WO2021114834A1 - Customer service question update method and system, terminal device, and computer storage medium - Google Patents

Customer service question update method and system, terminal device, and computer storage medium Download PDF

Info

Publication number
WO2021114834A1
WO2021114834A1 PCT/CN2020/118487 CN2020118487W WO2021114834A1 WO 2021114834 A1 WO2021114834 A1 WO 2021114834A1 CN 2020118487 W CN2020118487 W CN 2020118487W WO 2021114834 A1 WO2021114834 A1 WO 2021114834A1
Authority
WO
WIPO (PCT)
Prior art keywords
customer service
question
questions
clustering
sentence
Prior art date
Application number
PCT/CN2020/118487
Other languages
French (fr)
Chinese (zh)
Inventor
侯翠琴
文彬
李剑锋
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021114834A1 publication Critical patent/WO2021114834A1/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/332—Query formulation
    • G06F16/3329—Natural language query formulation or dialogue systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/23—Updating
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0281—Customer communication at a business location, e.g. providing product or service information, consulting

Definitions

  • This application relates to the technical field of customer service robots, and in particular to a method, system, terminal device, and computer storage medium for updating customer service questions.
  • Intelligent customer service robots are usually composed of modules such as automatic speech recognition (ASR), intention recognition, question and answer modules, knowledge base management, knowledge graphs, dialogue management, text generation, and text to speech (TTS).
  • ASR automatic speech recognition
  • question and answer modules knowledge base management
  • knowledge graphs knowledge graphs
  • dialogue management text generation
  • text to speech TTS
  • the knowledge base-based FAQ frequently Asked Questions, common question items and answers to corresponding questions
  • the Q&A module provides users with satisfactory answers by querying standard questions matching user questions in the knowledge base, which is the most important module in intelligent customer service robots.
  • the knowledge base is an important part of the FAQ module. It is composed of frequently asked questions by users and corresponding answer pairs. In order to adapt to different users’ different ways of asking the same question, each standard question will be generalized to several similarities. problem. With the development of the business, the question set of the knowledge base needs to be continuously updated to improve the answer rate of the customer service robot.
  • clustering technology is generally used to cluster the problems that the nearest machine customer service robot has not responded to, and then the operation personnel rely on the clustering results and the knowledge base to sort out new standard problems.
  • the corresponding similar questions are added to the knowledge base to update the question set.
  • the inventor realized that with the development of the business, a large number of new business problems will continue to emerge. It is no longer possible to cluster high-quality problems based on the clustering algorithm for clustering unresponsive problems. Update the problem set.
  • the main purpose of this application is to provide an update method, system, terminal device and computer storage medium for customer service problems, which aims to solve the existing method of clustering based on clustering algorithm for unresponsive problems, which cannot be clustered to obtain high
  • the quality issue is a technical issue that updates the problem set of the intelligent customer service robot.
  • an embodiment of the present application provides a method for updating customer service questions, and the method for updating customer service questions includes:
  • a new customer service question is generated according to the clustering result to update the customer service question set.
  • this application also provides a system for updating customer service questions, and the system for updating customer service questions includes:
  • the construction module is used to construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
  • the learning module is configured to train a similarity model based on the training data, and use the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
  • a clustering module configured to perform a clustering operation based on the clustering parameters to obtain a clustering result of the unresponsive customer service question
  • the update module is used to generate new customer service questions according to the clustering result to update the preset customer service question set
  • this application also provides a terminal device, the terminal device including: a memory, a processor, a communication bus, and an update program for customer service questions stored on the memory, and the communication bus is used to implement The communication connection between the processor and the memory; the processor is used to execute the update program of the customer service question, so as to implement the following steps:
  • a new customer service question is generated according to the clustering result to update the preset customer service question set.
  • the present application also provides a computer storage medium that stores one or more programs, and the one or more programs can be executed by one or more processors for :
  • a new customer service question is generated according to the clustering result to update the customer service question set.
  • the method, system, terminal device, and computer-readable storage medium for updating customer service questions construct question pairs based on customer service questions in a preset customer service question set, and combine the question pairs into training data; based on the training Data training similarity model, and use the training convergence similarity model to calculate the clustering parameters of the unresponsive customer service question; perform clustering operations based on the clustering parameters to obtain the clustering result of the unresponsive customer service question; according to the aggregation
  • the class result generates a new customer service question to update the preset customer service question set.
  • This application is based on the intelligent customer service robot extracting customer service questions from the preset customer service question set to construct question pairs, and then correspondingly combining the question pairs to obtain training data, using the training data to train the similarity model, and then training the convergent similarity model Calculate the clustering parameters of the customer service robot not responding to the customer service question, and then perform the subsequent update process of the customer service question set after clustering the non-responsive customer service question based on the clustering parameter, thereby realizing that the new question is obtained in the simple clustering problem
  • the clustering parameters of the unresponsive customer service questions are adaptively learned, so that more valuable high-quality questions are obtained during clustering to update the customer service question set, and the intelligent customer service robot is improved. The question is answered efficiently.
  • Figure 1 is a schematic structural diagram of a terminal device hardware operating environment involved in a method according to an embodiment of the application;
  • FIG. 2 is a schematic flowchart of an embodiment of a method for updating a customer service question application
  • Figure 3 is a schematic diagram of the functional modules of the update system for applying for customer service issues.
  • FIG. 1 is a schematic diagram of the device structure of the terminal device hardware operating environment involved in the solution of the embodiment of the present application.
  • the terminal device in the embodiment of the present application may be a smart customer service robot, or a terminal device such as a PC, a smart phone, a tablet computer, and a portable computer.
  • the terminal device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005.
  • the communication bus 1002 is used to implement connection and communication between these components.
  • the user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • the memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a magnetic disk memory.
  • the memory 1005 may also be a storage device independent of the aforementioned processor 1001.
  • the terminal equipment may also include a camera, RF (Radio Frequency (radio frequency) circuits, sensors, audio circuits, WiFi modules, etc.
  • sensors such as light sensors, motion sensors and other sensors.
  • the light sensor may include an ambient light sensor and a proximity sensor, where the ambient light sensor can adjust the brightness of the display screen according to the brightness of the ambient light, and the proximity sensor can turn off the display screen and/or backlight when the device is moved to the ear .
  • the gravity acceleration sensor can detect the magnitude of acceleration in various directions (usually three-axis), and can detect the magnitude and direction of gravity when it is stationary.
  • mobile terminals can also be equipped with other sensors such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc. Go into details again.
  • the device structure shown in FIG. 1 does not constitute a limitation on the terminal device.
  • the terminal device may also include more or less components than those shown in the figure, or a combination of certain components. Components, or different component arrangements.
  • the memory 1005 which is a computer storage medium, may include an operating system, a network communication module, a user interface module, and an update program for customer service issues.
  • the network interface 1004 is mainly used to connect to the back-end server and communicate with the back-end server;
  • the user interface 1003 is mainly used to connect to the client (user side) and communicate with the client; and the processing
  • the device 1001 can be used to call the update program for customer service questions stored in the memory 1005 and execute the following steps:
  • a new customer service question is generated according to the clustering result to update the preset customer service question set.
  • customer service questions include standard questions and similar questions
  • the processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
  • each similar problem corresponding to the first standard problem is constructed to obtain each similar problem pair;
  • each similar problem corresponding to the second standard problem in each standard problem is constructed to obtain each non-similar problem pair, wherein the first standard problem is different from the second standard problem ;
  • Each pair of similar questions is correspondingly combined with each pair of non-similar questions to form each training data.
  • processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
  • the clustering parameters include: the vector representation and similarity value of the question sentence.
  • the processor 1001 may be used to call the update program of the customer service question stored in the memory 1005, and further execute the following steps:
  • processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
  • a preset clustering algorithm based on sentence representation is invoked, and the vector representation is used to cluster the unresponsive customer service questions to obtain a clustering result of the unresponsive customer service questions.
  • processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
  • the new standard question, similar questions corresponding to the new standard question, and the service answer are added to the preset customer service question set to update the preset customer service question set.
  • processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
  • This application provides a method for updating customer service questions.
  • Figure 2 is a schematic flow chart of the first embodiment of the method for updating customer service questions of the application.
  • the first embodiment of the method for updating customer service questions of this application is applied to a smart customer service robot. Update methods include:
  • step S100 a question pair is constructed according to the customer service questions in the preset customer service question set, and the question pair is combined into training data;
  • the intelligent customer service robot extracts customer service questions from a set of preset customer service questions, then constructs each question pair based on the customer service question, and finally performs a one-to-one corresponding combination of each question pair to form training data.
  • the FAQ question and answer module in the intelligent customer service robot has a knowledge base, which is an important part of the FAQ question and answer module.
  • the preset customer service question set is the customer service question set saved in the knowledge base.
  • the customer service question set consists of many standard questions and one or more similar questions corresponding to each standard question. Therefore, the intelligent customer service robot asks questions through FAQs.
  • the module searches the knowledge base for standard questions that match the customer's customer service questions, so as to provide users with satisfactory answers.
  • the training data formed by a one-to-one corresponding combination of each question pair by the intelligent customer service robot may be a data set, which is composed of multiple pieces of training data.
  • the intelligent customer service robot periodically or randomly initiates an update operation for the customer service question set.
  • the intelligent customer service robot accesses the knowledge base of the FAQ module, and extracts standard questions and similarities from the customer service questions stored in the knowledge base. Questions, and then construct a question pair based on the standard question and similar questions, and finally form multiple pieces of training data by correspondingly combining the question pair according to the one-to-one rule to form a data set.
  • step S100 may include:
  • Step S101 extract each standard question and each similar question corresponding to each standard question from the preset customer service question set;
  • Step S102 constructing each similar question pair corresponding to the first standard question according to the first standard question among the standard questions;
  • the intelligent customer service robot accesses the knowledge base of the FAQ module to extract multiple standard questions and the multiple questions from the customer service question set stored in the knowledge base.
  • Each standard question corresponds to a similar question, and then, the intelligent customer service robot randomly combines any one of the multiple standard questions with multiple similar questions of the standard question, so that the component obtains multiple similar question pairs.
  • Step S103 constructing each similar question corresponding to the second standard question in each standard question according to the first standard question to obtain each non-similar question pair;
  • the first standard problem is different from the second standard problem.
  • the intelligent customer service robot accesses the knowledge base of the FAQ question and answer module, and extracts multiple standard questions and similar questions corresponding to each of the multiple standard questions from the customer service questions stored in the knowledge base,
  • the intelligent customer service robot can simultaneously execute (or asynchronously, of course) the operation of constructing non-similar question pairs while forming similar problem pairs. That is, the intelligent customer service robot can combine any one of the multiple standard questions with the same one.
  • multiple similar questions of any standard question except the standard question are randomly combined to obtain multiple non-similar question pairs.
  • Step S104 correspondingly combining each pair of similar questions with each pair of non-similar questions to form each training data.
  • the intelligent customer service robot combines the multiple similar question pairs and multiple dissimilar question pairs to form a piece of training data by correspondingly combining the rules of a similar question pair and a dissimilar question pair.
  • Multiple pieces of training data constructed according to the above rules constitute a data set.
  • the intelligent customer service robot when the intelligent customer service robot forms training data based on the combination of multiple similar question pairs and multiple dissimilar question pairs, it can also combine a similar question pair and multiple dissimilar question pairs.
  • a piece of training data is formed, and multiple pieces of training data are constructed to form a data set.
  • similar question pairs and non-similar questions that make up the same piece of training data have the same standard questions.
  • Similar question pairs and dissimilar question pairs may also be different for their respective standard questions.
  • the method for updating customer service questions in this application does not specifically limit the combination of similar question pairs and dissimilar question pairs in the training data.
  • Step S200 Train a similarity model based on the training data, and use the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
  • the intelligent customer service robot After the intelligent customer service robot extracts the customer service question set from the preset knowledge base, and then builds the training data based on the customer service question set, the intelligent customer service robot uses the training data to train the similarity model until the similarity model converges, and then the intelligent The customer service robot uses the similarity model trained to convergence based on the training data, and calculates the unanswered customer service question that has not been answered to obtain the clustering parameters for subsequent clustering of the non-responsive customer service question.
  • the similarity model may specifically be a bert model (Bidirectional Encoder Representations from Transformers is a very effective general pre-training language representation model proposed by Google recently), xlnet (a general autoregressive pre-training method) model, siamesecnn (twin convolution) network, siameselstm (twin) network, etc. Any of the models.
  • non-response customer service question refers to the customer service question recorded by the intelligent customer service robot when the customer service question raised by the user cannot be successfully matched to the standard question during the entire operation process of the intelligent customer service robot, so that the answer is not obtained to output to the user ( That is, unresponsive customer service questions are customer service questions that the intelligent customer service robot has not answered).
  • the intelligent customer service robot uses the bert model as the similarity model. After the intelligent customer service robot obtains multiple pieces of training data based on multiple similar question pairs and multiple non-similar question pairs, part of the multiple pieces of training data The training data is input to the bert model as training samples to train the bert model. After the intelligent customer service robot verifies that the bert model has converged, the bert model that has been trained and converged is used to calculate one or more pre-recorded unresponsive customer services The clustering parameter of the question, so as to perform a clustering operation on the unresponsive customer service question based on the clustering parameter.
  • Step S300 performing a clustering operation based on the clustering parameters to obtain a clustering result of the unresponsive customer service question
  • the intelligent customer service robot uses the existing mature clustering algorithm to calculate and output the clustering parameters of the unresponsive customer service question based on the training convergence similarity model, and then uses the distance parameter to perform the clustering operation on the unresponsive customer service question, thereby obtaining The clustering result that did not respond to customer service questions.
  • step S400 a new customer service question is generated according to the clustering result to update the customer service question set.
  • the intelligent customer service robot After the intelligent customer service robot obtains the clustering result of the unresponsive customer service question through the clustering operation, it automatically sorts out the new customer service question according to the clustering result and the customer service question set stored in the knowledge base of the FAQ module.
  • the customer service question is added to the customer service question set in the knowledge base, thereby completing the update of the customer service question set.
  • the intelligent customer service robot may also output the clustering result to the operator of the intelligent customer service robot after obtaining a clustering result that does not respond to customer service questions through a clustering operation, thereby assisting the According to the clustering result and the customer service question set stored in the knowledge base of the FAQ module, the operator sorts out new customer service questions, and then the intelligent customer service robot receives the new customer service question input by the operator, and adds the new customer service question To the customer service question set in the knowledge base to complete the update of the customer service question set.
  • an intelligent customer service robot extracts a customer service question set from a preset knowledge base, and then constructs training data based on the customer service question set; extracts a customer service question set from the preset knowledge base, and then constructs a customer service question set based on the customer service question set
  • the intelligent customer service robot uses the training data to train the similarity model until the similarity model converges. Then, the intelligent customer service robot uses the training data to train the similarity model to converge, and responds to unanswered unanswered questions.
