data set, configuring a training set address train_file by using the P-Tuning v2, verifying a set address valid_file, and storing the relevant Key and the model.

And S2.3, setting model parameters according to the configuration conditions of a CPU, a memory and a video memory of a computer, selecting a model quantification mode, changing model loading in codes into local loading, and adjusting the soft sample length and the training learning rate by using the mps rear end.

And S2.4, writing a training code, and running the bash train.

And S2.5, running a flash evaluation.sh to infer and evaluate the model, wherein evaluation indexes are Chinese Rouge score and BLEU-4, and using a model architecture to instantiate check points, the feasibility of the model can be checked and tested by pipeline additionally, and an open interface API (application program interface) is increased to complete the deployment of the model.

Step S3, integrating the trained ChatGLM model, creating an interactive robot dialogue interface, and constructing a task type multi-round dialogue by adopting a neural network algorithm, wherein the specific process realized by the method is shown in fig. 4, 5 and 6, and comprises the following steps:

step S3.1, according to the intelligent customer service system architecture of the figure 5, on the basis of the architecture required by the complete customer service system, a ChatGLM center module is added on a service capability module, and relevant parameters and presets are configured;

s3.2, writing an intelligent customer service system with a chat interaction page of the ChatGLM intelligent customer service robot, calling an API with an open model by taking a problem of a customer as a parameter, and rendering a return value of an API interface to the chat page so as to realize multi-round dialogue, and referring to FIG. 4;

step S3.3, after the project test is completed, running and deploying;

and S4, counting and analyzing a customer interaction process, and optimizing and updating the intelligent customer service by combining the related new functions of the micro-grid or the new product upgrading data set. The method specifically comprises the following steps:

step S4.1, adding a user interaction record collection function, and storing the data in the Pinecone, vespa or other vector databases.

And S4.2, on one hand, counting and analyzing the customer interaction process, and on the other hand, collecting new functions or new product upgrading data related to the micro-grid, further evaluating and adjusting the data set and the model, and retraining.

And S4.3, re-executing the steps S1-S3 to upgrade the intelligent customer service system.

The system is constructed by the construction method and based on the ChatGLM intelligent customer service system, and the system is composed of a bottom layer operation environment 1, a data storage layer 2, a service capability layer 3, a service capability layer 4, an interface access layer 5, a security protection layer 6, a display layer 7 and an access terminal layer 8, and has the functions of providing knowledge management technology, statistically analyzing information and generating a solution.

The underlying operating environment 1 includes a cloud host 10, a stand-alone server 11, a third party virtual host 12, and a docker container 13.

The data store layer 2 includes mysql20 database redis cache 21, oss object store 22, and ftp store 23.

The service capability layer 3 comprises a dubbo service framework 30 and a basic service middleware 31, wherein the basic service middleware 31 comprises a zk registry 310, a RocketMQ311, a search engine 312ES, a data synchronization Maxwell313 and a log component 314; the IM message module 300 in the dubbo service framework 30 is used for handling chat messages, the session module 301 is used for managing sessions, the customer service center 302 is used for handling customer service assignments, the platform management module 303 is used for querying with a configuration, the user center 304 is used for providing user management, and the ChatGLM center 305 is used for interfacing with ChatGLM and related configurations.

The business capability layer 4 includes a department personnel management module 40, a channel management module 41, a policy management module 42, an agent workbench 43 module, a commonly used term management module 44, a session management module 45, a client management module 46, a data statistics module 47, an agent configuration management module 48, and a platform configuration management module 49.

The interface access layer 5 unifies the api gateway 50, including black and white name, access control and authentication.

The security protection layer 6 is considered in public cloud deployment and comprises a firewall 60, DDOS high security IP61 and intrusion detection 62.

The presentation layer 7 includes vue/act 70, android/ios71, html725.

The access terminal layer 8 includes a pc seat side/administrator portal 80, a chatui mobile side: H5/APP81 and chatuiPC end 82.

