WO2021135441A1 - 基于深度学习的客户标签确定方法、装置、设备及介质 - Google Patents

基于深度学习的客户标签确定方法、装置、设备及介质 Download PDF

Info

Publication number
WO2021135441A1
WO2021135441A1 PCT/CN2020/117495 CN2020117495W WO2021135441A1 WO 2021135441 A1 WO2021135441 A1 WO 2021135441A1 CN 2020117495 W CN2020117495 W CN 2020117495W WO 2021135441 A1 WO2021135441 A1 WO 2021135441A1
Authority
WO
WIPO (PCT)
Prior art keywords
customer
intention
data set
classifier
result
Prior art date
Application number
PCT/CN2020/117495
Other languages
English (en)
French (fr)
Inventor
侯翠琴
李剑锋
文彬
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
Priority to US17/620,736 priority Critical patent/US20220414687A1/en
Publication of WO2021135441A1 publication Critical patent/WO2021135441A1/zh

Links

Images

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
    • 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
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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 intelligent decision-making, and in particular to a method, device, equipment, and medium for determining customer tags based on deep learning.
  • human-machine dialogue technology has attracted more and more attention and research from people from all walks of life, and human-machine dialogue products have also sprung up.
  • the human-machine dialogue system can provide customers with consulting, sales and other related services 24 hours a year, which can save a lot of manpower and cost. Therefore, intelligent customer service robots that serve customers are the most valuable , One of the man-machine dialogue products with the most usage scenarios.
  • This type of method is inaccurate in identifying customers' intentions, can only answer a few questions and the sentences are blunt, the customer feels poor, leading to the premature interruption of customer communication services, and the background cannot accurately determine the customer's product purchase intention based on a small amount of customer dialogue data , The accuracy of customer screening is not high, and high-value customers cannot be accurately selected.
  • This application provides a method, device, equipment, and medium for determining customer tags based on deep learning to solve the problem of inaccurate identification of customer intentions in the prior art, which leads to low accuracy of customer screening.
  • a method for determining customer labels based on deep learning including:
  • the content of the conversation is input into a preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier is based on customer conversations with multiple customer tags
  • the data is the intention classifier obtained by product purchase intention classification and dialogue sentence intention classification training.
  • the customer tags include high-intention customers who have purchase intentions for the product, low-intention customers who reject the product, and medium-sized customers who have not expressed their opinions on the product.
  • a customer label determination device based on deep learning including:
  • the first acquisition module is used to acquire the content of the dialogue between the customer and the robot customer service
  • the input module is configured to input the dialogue content into a preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier is based on a variety of
  • the customer dialogue data of the customer tag is the intention classifier obtained by product purchase intention classification and dialog sentence intention classification training.
  • the customer tag includes high-intention customers who have purchase intentions for the product, low-intention customers who reject the product, and non-intentional customers. Neutral customers who describe the product's position;
  • a setting module configured to set the customer tag of the customer according to the result of the product purchase intention recognition, and determine whether to provide manual service for the customer according to the customer tag of the customer;
  • the second obtaining module is configured to obtain the result of the manual service and the conversation data of the customer in the manual service if the manual service is provided for the customer;
  • the update module is used to update the customer tag of the customer according to the result of the manual service, and update the preset multi-factor intention classifier according to the conversation data of the customer.
  • a computer device includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor, and the processor implements the following steps when the processor executes the computer-readable instructions:
  • the content of the conversation is input into a preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier is based on customer conversations with multiple customer tags
  • the data is the intention classifier obtained by product purchase intention classification and dialogue sentence intention classification training.
  • the customer tags include high-intention customers who have purchase intentions for the product, low-intention customers who reject the product, and medium-sized customers who have not expressed their opinions on the product.
  • a computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the following steps are implemented:
  • the content of the conversation is input into a preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier is based on customer conversations with multiple customer tags
  • the data is the intention classifier obtained by product purchase intention classification and dialogue sentence intention classification training.
  • the customer tags include high-intention customers who have purchase intentions for the product, low-intention customers who reject the product, and medium-sized customers who have not expressed their opinions on the product.