  • the customer service question is calculated to obtain the subsequent clustering parameters for the unresponsive customer service question; after the clustering parameter of the unresponsive customer service question is calculated and output based on the similarity model of the training convergence, the intelligent customer service robot uses the distance parameter to obtain the clustering parameter for the unresponsive customer service question. Perform clustering operations in response to customer service questions to obtain the clustering results of the unresponsive customer service questions; finally, the intelligent customer service robot automatically sorts and obtains new customer service based on the clustering results and the customer service question set stored in the knowledge base of the FAQ module Questions, and then add the new customer service question to the customer service question set in the knowledge base, thereby completing the update of the customer service question set.
  • step S200 "training similarity based on the training data
  • the steps of "model” can include:
  • Step S201 dividing the training data into a first data set and a second data set
  • the intelligent customer service robot Based on a preset knowledge base, the intelligent customer service robot extracts standard questions and similar questions corresponding to the standard questions from the customer service questions stored in the preset knowledge base, and then combines pairs of similar questions and pairs of dissimilar questions. After constructing a data set of training data for the similar question pair and the dissimilar question pair, the intelligent customer service robot divides the data set according to a preset ratio or randomly to obtain the first data set and the second data set.
  • the preset ratio can be a ratio set by the operator based on design needs. It should be understood that the preset ratio can be set to any numerical ratio based on different design needs of actual applications. The method for updating customer service questions in this application does not specifically limit the size of the preset ratio.
  • Step S202 using the first data set to train the similarity model, and using the second data set to verify the trained similarity model;
  • the intelligent customer service robot After the intelligent customer service robot divides the training data data set into the first data set and the second data set according to a preset ratio or randomly, the intelligent customer service robot inputs the training data in the first data set into the similarity model to compare the training data.
  • the similarity model is trained, and after each round of training of the similarity model using the training data is completed, the training data in the second data set is then input to the similarity training model to according to the output result of the similarity model Verify whether the similarity model has converged after the end of the current round of training.
  • the intelligent customer service robot can also input the training data in the second data set to the similarity model to train the similarity model, and use the training data to target each similarity model. After one round of training is over, the training data in the first data set is then input to the similarity training model to verify whether the similarity model has converged after the end of the current round of training according to the output result of the similarity model.
  • Step S203 Mark the verified similarity model as a training converged similarity model, and store the trained convergent similarity model in the blockchain.
  • the intelligent customer service robot inputs the training data in the second data set to the similarity training model after training, and verifies that the similarity model has been trained and converged according to the output result of the similarity model, confirm to use the second data set
  • the verification of the similarity model has passed, and the similarity model is immediately marked as a training convergent similarity model, and then the training converged similarity model is stored in a node of the blockchain.
  • the trained convergent similarity model in order to ensure that the training converged similarity model will not be incorrectly modified or removed, can be stored in a node of a blockchain to ensure the index
  • the security of the label further ensures the accuracy of the subsequent clustering parameter calculation of the intelligent customer service robot's unresponsive customer service problems in the similarity model based on training convergence, and further improves the efficiency of the intelligent customer service robot's adaptive learning of new business problems.
  • the intelligent customer service robot divides the constructed training data data set into a first data set and a second data set; then, the training data in the first data set is used to train the similarity model, and After each round of training is over, the training data in the second data set is used to verify the trained similarity model; finally, the verified similarity model is marked as the training convergence similarity model, and the training The converged similarity model is stored in the blockchain for subsequent recall.
  • the method for updating customer service questions of this application may also include :
  • Step A Collect the unresponsive customer service questions at intervals of a preset length of time, and store the collected unresponsive customer service questions in the blockchain.
  • the intelligent customer service robot will encounter unresponsive customer service questions that cannot output answers to the user during the continuous operation, so the intelligent customer service robot can establish a temporary storage space in advance.
  • the unresponsive customer service question cannot output an answer to the user, the unresponsive customer service question is cached in the storage space.
  • the preset duration can be specifically 24 hours. It should be understood that, based on the different design needs of actual applications, the preset duration can of course also be any other duration. The method for updating customer service questions in this application does not apply to the preset duration. Specifically defined.
  • the intelligent customer service robot obtains from the pre-established temporary storage space in the 24 hours before the 0:00, what the intelligent customer service robot records during the operation process is the user
  • the non-response customer service question that outputs the answer to the raised customer service question is then stored in a node of the blockchain for subsequent calls.
  • the intelligent customer service robot can also directly store in the pre-established temporary storage space.
  • the information recorded by the intelligent customer service robot during the operation process is the user's information.
  • the non-response customer service question with the output answer of the proposed customer service question is persistently cached for subsequent calculation of the clustering parameter of the non-response customer service question and direct call when performing clustering operations.
  • the intelligent customer service robot uses the trained and converged similarity model to calculate the clustering parameters of the unresponsive customer service question, including but not limited to: the vector representation of the question sentence that does not respond to the customer service question, and the For the similarity value of the question sentence, in the above step S200, the step of "using the trained and converged similarity model to calculate the clustering parameters that did not respond to the customer service question" may include:
  • Step S204 Obtain the question sentence that does not respond to the customer service question and the duplicate sentence of the question sentence, and input the question sentence and the duplicate sentence into a training convergent similarity model;
  • the intelligent customer service robot After the intelligent customer service robot obtains the unresponsive customer service question, it uses the existing mature language processing technology to read the question sentence that does not respond to the customer service question, and copies the question sentence to obtain the duplicate sentence of the question sentence, and then the intelligent customer service The robot inputs the question sentence and the duplicate sentence of the question sentence into the similarity model that is trained to converge.
  • the intelligent customer service robot After the intelligent customer service robot obtains one or more unresponsive customer service questions from the nodes of the blockchain, it calls any existing mature natural language processing technology to read the text of the unresponsive customer service questions one by one As a question sentence, and continue to copy the question sentence to obtain the copy sentence corresponding to each question sentence, and then the intelligent customer service robot inputs the question sentence and the copy sentence corresponding to the question sentence one by one to the same node from the blockchain Among the similarity models that have been trained and converged obtained in.
  • Step S205 Obtain a training convergence similarity model and calculate a vector average of similarity values between the question sentence and the copied sentence;
  • Step S206 Use the vector average value as a vector representation of the question sentence
  • the similarity model calculates the similarity value between the question sentence and the copied sentence, Obtain the vector of the question sentence and the vector of the copied sentence, and then calculate the vector average of the vector of the question sentence and the vector of the copied sentence, and use the vector average as the vector of the question sentence that did not respond to the customer service question.
  • the step of "using the trained and converged similarity model to calculate the clustering parameters that did not respond to the customer service question" may further include:
  • Step S207 traverse the question sentences to construct sentence pairs, and input the sentence pairs to train a convergent similarity model
  • Step S208 Obtain the similarity value output by the trained and converged similarity model after calculating the similarity value of the sentence pair, and use the similarity value as the similarity value of the question sentence.
  • the intelligent customer service robot obtains the unresponsive customer service question, reads the question sentence that does not respond to the customer service question through the existing mature language processing technology, and traverses the question sentence to follow the current question sentence and other question sentences similar to the current question sentence To construct a sentence pair, and then input the sentence pair into the similarity model that has been trained and converged, so that the similarity model calculates the similarity value between the question sentence and other question sentences that are similar to the question sentence for the sentence pair. Finally, the intelligent customer service robot outputs the similarity value after calculating the similarity value based on the similarity model based on the input sentence pair, and uses the similarity value as the similarity value of the question sentence that does not respond to the customer service question.
  • performing a clustering operation based on the clustering parameters to obtain the clustering result of the unresponsive customer service question may include:
  • Step S301 calling a preset distance-based clustering algorithm, clustering the unresponsive customer service questions by using the similarity value to obtain a clustering result of the unresponsive customer service questions;
  • the intelligent customer service robot calls the existing mature preset distance-based clustering algorithm based on the similarity value of the question sentence output by the trained and converged similarity model that does not respond to the customer service question, and uses the similarity value to the unresponsive customer service question Model clustering operation to obtain the clustering result of the non-response customer service question.
  • the preset distance-based clustering algorithm includes, but is not limited to: affinity propagation (an AP clustering algorithm, in a broad sense, affinity Propagation is a type of message-passing algorithms), spectral clustering (spectral clustering algorithm).
  • step S300 may further include:
  • Step S302 Invoke a preset clustering algorithm based on sentence representation, and cluster the unresponsive customer service questions using the vector representation to obtain a clustering result of the unresponsive customer service questions.
  • the intelligent customer service robot is aimed at using a similarity model that has been trained and converged.
  • the similarity model calculates the similarity value between the question sentence and the copied sentence, it obtains the vector of the question sentence and the vector of the copied sentence, and calculates the resulting unresponsiveness
  • the vector representation of the question sentence of the customer service problem call the existing mature preset clustering algorithm based on sentence representation and use the vector to represent the clustering operation of the model that did not respond to the customer service problem, so as to obtain the clustering result of the non-responsive customer service problem .
  • the preset clustering algorithm based on sentence representation includes but is not limited to: k-means (k-means clustering algorithm: k-means clustering algorithm is an iterative solution clustering analysis algorithm), dbscan (DBSCAN (Density-Based Spatial Clustering of Applications with Noise, is a more representative density-based clustering algorithm) and hierarchical clustering algorithm.
  • the intelligent customer service robot collects unresponsive customer service questions at a preset interval, and stores the collected unresponsive customer service questions in the blockchain or directly caches the unresponsive customer service questions.
  • the intelligent customer service robot calculates the clustering parameters of the unresponsive customer service question (the vector representation and similarity value of the question sentence that does not respond to the customer service question), it obtains the question sentence and the duplicate sentence of the question sentence that did not respond to the customer service question, and then The question sentence and the copied sentence are input to the trained convergent similarity model; then the trained convergent similarity model is obtained to calculate the vector average value of the similarity values between the question sentence and the copied sentence; and the vector average value is used as a vector representation ; Or, construct a sentence pair by traversing the question sentence, and input the sentence pair into a convergent similarity model; then obtain the similarity value output after the similarity value of the sentence pair is calculated by the trained convergent similarity model, and the The similarity value is used as the similarity value of the question sentence.
  • the vector representation and similarity value of the question sentence that does not respond to the customer service question the vector representation and similarity value of the question sentence that did not respond to the customer service question
  • the intelligent customer service robot When the intelligent customer service robot performs a clustering operation for the unresponsive customer service question, it calls the preset distance-based clustering algorithm, and uses the similarity value of the question sentence to cluster the unresponsive customer service question to obtain the cluster of the unresponsive customer service question Result; or, call a preset clustering algorithm based on sentence representation, and use the vector representation of the question sentence to cluster the unresponsive customer service question to obtain the clustering result of the unresponsive customer service question.
  • the clustering parameters of unresponsive customer service questions are further adaptively learned according to the dynamic changes of knowledge base customer service questions, so as to deal with unresponsive customer service questions.
  • customer service questions are clustered, more valuable and high-quality questions can be obtained to update customer service questions, which improves the efficiency of answering questions of intelligent customer service robots.
  • a fourth embodiment of the method for updating customer service questions of this application is proposed.
  • the above step S400 generates a new method according to the clustering result.
  • the customer service questions can include:
  • Step S401 generating a new standard question and similar questions corresponding to the new standard question according to the clustering result and the customer service question set;
  • the intelligent customer service robot clusters the unresponsive customer service questions and obtains the clustering results, it combines the clustering results with the customer service questions stored in the knowledge base of the FAQ module to collect existing standard questions and similar questions, and organize them to generate new ones.
  • the standard problem is similar to the new standard problem.
  • Step S402 configuring respective corresponding service answers for the new standard question and similar questions corresponding to the new standard question;
  • Step S403 Add the new standard question, similar questions corresponding to the new standard question, and the service answer to the customer service question set to update the customer service question set.
  • the intelligent customer service robot After the intelligent customer service robot organizes and generates the new standard question and the similar questions corresponding to the new standard question, the intelligent customer service robot configures the corresponding service answers in different scenarios for the new standard question and the similar question, and then the intelligent customer service The robot establishes the association relationship between the new standard question and its corresponding service answer, the similar question of the new standard question and its corresponding service answer, and then combines the new standard question with its corresponding service answer and the new standard Similar questions and their corresponding service answers are added to the customer service question set stored in the knowledge base of the FAQ module to update the customer service question set.
  • the intelligent customer service robot clusters the unresponsive customer service questions to obtain the clustering result, it can also directly output the clustering result to the operator, so as to assist the operator according to the clustering.
  • Results and customer service questions stored in the knowledge base of the FAQ question and answer module focus on the existing standard questions and similar questions, sort out the new standard questions and similar questions of the new standard questions, and then implement the follow-up to the new standard by the operator The process of configuring the corresponding service answer for the question and the similar question of the new standard question, and then receiving the new standard question and its corresponding service answer input by the operator, and the similar question of the new standard question and its corresponding service answer, And add the new standard question and its corresponding service answer, the similar question of the new standard question and its corresponding service answer to the customer service question set stored in the knowledge base of the FAQ module, so as to realize the analysis of the customer service question set. Update.
  • the clustering operation is performed on the unresponsive customer service question to obtain the clustering result, and then the clustering result is obtained based on the clustering result and the existing
  • the customer service question set organize and generate the new standard question and the similar question corresponding to the new standard question; then configure the corresponding service answer for the new standard question and the similar question corresponding to the new standard question; finally put the new standard question and the similar question corresponding to the new standard question.
  • Standard questions, similar questions corresponding to the new standard questions, and their corresponding service answers are added to the existing customer service question set to update the customer service question set.
  • Figure 3 is a schematic diagram of the functional modules of the application update system for customer service issues.
  • the update system for customer service issues includes:
  • the construction module 101 is configured to construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
  • the learning module 102 is configured to train a similarity model based on the training data, and use the trained convergent similarity model to calculate clustering parameters that have not responded to customer service questions;
  • the clustering module 103 is configured to perform a clustering operation based on the clustering parameters to obtain the clustering result of the unresponsive customer service question;
  • the update module 104 is configured to generate a new customer service question according to the clustering result to update the preset customer service question set.
  • this application also provides a computer storage medium.
  • the computer storage medium may be non-volatile or volatile.
  • the computer storage medium stores one or more programs.
  • the program can also be executed by one or more processors for:
  • a new customer service question is generated according to the clustering result to update the preset customer service question set.
  • the one or more programs may also be executed by one or more processors and used to:
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the method described in each embodiment of the present application.
  • a terminal device which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.

Abstract

A customer service question update method and system, a terminal device, and a computer storage medium, relating to the technical field of customer service robots. The customer service question update method comprises: constructing question pairs on the basis of customer service questions in a preset customer service question set, and combining the question pairs into training data (S100); on the basis of the training data, training a similarity model, and using the trained converging similarity model to calculate clustering parameters of responseless customer service questions (S200); on the basis of the clustering parameters, performing a clustering operation to obtain a clustering result of the responseless customer service questions (S300); and, on the basis of the clustering result, generating a new customer service question and updating the preset customer service question set (S400).