The application process of the intelligent customer service system of the micro-grid is shown in fig. 6, a customer enters into a system consultation, the system loads customer information and invokes a model open interface API of the ChatGLM center module 305, the ChatGLM center module 305 carries out user intention recognition and carries out multiple rounds of dialogue, and if the customer problem is solved after the dialogue is finished, the customer selects to evaluate; otherwise, the ChatGLM center module 305 assists in answering the questions during the access of the artificial customer service, the customer selects to evaluate after the artificial customer service processes the questions, and the system enters the session to enter the quality inspection system after the customer consultation is completed, so as to perform optimization updating on the system.

The invention can be attached to the business requirement of an enterprise micro-grid, provides a fine-grained knowledge management technology for enterprises in related electric industries, establishes a quick and effective technical means based on natural language for communication between the enterprises and mass users, and simultaneously can provide statistical analysis information required by fine management for the enterprises. For the information mentioned by the clients, the doubt solution and proposal scheme wanted by the clients is generated more accurately, meanwhile, the time for the clients to communicate with the customer service is saved, the working efficiency of the manual customer service is greatly improved, the pressure of the manual customer service is reduced, each client is served accurately, and better customer service experience is provided for the clients.

While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made and equivalents will be apparent to those skilled in the art without departing from the scope of the invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims (10)

1. The utility model provides a little electric wire netting intelligence customer service system, this system is realized based on ChatGLM, customer service system includes bottom operational environment (1), data storage layer (2), service capability layer (3), business capability layer (4), interface access layer (5), security protection layer (6), show layer (7) and access terminal layer (8), its characterized in that, service capability layer (3) in including ChatGLM center module (305), when the user consults, customer service system call the model open interface API of ChatGLM center module (305), chatGLM center module (305) carries out user intention discernment and carries out many rounds of conversations.

2. The smart customer service system of claim 1, wherein the ChatGLM center module (305) accesses a manual customer service if the customer problem cannot be resolved.

3. A micro grid intelligent customer service system according to claim 2, wherein the ChatGLM center module (305) enters quality inspection after each customer consultation is completed, and is used for optimizing and updating the system.

4. A method for constructing the intelligent customer service system of the micro-grid, which is characterized by comprising the following steps:

step S1, collecting data text and constructing a data set;

step S2, training and fine tuning the data set constructed in the step S1 through a chatGLM artificial intelligent language model to generate a training chatGLM model;

step S3, adding a ChatGLM center module in a service capability layer of the intelligent customer service system, integrating a trained ChatGLM model in the module, and creating an interactive robot dialogue interface and constructing task multi-round dialogue;

and S4, counting and analyzing the interaction process with the client, upgrading the data set, and optimizing and updating the intelligent customer service system.

5. The construction method according to claim 4, wherein the step S1 specifically comprises the steps of:

step S1.1, collecting and sorting data from different sources into a data set;

s1.2, cleaning and preprocessing data;

s1.3, performing feature conversion and dimension reduction on the data;

and S1.4, merging the data sets into a training set, a verification set and a test set, wherein the training set is used for training a model, the verification set is used for adjusting model parameters and preventing overfitting, and the test set is used for evaluating the performance of the model.

6. The construction method according to claim 4, wherein the step S2 specifically comprises the steps of:

s2.1, downloading and constructing a ChatGLM model frame and configuring an environment;

step S2.2, loading a data set;

s2.3, setting model parameters;

s2.4, training and fine tuning are carried out;

and S2.5, reasoning and evaluating the model, and adding an open interface API to generate a final training model.

7. The method according to claim 6, wherein the interface API in step S2.5 is used for retrieving information in real time.

8. The method according to claim 4, wherein the ChatGLM center module in step S3 requires configuration parameters and presets.

9. The method according to claim 4, wherein the intelligent customer service system in step S3 uses the problem of the customer as a parameter, calls a model open interface API, and renders the return value of the API interface to the chat page, thereby implementing multiple rounds of conversations.

10. The construction method according to claim 4, wherein the step S4 specifically comprises the steps of:

step S4.1, adding a user interaction record collection function in the customer service system;

s4.2, counting and analyzing a customer interaction process, collecting new data, adjusting a data set and a model, and retraining;

and S4.3, re-executing the steps S1-S3 to upgrade the intelligent customer service system.