  • the correlation between the product purchase intention classification and the dialogue sentence intention classification is considered, which improves the accuracy of the preset multi-factor intention classifier in identifying the customer's product purchase intention , And then set the customer’s customer label according to the higher recognition result to provide worker services based on the customer’s label, update the customer label according to the result of the manual service, and improve the accuracy of the customer’s label so that subsequent accurate customer screening based on the customer’s label can be carried out. In turn, the quality of customer service is improved.
  • the preset multi-factor intention classifier is continuously updated based on the customer’s dialogue data during the manual service process, which further improves the identification of customer intentions by the preset multi-factor intention classifier.
  • the accuracy of customer selection thereby further improving the accuracy of customer screening.
  • FIG. 1 is a schematic diagram of an application environment of a method for determining customer tags based on deep learning in an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a method for determining customer tags based on deep learning in an embodiment of the present application
  • FIG. 3 is a schematic diagram of an implementation flow of step S30 in FIG. 2;
  • FIG. 4 is a schematic diagram of another implementation flow of step S30 in FIG. 2;
  • FIG. 5 is a schematic diagram of an implementation flow of step S50 in FIG. 2;
  • FIG. 6 is a schematic diagram of an acquisition process of a preset multi-factor intention classifier in an embodiment of the present application.
  • FIG. 7 is a schematic diagram of an implementation flow of step S06 in FIG. 6;
  • FIG. 8 is a schematic diagram of a structure of an apparatus for determining customer tags based on deep learning in an embodiment of the present application
  • Fig. 9 is a schematic structural diagram of a computer device in an embodiment of the present application.
  • the method for determining client labels based on deep learning can be applied in the application environment as shown in FIG. 1, in which the client communicates with the server through the network.
  • the server obtains the dialogue content between the client and the robot customer service in the client, and inputs the dialogue content into the preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and preset the multi-factor intention classification
  • the intention classifier is obtained by training the product purchase intention classification and the dialog sentence intention classification training based on the customer dialogue data of multiple customer tags.
  • the customer tags include high-intention customers who have purchase intentions, low-intention customers who reject products, and non-products.
  • the client can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices.
  • the server can be implemented as an independent server or a server cluster composed of multiple servers.
  • a method for determining customer labels based on deep learning is provided.
  • the method is applied to the server in FIG. 1 as an example for description, including the following steps:
  • the content of the dialog between the customer and the robot customer service is obtained, so that the customer's purchase intention of the product can be identified according to the dialog content input by the customer.
  • S20 Input the dialogue content into the preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier performs product purchase based on the customer dialogue data of multiple customer tags
  • customer tags include high-intention customers who have purchase intentions, low-intention customers who reject products, and neutral customers who have not expressed their opinions on products.
  • the preset multi-factor intention classifier is an intention classifier obtained by performing product purchase intention classification and dialogue sentence intention classification training based on the customer dialogue data of multiple customer tags.
  • the multiple customer tags include high-intention customers who have purchase intentions for the product, Low-intention customers who reject products and neutral customers who have not expressed their opinions on products further improve the diversity of the training data of the preset multi-factor intention classifier, and increase the accuracy of the preset multi-factor intention classifier.
  • the preset multi-factor intention classifier considers the correlation between the customer's product purchase intention and the customer's intention at different dialogue stages, and has a higher accuracy in identifying the customer's product purchase intention.
  • a preset multi-factor intention classifier is used to identify the customer's product purchase intention during the conversation, which realizes the automatic processing of artificial intelligence + intention recognition, and expresses and accurately obtains intention recognition without manual participation. As a result, the recognition efficiency and accuracy are improved.
  • the obtained relevant data information and the generated preset multi-factor intention classifier can be saved in the block Chain network.
  • Blockchain 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.
  • the preset multi-factor intention classifier and related data are stored in the blockchain network, which facilitates quick query of the target classifier and data, and improves the processing speed.
  • S30 Set the customer's customer label according to the product purchase intention recognition result, and determine whether to provide the customer with manual service according to the customer's customer label.