Description

客服问题的更新方法、系统、终端设备及计算机存储介质Update method, system, terminal equipment and computer storage medium for customer service issues
本申请要求于2020年6月24日提交中国专利局、申请号为202010595834.7、名称为“客服问题的更新方法、系统、终端设备及计算机存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。This application claims the priority of a Chinese patent application filed with the Chinese Patent Office, the application number is 202010595834.7, and the title is "Customer Service Problem Update Method, System, Terminal Equipment, and Computer Storage Medium" on June 24, 2020. The entire content of the application is approved. The reference is incorporated in this application.
技术领域Technical field
本申请涉及客服机器人技术领域，尤其涉及一种客服问题的更新方法、系统、终端设备以及计算机存储介质。This application relates to the technical field of customer service robots, and in particular to a method, system, terminal device, and computer storage medium for updating customer service questions.
背景技术Background technique
智能客服机器人通常由语音识别（Automatic Speech Recognition, ASR）、意图识别、问答模块、知识库管理、知识图谱、对话管理、文本生成、语音合成（Text to Speech, TTS）等模块构成。其中，基于知识库的FAQ（Frequently Asked Questions，常见的问题项目与对应问题的解答）问答模块通过在知识库中查询与用户问题相匹配的标准问题，为用户提供满意的答案，是智能客服机器人中最重要的模块。而知识库是FAQ问答模块重要的组成部分，它由用户高频问到的问题和对应的答案对构成，为了适应不同用户对同一问题的不同问法，每个标准问题都会泛化几个相似问题。随着业务的发展，知识库的问题集需要不断更新以提高客服机器人的回答率。Intelligent customer service robots are usually composed of modules such as automatic speech recognition (ASR), intention recognition, question and answer modules, knowledge base management, knowledge graphs, dialogue management, text generation, and text to speech (TTS). Among them, the knowledge base-based FAQ (Frequently Asked Questions, common question items and answers to corresponding questions) The Q&A module provides users with satisfactory answers by querying standard questions matching user questions in the knowledge base, which is the most important module in intelligent customer service robots. The knowledge base is an important part of the FAQ module. It is composed of frequently asked questions by users and corresponding answer pairs. In order to adapt to different users’ different ways of asking the same question, each standard question will be generalized to several similarities. problem. With the development of the business, the question set of the knowledge base needs to be continuously updated to improve the answer rate of the customer service robot.
目前，为了减少维护运营人员的工作量，一般是通过聚类技术把最近机客服机器人未响应的问题进行聚类，然后由运营人员依赖聚类结果和知识库的情况，整理出新的标准问题和对应的相似问题加入到该知识库中以对问题集进行更新。然而，发明人意识到，随着业务的发展会持续不断的涌现出大量新的业务问题，仅基于聚类算法针对未响应问题进行聚类的方式，已经无法聚类得出高质量的问题来对问题集进行更新。At present, in order to reduce the workload of maintenance and operation personnel, clustering technology is generally used to cluster the problems that the nearest machine customer service robot has not responded to, and then the operation personnel rely on the clustering results and the knowledge base to sort out new standard problems. The corresponding similar questions are added to the knowledge base to update the question set. However, the inventor realized that with the development of the business, a large number of new business problems will continue to emerge. It is no longer possible to cluster high-quality problems based on the clustering algorithm for clustering unresponsive problems. Update the problem set.
技术解决方案Technical solutions
本申请的主要目的在于提供一种客服问题的更新方法、系统、终端设备及计算机存储介质，旨在解决现有仅基于聚类算法针对未响应问题进行聚类的方式，无法聚类得出高质量的问题对智能客服机器人的问题集进行更新的技术问题。The main purpose of this application is to provide an update method, system, terminal device and computer storage medium for customer service problems, which aims to solve the existing method of clustering based on clustering algorithm for unresponsive problems, which cannot be clustered to obtain high The quality issue is a technical issue that updates the problem set of the intelligent customer service robot.
为实现上述目的，本申请实施例提供一种客服问题的更新方法，所述客服问题的更新方法包括：In order to achieve the foregoing objective, an embodiment of the present application provides a method for updating customer service questions, and the method for updating customer service questions includes:
根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
根据所述聚类结果生成新的客服问题以对所述客服问题集进行更新。A new customer service question is generated according to the clustering result to update the customer service question set.
此外，为实现上述目的，本申请还提供一种客服问题的更新系统，所述客服问题的更新系统包括：In addition, in order to achieve the above-mentioned purpose, this application also provides a system for updating customer service questions, and the system for updating customer service questions includes:
构建模块，用于根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；The construction module is used to construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
学习模块，用于基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；The learning module is configured to train a similarity model based on the training data, and use the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
聚类模块，用于基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；A clustering module, configured to perform a clustering operation based on the clustering parameters to obtain a clustering result of the unresponsive customer service question;
更新模块，用于根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新The update module is used to generate new customer service questions according to the clustering result to update the preset customer service question set
此外，为实现上述目的，本申请还提供一种终端设备，所述终端设备包括：存储器、处理器，通信总线以及存储在所述存储器上的客服问题的更新程序，所述通信总线用于实现处理器与存储器间的通信连接；所述处理器用于执行所述客服问题的更新程序，以实现以下步骤：In addition, in order to achieve the above object, this application also provides a terminal device, the terminal device including: a memory, a processor, a communication bus, and an update program for customer service questions stored on the memory, and the communication bus is used to implement The communication connection between the processor and the memory; the processor is used to execute the update program of the customer service question, so as to implement the following steps:
根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。A new customer service question is generated according to the clustering result to update the preset customer service question set.
此外，为实现上述目的，本申请还提供一种计算机存储介质，所述计算机存储介质存储有一个或者一个以上程序，所述一个或者一个以上程序可被一个或者一个以上的处理器执行以用于：In addition, in order to achieve the above object, the present application also provides a computer storage medium that stores one or more programs, and the one or more programs can be executed by one or more processors for :
根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
根据所述聚类结果生成新的客服问题以对所述客服问题集进行更新。A new customer service question is generated according to the clustering result to update the customer service question set.
本申请提供的客服问题的更新方法、系统、终端设备以及计算可读存储介质，通过根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。The method, system, terminal device, and computer-readable storage medium for updating customer service questions provided in this application construct question pairs based on customer service questions in a preset customer service question set, and combine the question pairs into training data; based on the training Data training similarity model, and use the training convergence similarity model to calculate the clustering parameters of the unresponsive customer service question; perform clustering operations based on the clustering parameters to obtain the clustering result of the unresponsive customer service question; according to the aggregation The class result generates a new customer service question to update the preset customer service question set.
本申请基于智能客服机器人从预设客服问题集提取客服问题来构建问题对，然后基于问题对进行对应组合得到训练数据，利用该训练数据对相似度模型进行训练，然后基于训练收敛的相似度模型计算客服机器人未响应客服问题的聚类参数，以基于该聚类参数针对该未响应客服问题进行聚类后进行后续更新客服问题集的过程，从而实现了，在简单聚类问题得到新问题的基础上，针对知识库客服问题的变化动态的对未响应客服问题的聚类参数进行自适应学习，从而在聚类时得到更加具有价值的高质量问题来更新客服问题集，提高了智能客服机器人的问题回答效率。This application is based on the intelligent customer service robot extracting customer service questions from the preset customer service question set to construct question pairs, and then correspondingly combining the question pairs to obtain training data, using the training data to train the similarity model, and then training the convergent similarity model Calculate the clustering parameters of the customer service robot not responding to the customer service question, and then perform the subsequent update process of the customer service question set after clustering the non-responsive customer service question based on the clustering parameter, thereby realizing that the new question is obtained in the simple clustering problem Based on the dynamic changes of the knowledge base customer service questions, the clustering parameters of the unresponsive customer service questions are adaptively learned, so that more valuable high-quality questions are obtained during clustering to update the customer service question set, and the intelligent customer service robot is improved. The question is answered efficiently.
附图说明Description of the drawings
图1为本申请实施例方法涉及的终端设备硬件运行环境的结构示意图;Figure 1 is a schematic structural diagram of a terminal device hardware operating environment involved in a method according to an embodiment of the application;
图2为本申请客服问题的更新方法一实施例的流程示意图；FIG. 2 is a schematic flowchart of an embodiment of a method for updating a customer service question application;
图3为本申请客服问题的更新系统的功能模块示意图。Figure 3 is a schematic diagram of the functional modules of the update system for applying for customer service issues.
本发明的实施方式Embodiments of the present invention
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.
如图1所示，图1是本申请实施例方案涉及的终端设备硬件运行环境的设备结构示意图。As shown in FIG. 1, FIG. 1 is a schematic diagram of the device structure of the terminal device hardware operating environment involved in the solution of the embodiment of the present application.
本申请实施例终端设备可以是智能客服机器人，也可以是PC、智能手机、平板电脑和便携计算机等终端设备。The terminal device in the embodiment of the present application may be a smart customer service robot, or a terminal device such as a PC, a smart phone, a tablet computer, and a portable computer.
如图1所示，该终端设备可以包括：处理器1001，例如CPU，通信总线1002，用户接口1003，网络接口1004，存储器1005。其中，通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏（Display）、输入单元比如键盘（Keyboard），可选的用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口（如WI-FI接口）。存储器1005可以是高速RAM存储器，也可以是稳定的存储器（non-volatile memory），例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。As shown in FIG. 1, the terminal device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to implement connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a magnetic disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.
该终端设备还可以包括摄像头、RF（Radio Frequency，射频）电路，传感器、音频电路、WiFi模块等等。其中，传感器比如光传感器、运动传感器以及其他传感器。具体地，光传感器可包括环境光传感器及接近传感器，其中，环境光传感器可根据环境光线的明暗来调节显示屏的亮度，接近传感器可在设备移动到耳边时，关闭显示屏和/或背光。作为运动传感器的一种，重力加速度传感器可检测各个方向上（一般为三轴）加速度的大小，静止时可检测出重力的大小及方向，可用于识别设备姿态的应用（比如横竖屏切换、相关游戏、磁力计姿态校准）、振动识别相关功能（比如计步器、敲击）等；当然，移动终端还可配置陀螺仪、气压计、湿度计、温度计、红外线传感器等其他传感器，在此不再赘述。The terminal equipment may also include a camera, RF (Radio Frequency (radio frequency) circuits, sensors, audio circuits, WiFi modules, etc. Among them, sensors such as light sensors, motion sensors and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, where the ambient light sensor can adjust the brightness of the display screen according to the brightness of the ambient light, and the proximity sensor can turn off the display screen and/or backlight when the device is moved to the ear . As a kind of motion sensor, the gravity acceleration sensor can detect the magnitude of acceleration in various directions (usually three-axis), and can detect the magnitude and direction of gravity when it is stationary. It can be used to identify the application of the device's posture (such as horizontal and vertical screen switching, related Games, magnetometer attitude calibration), vibration recognition related functions (such as pedometer, percussion), etc.; of course, mobile terminals can also be equipped with other sensors such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc. Go into details again.
本领域技术人员可以理解，图1中示出的设备结构并不构成对终端设备的限定，在其它实施方式当中，终端设备还可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。Those skilled in the art can understand that the device structure shown in FIG. 1 does not constitute a limitation on the terminal device. In other implementation manners, the terminal device may also include more or less components than those shown in the figure, or a combination of certain components. Components, or different component arrangements.
如图1所示，作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及客服问题的更新程序。As shown in FIG. 1, the memory 1005, which is a computer storage medium, may include an operating system, a network communication module, a user interface module, and an update program for customer service issues.
在图1所示的终端设备中，网络接口1004主要用于连接后台服务器，与后台服务器进行数据通信；用户接口1003主要用于连接客户端（用户端），与客户端进行数据通信；而处理器1001可以用于调用存储器1005中存储的客服问题的更新程序，并执行以下步骤：In the terminal device shown in FIG. 1, the network interface 1004 is mainly used to connect to the back-end server and communicate with the back-end server; the user interface 1003 is mainly used to connect to the client (user side) and communicate with the client; and the processing The device 1001 can be used to call the update program for customer service questions stored in the memory 1005 and execute the following steps:
根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。A new customer service question is generated according to the clustering result to update the preset customer service question set.
进一步地，所述客服问题包括标准问题和相似问题，处理器1001可以用于调用存储器1005中存储的客服问题的更新程序，还执行以下步骤：Further, the customer service questions include standard questions and similar questions, and the processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
从预设客服问题集中提取各标准问题和各所述标准问题各自对应的各相似问题；Extract each standard question and each similar question corresponding to each said standard question from the set of preset customer service questions;
根据各所述标准问题中的第一标准问题，和所述第一标准问题对应的各相似问题构建得到各相似问题对；According to the first standard problem among the standard problems, each similar problem corresponding to the first standard problem is constructed to obtain each similar problem pair;
根据所述第一标准问题，和各所述标准问题中的第二标准问题对应的各相似问题构建得到各非相似问题对，其中，所述第一标准问题与所述第二标准问题不相同；According to the first standard problem, each similar problem corresponding to the second standard problem in each standard problem is constructed to obtain each non-similar problem pair, wherein the first standard problem is different from the second standard problem ；
将各所述相似问题对与各所述非相似问题对进行对应组合形成各训练数据。Each pair of similar questions is correspondingly combined with each pair of non-similar questions to form each training data.
进一步地，处理器1001可以用于调用存储器1005中存储的客服问题的更新程序，还执行以下步骤：Further, the processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
将所述训练数据切分为第一数据集和第二数据集；Dividing the training data into a first data set and a second data set;
利用所述第一数据集对所述相似度模型进行训练，并利用所述第二数据集对经过训练的相似度模型进行验证；Using the first data set to train the similarity model, and using the second data set to verify the trained similarity model;
将通过验证的相似度模型标记为训练收敛的相似度模型，并将训练收敛的相似度模型存储在区块链中。Mark the verified similarity model as a training converged similarity model, and store the trained convergent similarity model in the blockchain.
进一步地，所述聚类参数包括：问题句子的向量表示和相似度值，处理器1001可以用于调用存储器1005中存储的客服问题的更新程序，还执行以下步骤：Further, the clustering parameters include: the vector representation and similarity value of the question sentence. The processor 1001 may be used to call the update program of the customer service question stored in the memory 1005, and further execute the following steps:
获取所述未响应客服问题的问题句子和所述问题句子的复制句子，并将所述问题句子和所述复制句子输入训练收敛的相似度模型；Acquiring the question sentence that did not respond to the customer service question and the duplicate sentence of the question sentence, and inputting the question sentence and the duplicate sentence into a training convergent similarity model;
获取训练收敛的相似度模型计算所述问题句子和所述复制句子之间的相似度值的向量平均值；Acquiring a similarity model that has been trained to converge and calculating a vector average of similarity values between the question sentence and the copied sentence;
将所述向量平均值作为所述问题句子的向量表示；Taking the vector average value as the vector representation of the question sentence;
或者，遍历所述问题句子构建句子对，并将所述句子对输入训练收敛的相似度模型；Or, traverse the question sentence to construct a sentence pair, and input the sentence pair to train a convergent similarity model;
获取训练收敛的相似度模型对所述句子对进行相似度值计算后输出的相似度值，并将所述相似度值作为所述问题句子的相似度值。Obtain the similarity value output after the similarity value calculation of the sentence pair is performed by the trained convergent similarity model, and the similarity value is used as the similarity value of the question sentence.
进一步地，处理器1001可以用于调用存储器1005中存储的客服问题的更新程序，还执行以下步骤：Further, the processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
调用预设基于距离的聚类算法，利用所述相似度值对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果；Calling a preset distance-based clustering algorithm, clustering the unresponsive customer service question by using the similarity value to obtain a clustering result of the unresponsive customer service question;
或者，调用预设基于句子表示的聚类算法，利用所述向量表示对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果。Alternatively, a preset clustering algorithm based on sentence representation is invoked, and the vector representation is used to cluster the unresponsive customer service questions to obtain a clustering result of the unresponsive customer service questions.
进一步地，处理器1001可以用于调用存储器1005中存储的客服问题的更新程序，还执行以下步骤：Further, the processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
根据所述聚类结果和所述客服问题集生成新的标准问题和所述新的标准问题对应的相似问题；Generating a new standard question and similar questions corresponding to the new standard question according to the clustering result and the customer service question set;
为所述新的标准问题和所述新的标准问题对应的相似问题配置各自对应的服务答案；Configure respective corresponding service answers for the new standard question and similar questions corresponding to the new standard question;
将所述新的标准问题、所述新的标准问题对应的相似问题以及所述服务答案，添加至所述预设客服问题集以对所述预设客服问题集进行更新。The new standard question, similar questions corresponding to the new standard question, and the service answer are added to the preset customer service question set to update the preset customer service question set.
进一步地，处理器1001可以用于调用存储器1005中存储的客服问题的更新程序，还执行以下步骤：Further, the processor 1001 may be used to call an update program for customer service questions stored in the memory 1005, and further execute the following steps:
间隔预设时长采集所述未响应客服问题，并将采集到的所述未响应客服问题存储在区块链中。Collect the unresponsive customer service questions at intervals of a preset length of time, and store the collected unresponsive customer service questions in the blockchain.
本申请客服问题的更新方法所涉及终端设备的具体实施例与下述客服问题的更新方法的各具体实施例基本相同，在此不作赘述。The specific embodiments of the terminal equipment involved in the method for updating customer service questions of this application are basically the same as the specific embodiments of the following method for updating customer service questions, and will not be repeated here.
本申请提供一种客服问题的更新方法。This application provides a method for updating customer service questions.
请参照图2，图2为本申请客服问题的更新方法第一实施例的流程示意图，本申请客服问题的更新方法第一实施例应用于智能客服机器人，在本实施例中，该客服问题的更新方法包括：Please refer to Figure 2. Figure 2 is a schematic flow chart of the first embodiment of the method for updating customer service questions of the application. The first embodiment of the method for updating customer service questions of this application is applied to a smart customer service robot. Update methods include:
步骤S100，根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；In step S100, a question pair is constructed according to the customer service questions in the preset customer service question set, and the question pair is combined into training data;
智能客服机器人从预设客服问题集中提取客服问题，然后基于该客服问题构建各问题对，最后将该各问题对进行一对一的对应组合形成训练数据。The intelligent customer service robot extracts customer service questions from a set of preset customer service questions, then constructs each question pair based on the customer service question, and finally performs a one-to-one corresponding combination of each question pair to form training data.
需要说明的是，在本实施例中，智能客服机器人中的FAQ问答模块具有知识库，该知识库为FAQ问答模块重要的组成部分。预设客服问题集即为该知识库中保存着的客服问题集，该客服问题集由众多标准问题和各标准问题各自对应的一个或者多个相似问题所组成，从而，智能客服机器人通过FAQ问答模块在该知识库中查询与用户提出的客服问题相匹配的标准问题，从而为用户提供满意的答案。此外，智能客服机器人将各问题对进行一对一的对应组合形成的训练数据可以为一个数据集，该数据集由多条训练数据组成。It should be noted that, in this embodiment, the FAQ question and answer module in the intelligent customer service robot has a knowledge base, which is an important part of the FAQ question and answer module. The preset customer service question set is the customer service question set saved in the knowledge base. The customer service question set consists of many standard questions and one or more similar questions corresponding to each standard question. Therefore, the intelligent customer service robot asks questions through FAQs. The module searches the knowledge base for standard questions that match the customer's customer service questions, so as to provide users with satisfactory answers. In addition, the training data formed by a one-to-one corresponding combination of each question pair by the intelligent customer service robot may be a data set, which is composed of multiple pieces of training data.
具体地，例如，智能客服机器人定期或者随机的发起针对客服问题集进行更新的操作，智能客服机器人访问FAQ问答模块的知识库，从该知识库中所存贮的客服问题集中提取标准问题和相似问题，然后基于该标准问题和相似问题构建问题对，最后按照一对一的规则通过对该问题对进行对应组合形成多条训练数据以组成数据集。Specifically, for example, the intelligent customer service robot periodically or randomly initiates an update operation for the customer service question set. The intelligent customer service robot accesses the knowledge base of the FAQ module, and extracts standard questions and similarities from the customer service questions stored in the knowledge base. Questions, and then construct a question pair based on the standard question and similar questions, and finally form multiple pieces of training data by correspondingly combining the question pair according to the one-to-one rule to form a data set.
进一步地，在一种实施例中，步骤S100中，可以包括：Further, in an embodiment, step S100 may include:
步骤S101，从预设客服问题集中提取各标准问题和各所述标准问题各自对应的各相似问题；Step S101, extract each standard question and each similar question corresponding to each standard question from the preset customer service question set;
步骤S102，根据各所述标准问题中的第一标准问题，和所述第一标准问题对应的各相似问题构建得到各相似问题对；Step S102, constructing each similar question pair corresponding to the first standard question according to the first standard question among the standard questions;
具体地，例如，智能客服机器人在发起针对客服问题集进行更新的操作之后，基于访问FAQ问答模块的知识库，以从该知识库中所存贮的客服问题集中提取多个标准问题和该多个标准问题各自所对应的相似问题，然后，智能客服机器人将该多个标准问题当中的任意一个标准问题与该标准问题的多个相似问题进行随机组合，从而组件得到多个相似问题对。Specifically, for example, after the intelligent customer service robot initiates an update operation for the customer service question set, it accesses the knowledge base of the FAQ module to extract multiple standard questions and the multiple questions from the customer service question set stored in the knowledge base. Each standard question corresponds to a similar question, and then, the intelligent customer service robot randomly combines any one of the multiple standard questions with multiple similar questions of the standard question, so that the component obtains multiple similar question pairs.
步骤S103，根据所述第一标准问题，和各所述标准问题中的第二标准问题对应的各相似问题构建得到各非相似问题对；Step S103, constructing each similar question corresponding to the second standard question in each standard question according to the first standard question to obtain each non-similar question pair;
需要说明的是，在本实施例中，其中，第一标准问题与所述第二标准问题不相同。It should be noted that, in this embodiment, the first standard problem is different from the second standard problem.
具体地，例如，在智能客服机器人基于访问FAQ问答模块的知识库，并从该知识库中所存贮的客服问题集中提取多个标准问题和该多个标准问题各自所对应的相似问题之后，智能客服机器人可以在组建得到相似问题对的同时，同步执行（当然也可以异步执行）组建非相似问题对的操作，即，智能客服机器人将多个标准问题当中的任意一个标准问题，与该多个标准问题当中除该标准问题之外的其他任意一个标准问题的多个相似问题进行随机组合，从而得到多个非相似问题对。Specifically, for example, after the intelligent customer service robot accesses the knowledge base of the FAQ question and answer module, and extracts multiple standard questions and similar questions corresponding to each of the multiple standard questions from the customer service questions stored in the knowledge base, The intelligent customer service robot can simultaneously execute (or asynchronously, of course) the operation of constructing non-similar question pairs while forming similar problem pairs. That is, the intelligent customer service robot can combine any one of the multiple standard questions with the same one. Among the standard questions, multiple similar questions of any standard question except the standard question are randomly combined to obtain multiple non-similar question pairs.
步骤S104，将各所述相似问题对与各所述非相似问题对进行对应组合形成各训练数据。Step S104, correspondingly combining each pair of similar questions with each pair of non-similar questions to form each training data.
具体地，例如，智能客服机器人将组合形成的该多个相似问题对和多个非相似问题对，按照一个相似问题对和一个非相似问题对的规则进行对应组合从而形成一条训练数据，然后将按照上述规则构建得到的多条训练数据组成数据集。Specifically, for example, the intelligent customer service robot combines the multiple similar question pairs and multiple dissimilar question pairs to form a piece of training data by correspondingly combining the rules of a similar question pair and a dissimilar question pair. Multiple pieces of training data constructed according to the above rules constitute a data set.
进一步地，在另一种实施例中，智能客服机器人在基于多个相似问题对和多个非相似问题对组合形成训练数据时，还可以按照一个相似问题对和多个非相似问题对来组合形成一条训练数据，从而构建得到多条训练数据以组成数据集。Further, in another embodiment, when the intelligent customer service robot forms training data based on the combination of multiple similar question pairs and multiple dissimilar question pairs, it can also combine a similar question pair and multiple dissimilar question pairs. A piece of training data is formed, and multiple pieces of training data are constructed to form a data set.
需要说明的是，在本实施例中，组成同一条训练数据的相似问题对和非相似问题对各自的标准问题相同，可以理解的时，基于实际应用的不同设计需要，组成同一条训练数据的相似问题对和非相似问题对各自的标准问题也可以不相同，本申请客服问题的更新方法并不对该训练数据中相似问题对和非相似问题对的组合形式进行具体限定。It should be noted that, in this embodiment, similar question pairs and non-similar questions that make up the same piece of training data have the same standard questions. When it is understandable, based on the different design needs of practical applications, the same piece of training data is made up. Similar question pairs and dissimilar question pairs may also be different for their respective standard questions. The method for updating customer service questions in this application does not specifically limit the combination of similar question pairs and dissimilar question pairs in the training data.
步骤S200，基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Step S200: Train a similarity model based on the training data, and use the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
智能客服机器人在从预设知识库中提取客服问题集，然后基于该客服问题集构建得到训练数据之后，智能客服机器人利用该训练数据对相似度模型进行训练直到该相似度模型收敛，然后，智能客服机器人利用基于该训练数据训练到收敛的相似度模型，针对未作出回答的未响应客服问题进行计算得到后续针对该未响应客服问题进行聚类的聚类参数。After the intelligent customer service robot extracts the customer service question set from the preset knowledge base, and then builds the training data based on the customer service question set, the intelligent customer service robot uses the training data to train the similarity model until the similarity model converges, and then the intelligent The customer service robot uses the similarity model trained to convergence based on the training data, and calculates the unanswered customer service question that has not been answered to obtain the clustering parameters for subsequent clustering of the non-responsive customer service question.
需要说明的是，在本实施例中，相似度模型具体可以为bert模型（Bidirectional Encoder Representations from Transformers，是近期Google提出的效果非常好的通用预训练语言表示模型）、xlnet（一种通用的自回归预训练方法）模型、siamesecnn（孪生卷积）网络、siameselstm（孪生）网络等模型中的任意一种。此外，未响应客服问题为智能客服机器人在整个运行过程中，针对用户所提出的客服问题未能够成功匹配到标准问题，从而没有得到答案以向用户输出时，智能客服机器人记录的该客服问题（即，未响应客服问题为智能客服机器人未做出回答的客服问题）。It should be noted that, in this embodiment, the similarity model may specifically be a bert model (Bidirectional Encoder Representations from Transformers is a very effective general pre-training language representation model proposed by Google recently), xlnet (a general autoregressive pre-training method) model, siamesecnn (twin convolution) network, siameselstm (twin) network, etc. Any of the models. In addition, non-response customer service question refers to the customer service question recorded by the intelligent customer service robot when the customer service question raised by the user cannot be successfully matched to the standard question during the entire operation process of the intelligent customer service robot, so that the answer is not obtained to output to the user ( That is, unresponsive customer service questions are customer service questions that the intelligent customer service robot has not answered).
具体地，例如，智能客服机器人以bert模型作为相似度模型，智能客服机器人在基于多个相似问题对和多个非相似问题对构建得到多条训练数据之后，将该多条训练数据中的部分训练数据作为训练样本输入至bert模型以对该bert模型进行训练，直到智能客服机器人校验到该bert模型收敛之后，利用已经训练收敛的该bert模型，计算预先记录的一个或者多个未响应客服问题的聚类参数，以便后续基于该聚类参数对该未响应客服问题进行聚类运算。Specifically, for example, the intelligent customer service robot uses the bert model as the similarity model. After the intelligent customer service robot obtains multiple pieces of training data based on multiple similar question pairs and multiple non-similar question pairs, part of the multiple pieces of training data The training data is input to the bert model as training samples to train the bert model. After the intelligent customer service robot verifies that the bert model has converged, the bert model that has been trained and converged is used to calculate one or more pre-recorded unresponsive customer services The clustering parameter of the question, so as to perform a clustering operation on the unresponsive customer service question based on the clustering parameter.
步骤S300，基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Step S300, performing a clustering operation based on the clustering parameters to obtain a clustering result of the unresponsive customer service question;
智能客服机器人通过调用现有成熟的聚类算法，在基于训练收敛的相似度模型计算输出未响应客服问题的聚类参数之后，利用该距离参数对该未响应客服问题进行聚类运算，从而得到该未响应客服问题的聚类结果。The intelligent customer service robot uses the existing mature clustering algorithm to calculate and output the clustering parameters of the unresponsive customer service question based on the training convergence similarity model, and then uses the distance parameter to perform the clustering operation on the unresponsive customer service question, thereby obtaining The clustering result that did not respond to customer service questions.
步骤S400，根据所述聚类结果生成新的客服问题以对所述客服问题集进行更新。In step S400, a new customer service question is generated according to the clustering result to update the customer service question set.
智能客服机器人在通过聚类运算得到未响应客服问题的聚类结果之后，自动根据该聚类结果和FAQ问答模块的知识库中贮存的客服问题集，整理得到新的客服问题，然后将该新的客服问题添加至该知识库中的客服问题集当中，从而完成对该客服问题集的更新。After the intelligent customer service robot obtains the clustering result of the unresponsive customer service question through the clustering operation, it automatically sorts out the new customer service question according to the clustering result and the customer service question set stored in the knowledge base of the FAQ module. The customer service question is added to the customer service question set in the knowledge base, thereby completing the update of the customer service question set.
需要说明的是，在本实施例中，智能客服机器人还可以在通过聚类运算得到未响应客服问题的聚类结果之后，将该聚类结果输出至该智能客服机器人的运营人员，从而辅助该运营人员根据该聚类结果和FAQ问答模块的知识库中贮存的客服问题集，整理得到新的客服问题，然后智能客服机器人接收运营人员输入的新的客服问题，并将该新的客服问题添加至该知识库中的客服问题集当中，从而完成对该客服问题集的更新。It should be noted that, in this embodiment, the intelligent customer service robot may also output the clustering result to the operator of the intelligent customer service robot after obtaining a clustering result that does not respond to customer service questions through a clustering operation, thereby assisting the According to the clustering result and the customer service question set stored in the knowledge base of the FAQ module, the operator sorts out new customer service questions, and then the intelligent customer service robot receives the new customer service question input by the operator, and adds the new customer service question To the customer service question set in the knowledge base to complete the update of the customer service question set.
在本实施例中，通过智能客服机器人从预设知识库中提取客服问题集，然后基于该客服问题集构建训练数据；在从预设知识库中提取客服问题集，然后基于该客服问题集构建得到训练数据之后，智能客服机器人利用该训练数据对相似度模型进行训练直到该相似度模型收敛，然后，智能客服机器人利用基于该训练数据训练到收敛的相似度模型，针对未作出回答的未响应客服问题进行计算得到后续针对该未响应客服问题进行聚类的聚类参数；在基于训练收敛的相似度模型计算输出未响应客服问题的聚类参数之后，智能客服机器人利用该距离参数对该未响应客服问题进行聚类运算，从而得到该未响应客服问题的聚类结果；最后，智能客服机器人自动根据该聚类结果和FAQ问答模块的知识库中贮存的客服问题集，整理得到新的客服问题，然后将该新的客服问题添加至该知识库中的客服问题集当中，从而完成对该客服问题集的更新。In this embodiment, an intelligent customer service robot extracts a customer service question set from a preset knowledge base, and then constructs training data based on the customer service question set; extracts a customer service question set from the preset knowledge base, and then constructs a customer service question set based on the customer service question set After obtaining the training data, the intelligent customer service robot uses the training data to train the similarity model until the similarity model converges. Then, the intelligent customer service robot uses the training data to train the similarity model to converge, and responds to unanswered unanswered questions. The customer service question is calculated to obtain the subsequent clustering parameters for the unresponsive customer service question; after the clustering parameter of the unresponsive customer service question is calculated and output based on the similarity model of the training convergence, the intelligent customer service robot uses the distance parameter to obtain the clustering parameter for the unresponsive customer service question. Perform clustering operations in response to customer service questions to obtain the clustering results of the unresponsive customer service questions; finally, the intelligent customer service robot automatically sorts and obtains new customer service based on the clustering results and the customer service question set stored in the knowledge base of the FAQ module Questions, and then add the new customer service question to the customer service question set in the knowledge base, thereby completing the update of the customer service question set.
实现了基于智能客服机器人从预设客服问题集提取客服问题来构建问题对，然后基于问题对进行对应组合得到训练数据，利用该训练数据对相似度模型进行训练，然后基于训练收敛的相似度模型计算客服机器人未响应客服问题的聚类参数，以基于该聚类参数针对该未响应客服问题进行聚类后进行后续更新客服问题集的过程，从而实现了，在简单聚类问题得到新问题的基础上，针对知识库客服问题的变化动态的对未响应客服问题的聚类参数进行自适应学习，从而在聚类时得到更加具有价值的高质量问题来更新客服问题集，提高了智能客服机器人的问题回答效率。Realized that based on the intelligent customer service robot extracting customer service questions from the preset customer service question set to construct the question pair, and then based on the corresponding combination of the question pair to obtain the training data, using the training data to train the similarity model, and then based on the training convergence similarity model Calculate the clustering parameters of the customer service robot not responding to the customer service question, and then perform the subsequent update process of the customer service question set after clustering the non-response customer service question based on the clustering parameter, thereby realizing that the new question is obtained in the simple clustering problem Based on the dynamic changes of the knowledge base customer service questions, the clustering parameters of the unresponsive customer service questions are adaptively learned, so that more valuable high-quality questions are obtained during clustering to update the customer service question set, and the intelligent customer service robot is improved. The question is answered efficiently.
进一步地，在本申请客服问题的更新方法第一实施例的基础上，提出客服问题的更新方法第二实施例，在本实施例中，上述步骤S200中，“基于所述训练数据训练相似度模型”的步骤，可以包括：Further, on the basis of the first embodiment of the method for updating customer service questions of this application, a second embodiment of the method for updating customer service questions is proposed. In this embodiment, in the above step S200, "training similarity based on the training data The steps of "model" can include:
步骤S201，将所述训练数据切分为第一数据集和第二数据集；Step S201, dividing the training data into a first data set and a second data set;
智能客服机器人在基于预设知识库，以从该预设知识库中所存贮的客服问题集中提取标准问题和该标准问题各自所对应的相似问题，然后组合得到相似问题对以及非相似问题对，并将该相似问题对和非相似问题对构建得到训练数据的数据集之后，智能客服机器人按照预设比例或者随机对该数据集进行切分，从而得到第一数据集和第二数据集。Based on a preset knowledge base, the intelligent customer service robot extracts standard questions and similar questions corresponding to the standard questions from the customer service questions stored in the preset knowledge base, and then combines pairs of similar questions and pairs of dissimilar questions. After constructing a data set of training data for the similar question pair and the dissimilar question pair, the intelligent customer service robot divides the data set according to a preset ratio or randomly to obtain the first data set and the second data set.
需要说明的是，在本实施例中，预设比例可以为运营人员基于设计需要设置的比例，应当理解的是，基于实际应用的不同设计需要，该预设比例可以被设置为任意数值的比例，本申请客服问题的更新方法并不对该预设比例的大小进行具体的限定。It should be noted that, in this embodiment, the preset ratio can be a ratio set by the operator based on design needs. It should be understood that the preset ratio can be set to any numerical ratio based on different design needs of actual applications. The method for updating customer service questions in this application does not specifically limit the size of the preset ratio.
步骤S202，利用所述第一数据集对所述相似度模型进行训练，并利用所述第二数据集对经过训练的相似度模型进行验证；Step S202, using the first data set to train the similarity model, and using the second data set to verify the trained similarity model;
智能客服机器人在按照预设比例或者随机的将训练数据的数据集切分为第一数据集和第二数据集之后，智能客服机器人将第一数据集中的训练数据输入至相似度模型以对该相似度模型进行训练，并在利用该训练数据针对相似度模型的每一轮训练结束之后，随即将第二数据集中的训练数据输入至该相似度训练模型，以根据该相似度模型的输出结果验证该相似度模型是否在当前轮的训练结束之后即已经收敛。After the intelligent customer service robot divides the training data data set into the first data set and the second data set according to a preset ratio or randomly, the intelligent customer service robot inputs the training data in the first data set into the similarity model to compare the training data. The similarity model is trained, and after each round of training of the similarity model using the training data is completed, the training data in the second data set is then input to the similarity training model to according to the output result of the similarity model Verify whether the similarity model has converged after the end of the current round of training.
进一步地，在另一种实施例中，智能客服机器人也可以将第二数据集中的训练数据输入至相似度模型以对该相似度模型进行训练，并在利用该训练数据针对相似度模型的每一轮训练结束之后，随即将第一数据集中的训练数据输入至该相似度训练模型，以根据该相似度模型的输出结果验证该相似度模型是否在当前轮的训练结束之后即已经收敛。Further, in another embodiment, the intelligent customer service robot can also input the training data in the second data set to the similarity model to train the similarity model, and use the training data to target each similarity model. After one round of training is over, the training data in the first data set is then input to the similarity training model to verify whether the similarity model has converged after the end of the current round of training according to the output result of the similarity model.
步骤S203，将通过验证的相似度模型标记为训练收敛的相似度模型，并将训练收敛的相似度模型存储在区块链中。Step S203: Mark the verified similarity model as a training converged similarity model, and store the trained convergent similarity model in the blockchain.
若智能客服机器人在将第二数据集中的训练数据，输入至经过训练之后的相似度训练模型，并根据该相似度模型的输出结果验证该相似度模型已经训练收敛之后，确认利用第二数据集针对该相似度模型的验证已经通过，并即时将该相似度模型标记为训练收敛的相似度模型，然后将该训练收敛的相似度模型存储在一区块链的节点当中。If the intelligent customer service robot inputs the training data in the second data set to the similarity training model after training, and verifies that the similarity model has been trained and converged according to the output result of the similarity model, confirm to use the second data set The verification of the similarity model has passed, and the similarity model is immediately marked as a training convergent similarity model, and then the training converged similarity model is stored in a node of the blockchain.
需要说明的是，在本实施例中，为了保证训练收敛的相似度模型不会被错误修改或者移除，可以将训练收敛的相似度模型存储于一区块链的节点中，确保了该索引标签的安全性，更确保了后续在基于训练收敛的相似度模型针对智能客服机器人的未响应客服问题进行聚类参数计算的准确性，进一步提升了智能客服机器人自适应学习新业务问题的效率。It should be noted that in this embodiment, in order to ensure that the training converged similarity model will not be incorrectly modified or removed, the trained convergent similarity model can be stored in a node of a blockchain to ensure the index The security of the label further ensures the accuracy of the subsequent clustering parameter calculation of the intelligent customer service robot's unresponsive customer service problems in the similarity model based on training convergence, and further improves the efficiency of the intelligent customer service robot's adaptive learning of new business problems.
在本实施例中，智能客服机器人通过将构建得到的训练数据的数据集切分为第一数据集和第二数据集；然后，利用第一数据集中的训练数据对相似度模型进行训练，并在每一轮次的训练结束之后，利用第二数据集中的训练数据对经过训练的相似度模型进行验证；最后，将通过验证的相似度模型标记为训练收敛的相似度模型，并将该训练收敛的相似度模型存储在区块链中以供后续调用。实现了，根据目前FAQ问答模块的知识库中贮存的客服问题集（标准问题+标准问题的相似问题）构建由相似问题对和非相似问题对组成的数据集，然后在该数据集的基础上训练相似度模型以用以后续自适应学习未响应客服问题的聚类参数，从而保证在针对该未响应客服问题进行聚类时，能够得到更有价值的新问题以对客服问题集进行更新，提高了智能客服机器人的问题回答效率。In this embodiment, the intelligent customer service robot divides the constructed training data data set into a first data set and a second data set; then, the training data in the first data set is used to train the similarity model, and After each round of training is over, the training data in the second data set is used to verify the trained similarity model; finally, the verified similarity model is marked as the training convergence similarity model, and the training The converged similarity model is stored in the blockchain for subsequent recall. Realized, according to the customer service question set (standard question + similar question of standard question) stored in the knowledge base of the current FAQ question and answer module, a data set consisting of similar question pairs and non-similar question pairs is constructed, and then on the basis of this data set The similarity model is trained for subsequent adaptive learning of clustering parameters of unresponsive customer service questions, so as to ensure that when clustering the unresponsive customer service questions, more valuable new questions can be obtained to update the customer service question set. Improved the efficiency of answering questions for intelligent customer service robots.
进一步地，在本申请客服问题的更新方法第一实施例的基础上，提出本申请客服问题的更新方法的第三实施例，在本实施例中，本申请客服问题的更新方法，还可以包括：Further, on the basis of the first embodiment of the method for updating customer service questions of this application, a third embodiment of the method for updating customer service questions of this application is proposed. In this embodiment, the method for updating customer service questions of this application may also include :
步骤A，间隔预设时长采集所述未响应客服问题，并将采集到的所述未响应客服问题存储在区块链中。Step A: Collect the unresponsive customer service questions at intervals of a preset length of time, and store the collected unresponsive customer service questions in the blockchain.
需要说明的是，在本实施例中，智能客服机器人在持续运营的过程中会遭遇到无法向用户输出答案的未响应客服问题，从而智能客服机器人可以预先建立一个临时的存储空间，一旦在遭遇到无法向用户输出答案的未响应客服问题时，便将该未响应客服问题缓存至该存储空间当中。此外，预设时长具体可以为24小时，应当理解的是，基于实际应用的不同设计需要，该预设时长当然也可以为其它任意时长，本申请客服问题的更新方法并不对该预设时长进行具体地限定。It should be noted that, in this embodiment, the intelligent customer service robot will encounter unresponsive customer service questions that cannot output answers to the user during the continuous operation, so the intelligent customer service robot can establish a temporary storage space in advance. When an unresponsive customer service question cannot output an answer to the user, the unresponsive customer service question is cached in the storage space. In addition, the preset duration can be specifically 24 hours. It should be understood that, based on the different design needs of actual applications, the preset duration can of course also be any other duration. The method for updating customer service questions in this application does not apply to the preset duration. Specifically defined.
具体地，例如，智能客服机器人在每天的凌晨0:00，从预先建立的临时存储空间中，获取自该0:00之前的24小时内，智能客服机器人在运营过程中所记录的为对用户所提出的客服问题输出答案的未响应客服问题，然后将该未响应客服问题存储在一区块链的节点当中以供后续调用。Specifically, for example, at 0:00 in the morning every day, the intelligent customer service robot obtains from the pre-established temporary storage space in the 24 hours before the 0:00, what the intelligent customer service robot records during the operation process is the user The non-response customer service question that outputs the answer to the raised customer service question is then stored in a node of the blockchain for subsequent calls.
进一步地，在另一种实施例中，智能客服机器人还可以直接在预先建立的临时存储空间中，对0:00之前的24小时内，智能客服机器人在运营过程中所记录的为对用户所提出的客服问题输出答案的未响应客服问题，进行持久化缓存，以供后续计算该未响应客服问题的聚类参数和进行聚类运算时直接调用。Further, in another embodiment, the intelligent customer service robot can also directly store in the pre-established temporary storage space. In the 24 hours before 0:00, the information recorded by the intelligent customer service robot during the operation process is the user's information. The non-response customer service question with the output answer of the proposed customer service question is persistently cached for subsequent calculation of the clustering parameter of the non-response customer service question and direct call when performing clustering operations.
进一步地，在本实施例中，智能客服机器人利用已经训练收敛的相似度模型计算得到的未响应客服问题的聚类参数，包括但不限于：未响应客服问题的问题句子的向量表示，和该问题句子的相似度值，上述步骤S200中，“利用训练收敛的相似度模型计算未响应客服问题的聚类参数”的步骤，可以包括：Further, in this embodiment, the intelligent customer service robot uses the trained and converged similarity model to calculate the clustering parameters of the unresponsive customer service question, including but not limited to: the vector representation of the question sentence that does not respond to the customer service question, and the For the similarity value of the question sentence, in the above step S200, the step of "using the trained and converged similarity model to calculate the clustering parameters that did not respond to the customer service question" may include:
步骤S204，获取所述未响应客服问题的问题句子和所述问题句子的复制句子，并将所述问题句子和所述复制句子输入训练收敛的相似度模型；Step S204: Obtain the question sentence that does not respond to the customer service question and the duplicate sentence of the question sentence, and input the question sentence and the duplicate sentence into a training convergent similarity model;
智能客服机器人在获取得到未响应客服问题之后，通过现有成熟的语言处理技术读取该未响应客服问题的问题句子，并对该问题句子进行复制得到该问题句子的复制句子，然后，智能客服机器人将该问题句子和该问题句子的复制句子输入至训练收敛的相似度模型当中。After the intelligent customer service robot obtains the unresponsive customer service question, it uses the existing mature language processing technology to read the question sentence that does not respond to the customer service question, and copies the question sentence to obtain the duplicate sentence of the question sentence, and then the intelligent customer service The robot inputs the question sentence and the duplicate sentence of the question sentence into the similarity model that is trained to converge.
具体地，例如，在智能客服机器人将从区块链的节点中获取到一个或者多个未响应客服问题之后，调用现有任意成熟的自然语言处理技术，逐一读取该未响应客服问题的文本作为问题句子，并继续针对该问题句子进行复制得到各问题句子对应的复制句子，然后，智能客服机器人逐一将该问题句子和该问题句子所对应的复制句子，输入到同样从区块链的节点中获取到的已经训练收敛的相似度模型当中。Specifically, for example, after the intelligent customer service robot obtains one or more unresponsive customer service questions from the nodes of the blockchain, it calls any existing mature natural language processing technology to read the text of the unresponsive customer service questions one by one As a question sentence, and continue to copy the question sentence to obtain the copy sentence corresponding to each question sentence, and then the intelligent customer service robot inputs the question sentence and the copy sentence corresponding to the question sentence one by one to the same node from the blockchain Among the similarity models that have been trained and converged obtained in.
步骤S205，获取训练收敛的相似度模型计算所述问题句子和所述复制句子之间的相似度值的向量平均值；Step S205: Obtain a training convergence similarity model and calculate a vector average of similarity values between the question sentence and the copied sentence;
步骤S206，将所述向量平均值作为所述问题句子的向量表示；Step S206: Use the vector average value as a vector representation of the question sentence;
智能客服机器人在将未响应客服问题的问题句子和问题句子对应的复制句子输入至已经训练收敛的相似度模型当中之后，在该相似度模型计算该问题句子与复制句子之间相似度值时，获取该问题句子的向量与复制句子的向量，然后计算该问题句子的向量与复制句子的向量的向量平均值，并将该向量平均值作为未响应客服问题的问题句子的向量表示。After the intelligent customer service robot inputs the question sentence that has not responded to the customer service question and the copied sentence corresponding to the question sentence into the similarity model that has been trained and converged, when the similarity model calculates the similarity value between the question sentence and the copied sentence, Obtain the vector of the question sentence and the vector of the copied sentence, and then calculate the vector average of the vector of the question sentence and the vector of the copied sentence, and use the vector average as the vector of the question sentence that did not respond to the customer service question.
进一步地，在另一种实施例中，上述步骤S200中，“利用训练收敛的相似度模型计算未响应客服问题的聚类参数”的步骤，还可以包括：Further, in another embodiment, in the above step S200, the step of "using the trained and converged similarity model to calculate the clustering parameters that did not respond to the customer service question" may further include:
步骤S207，遍历所述问题句子构建句子对，并将所述句子对输入训练收敛的相似度模型；Step S207, traverse the question sentences to construct sentence pairs, and input the sentence pairs to train a convergent similarity model;
步骤S208，获取训练收敛的相似度模型对所述句子对进行相似度值计算后输出的相似度值，并将所述相似度值作为所述问题句子的相似度值。Step S208: Obtain the similarity value output by the trained and converged similarity model after calculating the similarity value of the sentence pair, and use the similarity value as the similarity value of the question sentence.
智能客服机器人在获取得到未响应客服问题，并通过现有成熟的语言处理技术读取该未响应客服问题的问题句子，遍历该问题句子以按照当前问题句子和与当前问题句子相似的其他问题句子来构建句子对，然后将该句子对输入到已经训练收敛的相似度模型当中，以供该相似度模型针对该句子对计算问题句子与问题句子相似的其他问题句子之间的相似度值。最后，智能客服机器人将该相似度模型基于输入的句子对进行相似度值计算后输出的该相似度值，并将该相似度值作为未响应客服问题的问题句子的相似度值。The intelligent customer service robot obtains the unresponsive customer service question, reads the question sentence that does not respond to the customer service question through the existing mature language processing technology, and traverses the question sentence to follow the current question sentence and other question sentences similar to the current question sentence To construct a sentence pair, and then input the sentence pair into the similarity model that has been trained and converged, so that the similarity model calculates the similarity value between the question sentence and other question sentences that are similar to the question sentence for the sentence pair. Finally, the intelligent customer service robot outputs the similarity value after calculating the similarity value based on the similarity model based on the input sentence pair, and uses the similarity value as the similarity value of the question sentence that does not respond to the customer service question.
进一步地，在本实施例中，上述步骤S300，基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果，可以包括：Further, in this embodiment, in the above step S300, performing a clustering operation based on the clustering parameters to obtain the clustering result of the unresponsive customer service question may include:
步骤S301，调用预设基于距离的聚类算法，利用所述相似度值对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果；Step S301, calling a preset distance-based clustering algorithm, clustering the unresponsive customer service questions by using the similarity value to obtain a clustering result of the unresponsive customer service questions;
智能客服机器人针对利用已经训练收敛的相似度模型输出的未响应客服问题的问题句子的相似度值，调用现有成熟的预设基于距离的聚类算法利用该相似度值对该未响应客服问题机型聚类运算，从而得到该未响应客服问题的聚类结果。The intelligent customer service robot calls the existing mature preset distance-based clustering algorithm based on the similarity value of the question sentence output by the trained and converged similarity model that does not respond to the customer service question, and uses the similarity value to the unresponsive customer service question Model clustering operation to obtain the clustering result of the non-response customer service question.
需要说明的是，在本实施例中，预设基于距离的聚类算法包括但不限于：affinity propagation（一种AP聚类算法，从广义的角度说，affinity propagation属于message-passing algorithms（消息传递算法）的一种）、spectral clustering（谱聚类算法）。It should be noted that, in this embodiment, the preset distance-based clustering algorithm includes, but is not limited to: affinity propagation (an AP clustering algorithm, in a broad sense, affinity Propagation is a type of message-passing algorithms), spectral clustering (spectral clustering algorithm).
进一步地，在本实施例中，上述步骤S300，还可以包括：Further, in this embodiment, the above step S300 may further include:
步骤S302，调用预设基于句子表示的聚类算法，利用所述向量表示对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果。Step S302: Invoke a preset clustering algorithm based on sentence representation, and cluster the unresponsive customer service questions using the vector representation to obtain a clustering result of the unresponsive customer service questions.
智能客服机器人针对利用已经训练收敛的相似度模型，在该相似度模型计算该问题句子与复制句子之间相似度值时，获取该问题句子的向量与复制句子的向量，并计算得到的未响应客服问题的问题句子的向量表示，调用现有成熟的预设基于句子表示的聚类算法利用该向量表示对该未响应客服问题机型聚类运算，从而得到该未响应客服问题的聚类结果。The intelligent customer service robot is aimed at using a similarity model that has been trained and converged. When the similarity model calculates the similarity value between the question sentence and the copied sentence, it obtains the vector of the question sentence and the vector of the copied sentence, and calculates the resulting unresponsiveness The vector representation of the question sentence of the customer service problem, call the existing mature preset clustering algorithm based on sentence representation and use the vector to represent the clustering operation of the model that did not respond to the customer service problem, so as to obtain the clustering result of the non-responsive customer service problem .
需要说明的是，在本实施例中，预设基于句子表示的聚类算法包括但不限于：k-means（k-means clustering algorithm：k均值聚类算法，是一种迭代求解的聚类分析算法）、dbscan（DBSCAN(Density-Based Spatial Clustering of Applications with Noise，是一个比较有代表性的基于密度的聚类算法）以及层次聚类算法。It should be noted that, in this embodiment, the preset clustering algorithm based on sentence representation includes but is not limited to: k-means (k-means clustering algorithm: k-means clustering algorithm is an iterative solution clustering analysis algorithm), dbscan (DBSCAN (Density-Based Spatial Clustering of Applications with Noise, is a more representative density-based clustering algorithm) and hierarchical clustering algorithm.
在本实施例中，智能客服机器人间隔预设时长采集未响应客服问题，并将采集到的未响应客服问题存储在区块链中或者直接针对该未响应客服问题做持久缓存。In this embodiment, the intelligent customer service robot collects unresponsive customer service questions at a preset interval, and stores the collected unresponsive customer service questions in the blockchain or directly caches the unresponsive customer service questions.
在智能客服机器人计算该未响应客服问题的聚类参数（未响应客服问题的问题句子的向量表示和相似度值）时，获取该未响应客服问题的问题句子和问题句子的复制句子，并将问题句子和复制句子输入训练收敛的相似度模型；然后获取训练收敛的相似度模型在计算该问题句子和该复制句子之间的相似度值的向量平均值；并将该向量平均值作为向量表示；或者，通过遍历问题句子以构建句子对，并将句子对输入训练收敛的相似度模型；然后获取训练收敛的相似度模型对句子对进行相似度值计算后输出的相似度值，并将该相似度值作为问题句子的相似度值。此外，When the intelligent customer service robot calculates the clustering parameters of the unresponsive customer service question (the vector representation and similarity value of the question sentence that does not respond to the customer service question), it obtains the question sentence and the duplicate sentence of the question sentence that did not respond to the customer service question, and then The question sentence and the copied sentence are input to the trained convergent similarity model; then the trained convergent similarity model is obtained to calculate the vector average value of the similarity values between the question sentence and the copied sentence; and the vector average value is used as a vector representation ; Or, construct a sentence pair by traversing the question sentence, and input the sentence pair into a convergent similarity model; then obtain the similarity value output after the similarity value of the sentence pair is calculated by the trained convergent similarity model, and the The similarity value is used as the similarity value of the question sentence. In addition,
在智能客服机器人针对该未响应客服问题进行聚类操作时，调用预设基于距离的聚类算法，利用问题句子的相似度值对未响应客服问题进行聚类得到该未响应客服问题的聚类结果；或者，调用预设基于句子表示的聚类算法，利用问题句子的向量表示对未响应客服问题进行聚类得到该未响应客服问题的聚类结果。When the intelligent customer service robot performs a clustering operation for the unresponsive customer service question, it calls the preset distance-based clustering algorithm, and uses the similarity value of the question sentence to cluster the unresponsive customer service question to obtain the cluster of the unresponsive customer service question Result; or, call a preset clustering algorithm based on sentence representation, and use the vector representation of the question sentence to cluster the unresponsive customer service question to obtain the clustering result of the unresponsive customer service question.
实现了，在传统简单针对未响应客服问题进行聚类得到新问题的基础上，进一步针对知识库客服问题的变化动态的对未响应客服问题的聚类参数进行自适应学习，从而在针对未响应客服问题进行聚类时能够得到更加具有价值的高质量问题来更新客服问题，提高了智能客服机器人的问题回答效率。Achieved, based on the traditional simple clustering of unresponsive customer service questions to obtain new questions, the clustering parameters of unresponsive customer service questions are further adaptively learned according to the dynamic changes of knowledge base customer service questions, so as to deal with unresponsive customer service questions. When customer service questions are clustered, more valuable and high-quality questions can be obtained to update customer service questions, which improves the efficiency of answering questions of intelligent customer service robots.
进一步地，在本申请客服问题的更新方法第一实施例的基础上，提出本申请客服问题的更新方法第四实施例，在本实施例中，上述步骤S400，根据所述聚类结果生成新的客服问题以对所述客服问题集进行更新，可以包括：Further, on the basis of the first embodiment of the method for updating customer service questions of this application, a fourth embodiment of the method for updating customer service questions of this application is proposed. In this embodiment, the above step S400 generates a new method according to the clustering result. To update the set of customer service questions, the customer service questions can include:
步骤S401，根据所述聚类结果和所述客服问题集生成新的标准问题和所述新的标准问题对应的相似问题；Step S401, generating a new standard question and similar questions corresponding to the new standard question according to the clustering result and the customer service question set;
智能客服机器人在针对未响应客服问题进行聚类得到聚类结果之后，结合该聚类结果，和FAQ问答模块的知识库中贮存的客服问题集中现有的标准问题和相似问题，整理生成新的标准问题和该新的标准问题的相似问题。After the intelligent customer service robot clusters the unresponsive customer service questions and obtains the clustering results, it combines the clustering results with the customer service questions stored in the knowledge base of the FAQ module to collect existing standard questions and similar questions, and organize them to generate new ones. The standard problem is similar to the new standard problem.
步骤S402，为所述新的标准问题和所述新的标准问题对应的相似问题配置各自对应的服务答案；Step S402, configuring respective corresponding service answers for the new standard question and similar questions corresponding to the new standard question;
步骤S403，将所述新的标准问题、所述新的标准问题对应的相似问题以及所述服务答案，添加至所述客服问题集以对所述客服问题集进行更新。Step S403: Add the new standard question, similar questions corresponding to the new standard question, and the service answer to the customer service question set to update the customer service question set.
智能客服机器人在整理生成新的标准问题以及该新的标准问题对应的相似问题之后，智能客服机器人分别针对该新的标准问题和该相似问题配置不同场景下各自对应的服务答案，然后，智能客服机器人建立该新的标准问题与其对应的服务答案之间、该新的标准问题的相似问题与其对应的服务答案之间的关联关系，然后将新的标准问题和其对应的服务答案、新的标准问题的相似问题与其对应的服务答案，添加至FAQ问答模块的知识库中贮存的客服问题集中，以实现对该客服问题集的更新。After the intelligent customer service robot organizes and generates the new standard question and the similar questions corresponding to the new standard question, the intelligent customer service robot configures the corresponding service answers in different scenarios for the new standard question and the similar question, and then the intelligent customer service The robot establishes the association relationship between the new standard question and its corresponding service answer, the similar question of the new standard question and its corresponding service answer, and then combines the new standard question with its corresponding service answer and the new standard Similar questions and their corresponding service answers are added to the customer service question set stored in the knowledge base of the FAQ module to update the customer service question set.
进一步地，在一种实施例中，智能客服机器人在针对未响应客服问题进行聚类得到聚类结果之后，还可以将该聚类结果直接面向运营人员进行输出，以辅助运营人员根据该聚类结果和FAQ问答模块的知识库中贮存的客服问题集中现有的标准问题和相似问题，整理生成新的标准问题和该新的标准问题的相似问题，并由该运营人员执行后续为新的标准问题和该新的标准问题的相似问题配置各自对应的服务答案的过程，然后接收运营人员所输入的新的标准问题和其对应的服务答案、新的标准问题的相似问题与其对应的服务答案，并将该新的标准问题和其对应的服务答案、新的标准问题的相似问题与其对应的服务答案，添加至FAQ问答模块的知识库中贮存的客服问题集中，以实现对该客服问题集的更新。Further, in an embodiment, after the intelligent customer service robot clusters the unresponsive customer service questions to obtain the clustering result, it can also directly output the clustering result to the operator, so as to assist the operator according to the clustering. Results and customer service questions stored in the knowledge base of the FAQ question and answer module focus on the existing standard questions and similar questions, sort out the new standard questions and similar questions of the new standard questions, and then implement the follow-up to the new standard by the operator The process of configuring the corresponding service answer for the question and the similar question of the new standard question, and then receiving the new standard question and its corresponding service answer input by the operator, and the similar question of the new standard question and its corresponding service answer, And add the new standard question and its corresponding service answer, the similar question of the new standard question and its corresponding service answer to the customer service question set stored in the knowledge base of the FAQ module, so as to realize the analysis of the customer service question set. Update.
在本实施例中，根据利用训练收敛的相似度模型所计算得到的未响应客服问题的聚类参数，针对未响应客服问题进行聚类运算得到聚类结果，然后基于该聚类结果和现有的客服问题集，整理生成新的标准问题和该新的标准问题对应的相似问题；然后为该新的标准问题和该新的标准问题对应的相似问题配置各自对应的服务答案；最后将新的标准问题、新的标准问题对应的相似问题以及各自对应的服务答案，添加至现有的客服问题集以对客服问题集进行更新。实现了利用机器学习模型动态的对未响应客服问题的聚类参数进行自适应学习，从而在针对未响应客服问题进行聚类时能够得到更加具有价值的高质量问题来更新客服问题，提高了智能客服机器人的问题回答效率。In this embodiment, according to the clustering parameters of the unresponsive customer service question calculated by using the training convergence similarity model, the clustering operation is performed on the unresponsive customer service question to obtain the clustering result, and then the clustering result is obtained based on the clustering result and the existing The customer service question set, organize and generate the new standard question and the similar question corresponding to the new standard question; then configure the corresponding service answer for the new standard question and the similar question corresponding to the new standard question; finally put the new standard question and the similar question corresponding to the new standard question. Standard questions, similar questions corresponding to the new standard questions, and their corresponding service answers are added to the existing customer service question set to update the customer service question set. Realize the use of machine learning model to dynamically adaptively learn the clustering parameters of unresponsive customer service questions, so that when clustering unresponsive customer service questions, more valuable and high-quality questions can be obtained to update customer service questions and improve intelligence The customer service robot’s question answering efficiency.
此外，本申请还提供了客服问题的更新系统，请参照图3，图3为本申请客服问题的更新系统的功能模块示意图，该客服问题的更新系统包括：In addition, this application also provides an update system for customer service issues. Please refer to Figure 3. Figure 3 is a schematic diagram of the functional modules of the application update system for customer service issues. The update system for customer service issues includes:
构建模块101，用于根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；The construction module 101 is configured to construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
学习模块102，用于基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；The learning module 102 is configured to train a similarity model based on the training data, and use the trained convergent similarity model to calculate clustering parameters that have not responded to customer service questions;
聚类模块103，用于基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；The clustering module 103 is configured to perform a clustering operation based on the clustering parameters to obtain the clustering result of the unresponsive customer service question;
更新模块104，用于根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。The update module 104 is configured to generate a new customer service question according to the clustering result to update the preset customer service question set.
此外，本申请还提供了一种计算机存储介质，所述计算机存储介质可以是非易失性的，也可以是易失性的，该计算机存储介质存储有一个或者一个以上程序，该一个或者一个以上程序还可被一个或者一个以上的处理器执行以用于：In addition, this application also provides a computer storage medium. The computer storage medium may be non-volatile or volatile. The computer storage medium stores one or more programs. The program can also be executed by one or more processors for:
根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。A new customer service question is generated according to the clustering result to update the preset customer service question set.
此外，该一个或者一个以上程序还可被一个或者一个以上的处理器执行还用于：In addition, the one or more programs may also be executed by one or more processors and used to:
间隔预设时长采集所述未响应客服问题，并将采集到的所述未响应客服问题存储在区块链中。Collect the unresponsive customer service questions at intervals of a preset length of time, and store the collected unresponsive customer service questions in the blockchain.
本申请计算机存储介质具体实施方式与上述客服问题的更新方法各实施例基本相同，在此不再赘述。The specific implementation of the computer storage medium of this application is basically the same as the above embodiments of the method for updating customer service questions, and will not be repeated here.
需要说明的是，本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。此外，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。It should be noted that the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer. In addition, in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also Other elements that are not explicitly listed, or also include elements inherent to the process, method, article, or device. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, article, or device that includes the element.
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority or inferiority of the embodiments.
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质（如ROM/RAM、磁碟、光盘）中，包括若干指令用以使得一台终端设备（可以是手机、计算机、服务器、空调器或者网络设备等）执行本申请各个实施例所述的方法。Through the description of the above implementation manners, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the method described in each embodiment of the present application.
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims (20)

  1. 一种客服问题的更新方法，其中，所述客服问题的更新方法包括：A method for updating customer service questions, wherein the method for updating customer service questions includes:
    根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
    基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
    基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
    根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。A new customer service question is generated according to the clustering result to update the preset customer service question set.
  2. 如权利要求1所述的客服问题的更新方法，其中，所述客服问题包括标准问题和相似问题，所述根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据的步骤，包括：The method for updating customer service questions according to claim 1, wherein the customer service questions include standard questions and similar questions, and the question pairs are constructed according to the customer service questions in the preset customer service question set, and the question pairs are combined into training The data steps include:
    从预设客服问题集中提取各标准问题和各所述标准问题各自对应的各相似问题；Extract each standard question and each similar question corresponding to each said standard question from the set of preset customer service questions;
    根据各所述标准问题中的第一标准问题，和所述第一标准问题对应的各相似问题构建得到各相似问题对；According to the first standard problem among the standard problems, each similar problem corresponding to the first standard problem is constructed to obtain each similar problem pair;
    根据所述第一标准问题，和各所述标准问题中的第二标准问题对应的各相似问题构建得到各非相似问题对，其中，所述第一标准问题与所述第二标准问题不相同；According to the first standard problem, each similar problem corresponding to the second standard problem in each standard problem is constructed to obtain each non-similar problem pair, wherein the first standard problem is different from the second standard problem ；
    将各所述相似问题对与各所述非相似问题对进行对应组合形成各训练数据。Each pair of similar questions is correspondingly combined with each pair of non-similar questions to form each training data.
  3. 如权利要求1所述的客服问题的更新方法，其中，所述基于所述训练数据训练相似度模型的步骤，包括：The method for updating customer service questions according to claim 1, wherein the step of training a similarity model based on the training data comprises:
    将所述训练数据切分为第一数据集和第二数据集；Dividing the training data into a first data set and a second data set;
    利用所述第一数据集对所述相似度模型进行训练，并利用所述第二数据集对经过训练的相似度模型进行验证；Using the first data set to train the similarity model, and using the second data set to verify the trained similarity model;
    将通过验证的相似度模型标记为训练收敛的相似度模型，并将训练收敛的相似度模型存储在区块链中。Mark the verified similarity model as a training converged similarity model, and store the trained convergent similarity model in the blockchain.
  4. 如权利要求1所述的客服问题的更新方法，其中，所述聚类参数包括：问题句子的向量表示和相似度值，所述利用训练收敛的相似度模型计算未响应客服问题的聚类参数的步骤，包括：The method for updating customer service questions according to claim 1, wherein the clustering parameters include: vector representations and similarity values of the questioned sentences, and the training and convergence similarity model is used to calculate the clustering parameters for the customer service questions that are not responded to The steps include:
    获取所述未响应客服问题的问题句子和所述问题句子的复制句子，并将所述问题句子和所述复制句子输入训练收敛的相似度模型；Acquiring the question sentence that did not respond to the customer service question and the duplicate sentence of the question sentence, and inputting the question sentence and the duplicate sentence into a training convergent similarity model;
    获取训练收敛的相似度模型计算所述问题句子和所述复制句子之间的相似度值的向量平均值；Acquiring a similarity model that has been trained to converge and calculating a vector average of similarity values between the question sentence and the copied sentence;
    将所述向量平均值作为所述问题句子的向量表示；Taking the vector average value as the vector representation of the question sentence;
    或者，遍历所述问题句子构建句子对，并将所述句子对输入训练收敛的相似度模型；Or, traverse the question sentence to construct a sentence pair, and input the sentence pair to train a convergent similarity model;
    获取训练收敛的相似度模型对所述句子对进行相似度值计算后输出的相似度值，并将所述相似度值作为所述问题句子的相似度值。Obtain the similarity value output after the similarity value calculation of the sentence pair is performed by the trained convergent similarity model, and the similarity value is used as the similarity value of the question sentence.
  5. 如权利要求4所述的客服问题的更新方法，其中，所述基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果的步骤，包括：The method for updating customer service questions according to claim 4, wherein the step of performing clustering operations based on the clustering parameters to obtain the clustering results of the unresponsive customer service questions comprises:
    调用预设基于距离的聚类算法，利用所述相似度值对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果；Calling a preset distance-based clustering algorithm, clustering the unresponsive customer service question by using the similarity value to obtain a clustering result of the unresponsive customer service question;
    或者，调用预设基于句子表示的聚类算法，利用所述向量表示对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果。Alternatively, a preset clustering algorithm based on sentence representation is invoked, and the vector representation is used to cluster the unresponsive customer service questions to obtain a clustering result of the unresponsive customer service questions.
  6. 如权利要求1所述的客服问题的更新方法，其中，所述根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新的步骤，包括：The method for updating customer service questions according to claim 1, wherein the step of generating new customer service questions according to the clustering result to update the preset customer service question set comprises:
    根据所述聚类结果和所述客服问题集生成新的标准问题和所述新的标准问题对应的相似问题；Generating a new standard question and similar questions corresponding to the new standard question according to the clustering result and the customer service question set;
    为所述新的标准问题和所述新的标准问题对应的相似问题配置各自对应的服务答案；Configure respective corresponding service answers for the new standard question and similar questions corresponding to the new standard question;
    将所述新的标准问题、所述新的标准问题对应的相似问题以及所述服务答案，添加至所述预设客服问题集以对所述预设客服问题集进行更新。The new standard question, similar questions corresponding to the new standard question, and the service answer are added to the preset customer service question set to update the preset customer service question set.
  7. 如权利要求1所述的客服问题的更新方法，其中，所述客服问题的更新方法，还包括：The method for updating customer service questions according to claim 1, wherein the method for updating customer service questions further comprises:
    间隔预设时长采集所述未响应客服问题，并将采集到的所述未响应客服问题存储在区块链中。Collect the unresponsive customer service questions at intervals of a preset length of time, and store the collected unresponsive customer service questions in the blockchain.
  8. 一种客服问题的更新系统，其中，所述客服问题的更新系统包括：A system for updating customer service questions, wherein the system for updating customer service questions includes:
    构建模块，用于根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；The construction module is used to construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
    学习模块，用于基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；The learning module is configured to train a similarity model based on the training data, and use the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
    聚类模块，用于基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；A clustering module, configured to perform a clustering operation based on the clustering parameters to obtain a clustering result of the unresponsive customer service question;
    更新模块，用于根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。The update module is used to generate new customer service questions according to the clustering result to update the preset customer service question set.
  9. 一种终端设备，其中，所述终端设备包括：存储器、处理器，通信总线以及存储在所述存储器上的客服问题的更新程序，A terminal device, wherein the terminal device includes a memory, a processor, a communication bus, and an update program for customer service questions stored on the memory,
    所述通信总线用于实现处理器与存储器间的通信连接；The communication bus is used to realize the communication connection between the processor and the memory;
    所述处理器用于执行所述基于互联网的客服问题的更新程序，以实现如下步骤：The processor is configured to execute the update program of the Internet-based customer service question, so as to implement the following steps:
    根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
    基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
    基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
    根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。A new customer service question is generated according to the clustering result to update the preset customer service question set.
  10. 如权利要求9所述的终端设备，其中，所述客服问题包括标准问题和相似问题，所述根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据的步骤，包括：The terminal device according to claim 9, wherein the customer service questions include standard questions and similar questions, and the step of constructing question pairs according to customer service questions in a preset customer service question set, and combining the question pairs into training data ,include:
    从预设客服问题集中提取各标准问题和各所述标准问题各自对应的各相似问题；Extract each standard question and each similar question corresponding to each said standard question from the set of preset customer service questions;
    根据各所述标准问题中的第一标准问题，和所述第一标准问题对应的各相似问题构建得到各相似问题对；According to the first standard problem among the standard problems, each similar problem corresponding to the first standard problem is constructed to obtain each similar problem pair;
    根据所述第一标准问题，和各所述标准问题中的第二标准问题对应的各相似问题构建得到各非相似问题对，其中，所述第一标准问题与所述第二标准问题不相同；According to the first standard problem, each similar problem corresponding to the second standard problem in each standard problem is constructed to obtain each non-similar problem pair, wherein the first standard problem is different from the second standard problem ；
    将各所述相似问题对与各所述非相似问题对进行对应组合形成各训练数据。Each pair of similar questions is correspondingly combined with each pair of non-similar questions to form each training data.
  11. 如权利要求9所述的终端设备，其中，所述基于所述训练数据训练相似度模型的步骤，包括：The terminal device according to claim 9, wherein the step of training a similarity model based on the training data comprises:
    将所述训练数据切分为第一数据集和第二数据集；Dividing the training data into a first data set and a second data set;
    利用所述第一数据集对所述相似度模型进行训练，并利用所述第二数据集对经过训练的相似度模型进行验证；Using the first data set to train the similarity model, and using the second data set to verify the trained similarity model;
    将通过验证的相似度模型标记为训练收敛的相似度模型，并将训练收敛的相似度模型存储在区块链中。Mark the verified similarity model as a training converged similarity model, and store the trained convergent similarity model in the blockchain.
  12. 如权利要求9所述的终端设备，其中，所述聚类参数包括：问题句子的向量表示和相似度值，所述利用训练收敛的相似度模型计算未响应客服问题的聚类参数的步骤，包括：9. The terminal device according to claim 9, wherein the clustering parameters comprise: vector representations and similarity values of the questioned sentences, and the step of calculating the clustering parameters for non-response customer service questions by using the trained and converged similarity model, include:
    获取所述未响应客服问题的问题句子和所述问题句子的复制句子，并将所述问题句子和所述复制句子输入训练收敛的相似度模型；Acquiring the question sentence that did not respond to the customer service question and the duplicate sentence of the question sentence, and inputting the question sentence and the duplicate sentence into a training convergent similarity model;
    获取训练收敛的相似度模型计算所述问题句子和所述复制句子之间的相似度值的向量平均值；Acquiring a similarity model that has been trained to converge and calculating a vector average of similarity values between the question sentence and the copied sentence;
    将所述向量平均值作为所述问题句子的向量表示；Taking the vector average value as the vector representation of the question sentence;
    或者，遍历所述问题句子构建句子对，并将所述句子对输入训练收敛的相似度模型；Or, traverse the question sentence to construct a sentence pair, and input the sentence pair to train a convergent similarity model;
    获取训练收敛的相似度模型对所述句子对进行相似度值计算后输出的相似度值，并将所述相似度值作为所述问题句子的相似度值。Obtain the similarity value output after the similarity value calculation of the sentence pair is performed by the trained convergent similarity model, and the similarity value is used as the similarity value of the question sentence.
  13. 如权利要求12所述的终端设备，其中，所述基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果的步骤，包括：The terminal device according to claim 12, wherein the step of performing a clustering operation based on the clustering parameter to obtain the clustering result of the unresponsive customer service question comprises:
    调用预设基于距离的聚类算法，利用所述相似度值对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果；Calling a preset distance-based clustering algorithm, clustering the unresponsive customer service question by using the similarity value to obtain a clustering result of the unresponsive customer service question;
    或者，调用预设基于句子表示的聚类算法，利用所述向量表示对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果。Alternatively, a preset clustering algorithm based on sentence representation is invoked, and the vector representation is used to cluster the unresponsive customer service questions to obtain a clustering result of the unresponsive customer service questions.
  14. 如权利要求9所述的终端设备，其中，所述根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新的步骤，包括：9. The terminal device of claim 9, wherein the step of generating a new customer service question according to the clustering result to update the preset customer service question set comprises:
    根据所述聚类结果和所述客服问题集生成新的标准问题和所述新的标准问题对应的相似问题；Generating a new standard question and similar questions corresponding to the new standard question according to the clustering result and the customer service question set;
    为所述新的标准问题和所述新的标准问题对应的相似问题配置各自对应的服务答案；Configure respective corresponding service answers for the new standard question and similar questions corresponding to the new standard question;
    将所述新的标准问题、所述新的标准问题对应的相似问题以及所述服务答案，添加至所述预设客服问题集以对所述预设客服问题集进行更新。The new standard question, similar questions corresponding to the new standard question, and the service answer are added to the preset customer service question set to update the preset customer service question set.
  15. 如权利要求9所述的终端设备，其中，所述处理器用于执行所述基于互联网的客服问题的更新程序，还实现如下步骤：The terminal device according to claim 9, wherein the processor is configured to execute the update program of the Internet-based customer service question, and further implements the following steps:
    间隔预设时长采集所述未响应客服问题，并将采集到的所述未响应客服问题存储在区块链中。Collect the unresponsive customer service questions at intervals of a preset length of time, and store the collected unresponsive customer service questions in the blockchain.
  16. 一种计算机存储介质，其中，所述计算机存储介质上存储有客服问题的更新程序，所述客服问题的更新程序被处理器执行时实现如下步骤：A computer storage medium, wherein an update program for customer service issues is stored on the computer storage medium, and the following steps are implemented when the update program for customer service issues is executed by a processor:
    根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据；Construct question pairs according to the customer service questions in the preset customer service question set, and combine the question pairs into training data;
    基于所述训练数据训练相似度模型，并利用训练收敛的相似度模型计算未响应客服问题的聚类参数；Training a similarity model based on the training data, and using the trained convergent similarity model to calculate clustering parameters that do not respond to customer service questions;
    基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果；Performing a clustering operation based on the clustering parameter to obtain a clustering result of the unresponsive customer service question;
    根据所述聚类结果生成新的客服问题以对所述预设客服问题集进行更新。A new customer service question is generated according to the clustering result to update the preset customer service question set.
  17. 如权利要求16所述的计算机存储介质，其中，所述客服问题包括标准问题和相似问题，所述根据预设客服问题集中的客服问题构建问题对，并将所述问题对组合为训练数据的步骤，包括：The computer storage medium of claim 16, wherein the customer service questions include standard questions and similar questions, and the question pairs are constructed according to the customer service questions in the preset customer service question set, and the question pairs are combined into training data. The steps include:
    从预设客服问题集中提取各标准问题和各所述标准问题各自对应的各相似问题；Extract each standard question and each similar question corresponding to each said standard question from the set of preset customer service questions;
    根据各所述标准问题中的第一标准问题，和所述第一标准问题对应的各相似问题构建得到各相似问题对；According to the first standard problem among the standard problems, each similar problem corresponding to the first standard problem is constructed to obtain each similar problem pair;
    根据所述第一标准问题，和各所述标准问题中的第二标准问题对应的各相似问题构建得到各非相似问题对，其中，所述第一标准问题与所述第二标准问题不相同；According to the first standard problem, each similar problem corresponding to the second standard problem in each standard problem is constructed to obtain each non-similar problem pair, wherein the first standard problem is different from the second standard problem ；
    将各所述相似问题对与各所述非相似问题对进行对应组合形成各训练数据。Each pair of similar questions is correspondingly combined with each pair of non-similar questions to form each training data.
  18. 如权利要求16所述的计算机存储介质，其中，所述基于所述训练数据训练相似度模型的步骤，包括：15. The computer storage medium of claim 16, wherein the step of training a similarity model based on the training data comprises:
    将所述训练数据切分为第一数据集和第二数据集；Dividing the training data into a first data set and a second data set;
    利用所述第一数据集对所述相似度模型进行训练，并利用所述第二数据集对经过训练的相似度模型进行验证；Using the first data set to train the similarity model, and using the second data set to verify the trained similarity model;
    将通过验证的相似度模型标记为训练收敛的相似度模型，并将训练收敛的相似度模型存储在区块链中。Mark the verified similarity model as a training converged similarity model, and store the trained convergent similarity model in the blockchain.
  19. 如权利要求16所述的计算机存储介质，其中，所述聚类参数包括：问题句子的向量表示和相似度值，所述利用训练收敛的相似度模型计算未响应客服问题的聚类参数的步骤，包括：The computer storage medium according to claim 16, wherein the clustering parameters include: vector representations and similarity values of the questioned sentences, and the step of using the trained and converged similarity model to calculate the clustering parameters for non-response customer service questions ,include:
    获取所述未响应客服问题的问题句子和所述问题句子的复制句子，并将所述问题句子和所述复制句子输入训练收敛的相似度模型；Acquiring the question sentence that did not respond to the customer service question and the duplicate sentence of the question sentence, and inputting the question sentence and the duplicate sentence into a training convergent similarity model;
    获取训练收敛的相似度模型计算所述问题句子和所述复制句子之间的相似度值的向量平均值；Acquiring a similarity model that has been trained to converge and calculating a vector average of similarity values between the question sentence and the copied sentence;
    将所述向量平均值作为所述问题句子的向量表示；Taking the vector average value as the vector representation of the question sentence;
    或者，遍历所述问题句子构建句子对，并将所述句子对输入训练收敛的相似度模型；Or, traverse the question sentence to construct a sentence pair, and input the sentence pair to train a convergent similarity model;
    获取训练收敛的相似度模型对所述句子对进行相似度值计算后输出的相似度值，并将所述相似度值作为所述问题句子的相似度值。Obtain the similarity value output after the similarity value calculation of the sentence pair is performed by the trained convergent similarity model, and the similarity value is used as the similarity value of the question sentence.
  20. 如权利要求19所述的计算机存储介质，其中，所述基于所述聚类参数进行聚类运算得到所述未响应客服问题的聚类结果的步骤，包括：19. The computer storage medium of claim 19, wherein the step of performing a clustering operation based on the clustering parameters to obtain a clustering result of the unresponsive customer service question comprises:
    调用预设基于距离的聚类算法，利用所述相似度值对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果；Calling a preset distance-based clustering algorithm, clustering the unresponsive customer service question by using the similarity value to obtain a clustering result of the unresponsive customer service question;
    或者，调用预设基于句子表示的聚类算法，利用所述向量表示对所述未响应客服问题进行聚类得到所述未响应客服问题的聚类结果。Alternatively, a preset clustering algorithm based on sentence representation is invoked, and the vector representation is used to cluster the unresponsive customer service questions to obtain a clustering result of the unresponsive customer service questions.
PCT/CN2020/118487 2020-06-24 2020-09-28 Customer service question update method and system, terminal device, and computer storage medium WO2021114834A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010595834.7A CN111813910A (en) 2020-06-24 2020-06-24 Method, system, terminal device and computer storage medium for updating customer service problem
CN202010595834.7 2020-06-24

Publications (1)

Publication Number Publication Date
WO2021114834A1 true WO2021114834A1 (en) 2021-06-17

Family

ID=72855522

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/118487 WO2021114834A1 (en) 2020-06-24 2020-09-28 Customer service question update method and system, terminal device, and computer storage medium

Country Status (2)

Country Link
CN (1) CN111813910A (en)
WO (1) WO2021114834A1 (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112328748A (en) * 2020-11-11 2021-02-05 上海昌投网络科技有限公司 Method for identifying insurance configuration intention
CN112541360A (en) * 2020-12-07 2021-03-23 国泰君安证券股份有限公司 Cross-platform anomaly identification and translation method, device, processor and storage medium for clustering by using hyper-parametric self-adaptive DBSCAN (direct media Access controller area network)
CN112988999A (en) * 2021-03-17 2021-06-18 平安科技（深圳）有限公司 Construction method, device, equipment and storage medium of Buddha question and answer pair
CN116910259B (en) * 2023-09-12 2024-04-16 深圳须弥云图空间科技有限公司 Knowledge diagnosis method and device for knowledge base

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9020862B2 (en) * 2011-07-29 2015-04-28 International Business Machines Corporation Method and system for computer question-answering
CN105975460A (en) * 2016-05-30 2016-09-28 上海智臻智能网络科技股份有限公司 Question information processing method and device
CN107562789A (en) * 2017-07-28 2018-01-09 深圳前海微众银行股份有限公司 Knowledge base problem update method, customer service robot and readable storage medium storing program for executing
CN108804567A (en) * 2018-05-22 2018-11-13 平安科技（深圳）有限公司 Improve method, equipment, storage medium and the device of intelligent customer service response rate
CN110555101A (en) * 2019-09-09 2019-12-10 浙江诺诺网络科技有限公司 customer service knowledge base updating method, device, equipment and storage medium
CN110909165A (en) * 2019-11-25 2020-03-24 杭州网易再顾科技有限公司 Data processing method, device, medium and electronic equipment
CN110955766A (en) * 2019-11-29 2020-04-03 支付宝(杭州)信息技术有限公司 Method and system for automatically expanding intelligent customer service standard problem pairs

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8204838B2 (en) * 2009-04-10 2012-06-19 Microsoft Corporation Scalable clustering
CN106355449B (en) * 2016-08-31 2021-09-07 腾讯科技（深圳）有限公司 User selection method and device
CN107957989B9 (en) * 2017-10-23 2021-01-12 创新先进技术有限公司 Cluster-based word vector processing method, device and equipment
CN110032632A (en) * 2019-04-04 2019-07-19 平安科技（深圳）有限公司 Intelligent customer service answering method, device and storage medium based on text similarity
CN110263141A (en) * 2019-06-25 2019-09-20 杭州微洱网络科技有限公司 A kind of customer service question answering system based on BERT
CN111125334B (en) * 2019-12-20 2023-09-12 神思电子技术股份有限公司 Search question-answering system based on pre-training

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9020862B2 (en) * 2011-07-29 2015-04-28 International Business Machines Corporation Method and system for computer question-answering
CN105975460A (en) * 2016-05-30 2016-09-28 上海智臻智能网络科技股份有限公司 Question information processing method and device
CN107562789A (en) * 2017-07-28 2018-01-09 深圳前海微众银行股份有限公司 Knowledge base problem update method, customer service robot and readable storage medium storing program for executing
CN108804567A (en) * 2018-05-22 2018-11-13 平安科技（深圳）有限公司 Improve method, equipment, storage medium and the device of intelligent customer service response rate
CN110555101A (en) * 2019-09-09 2019-12-10 浙江诺诺网络科技有限公司 customer service knowledge base updating method, device, equipment and storage medium
CN110909165A (en) * 2019-11-25 2020-03-24 杭州网易再顾科技有限公司 Data processing method, device, medium and electronic equipment
CN110955766A (en) * 2019-11-29 2020-04-03 支付宝(杭州)信息技术有限公司 Method and system for automatically expanding intelligent customer service standard problem pairs

Also Published As

Publication number Publication date
CN111813910A (en) 2020-10-23

Similar Documents

Publication Publication Date Title
WO2021114834A1 (en) Customer service question update method and system, terminal device, and computer storage medium
US20210201147A1 (en) Model training method, machine translation method, computer device, and storage medium
US11790904B2 (en) Voice application platform
US11887597B2 (en) Voice application platform
JP7104683B2 (en) How and equipment to generate information
CN110795528B (en) Data query method and device, electronic equipment and storage medium
US20190371313A1 (en) Voice application platform
TW201935273A (en) A statement user intention identification method and device
CN109947919A (en) Method and apparatus for generating text matches model
WO2021175007A1 (en) Online customer service consultation method and apparatus, medium, and electronic device
CN107871501A (en) The automated software identified using intelligent sound performs method
WO2018067368A1 (en) Hierarchical annotation of dialog acts
US10979242B2 (en) Intelligent personal assistant controller where a voice command specifies a target appliance based on a confidence score without requiring uttering of a wake-word
US11437029B2 (en) Voice application platform
WO2019000905A1 (en) Triage dialogue method, device, and system
WO2022022024A1 (en) Training sample construction method, apparatus, and device, and computer-readable storage medium
CN111105852A (en) Electronic medical record recommendation method and device, terminal and storage medium
WO2022105297A1 (en) Table structure recovery method and system, computer device, and storage medium
CN108694941A (en) For the method for interactive session, information processing unit and product
CN113868427A (en) Data processing method and device and electronic equipment
CN110619050A (en) Intention recognition method and equipment
CN114564595A (en) Knowledge graph updating method and device and electronic equipment
WO2021068493A1 (en) Method and apparatus for processing information
CN114694226A (en) Face recognition method, system and storage medium
CN111651989B (en) Named entity recognition method and device, storage medium and electronic device

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20898044

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20898044

Country of ref document: EP

Kind code of ref document: A1