  • the customer's customer label is set according to the product purchase intention recognition result, and whether to provide the customer with manual service is determined according to the set customer label.
  • the robot customer service will continue to control the dialogue with the customer until the dialogue ends.
  • the customer will be transferred to the manual customer so that the experienced customer service staff can provide personalized manual service for the customer.
  • record The conversation data of the customer in the manual service and the result of the manual service, so that the result of the manual service and the conversation data of the customer in the manual service can be inquired and extracted subsequently.
  • the manual service may be a product promotion service. In other embodiments, the manual service may also be other types of services. In this embodiment, a manual service is used as a product promotion service as an example for description.
  • S50 Update the customer label of the customer according to the result of the manual service, and update the preset multi-factor intention classifier according to the conversation data of the customer.
  • the preset multi-factor intention classifier will be updated according to the dialogue data of the customer in the manual service, that is, the preset multi-factor intention classifier will be retrained according to the dialogue data in the manual service process to improve the preset multi-factor intention classifier
  • the accuracy of the recognition of the customer's product purchase intention makes the accuracy of the result predicted by the preset multi-factor intention classifier become higher and higher with the continuous operation of the server.
  • the content of the dialogue between the customer and the robot customer service is obtained, and the content of the dialogue is input into the preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and then according to the product purchase intention Set the customer’s customer tag for the identification result, and determine whether to provide manual service for the customer according to the customer’s customer tag.
  • the training process of the preset multi-factor intention classifier takes into account the correlation between the product purchase intention classification and the dialogue sentence intention classification, which improves
  • the preset multi-factor intent classifier recognizes the accuracy of the customer’s product purchase intention, and then sets the customer’s customer label according to the recognition result to provide worker services according to the customer’s label, and update the customer label according to the result of the manual service, which improves the quality of the customer’s label.
  • the preset multi-factor intention classifier is continuously updated based on the customer’s dialogue data during the manual service process. It further improves the accuracy of the preset multi-factor intention classifier to identify customer intentions, thereby further improving the accuracy of customer screening.
  • step S30 after obtaining the product purchase intention recognition result output by the preset multi-factor intention classifier, as shown in FIG. 3, in step S30, the customer tag of the customer is set according to the product purchase intention recognition result, which specifically includes the following step:
  • the preset multi-factor intention classifier After obtaining the product purchase intention recognition result output by the preset multi-factor intention classifier, it is determined whether the product purchase intention recognition result is a high purchase intention, and the customer label is set according to the determination result.
  • the customer tag of the customer is set as a low intention customer.
  • the customer’s customer tag is set as a high-intention customer, so as to provide personalized manual services to high-intention customers based on the customer tag in order to improve the customer’s service experience.
  • the preset multi-factor intention classifier After obtaining the product purchase intention recognition result output by the preset multi-factor intention classifier, it is determined whether the product purchase intention recognition result is high purchase intention, and the product purchase intention recognition result is determined to be high purchase intention, then in the dialogue process , Determine whether the preset multi-factor intention classifier outputs high purchase intentions more than the preset number of times, if the preset multi-factor intention classifier outputs high purchase intentions more than the preset number of times, set the customer’s customer tag to high Intentional customers, detailed the steps of setting customers’ customer labels based on the results of product purchase intention recognition, and clarified the process of identifying high-intent customers based on the number of times the preset multi-factor intention classifier outputs high purchase intentions, which is the follow-up label for different customers Customers provide the basis for different service strategies.
  • the customer tag of the customer is set as a low intention customer; If it is determined that the product purchase intention recognition result is a neutral intention that does not express a statement on the product, and the number of times that the preset multi-factor intention classifier outputs neutral intentions is greater than the second preset threshold, the customer’s customer label is set as a neutral customer, In order to subsequently screen the customer groups according to the customer tags of the customers, and then provide different customer services.
  • step S30 that is, determining whether to provide manual service for the customer according to the customer's customer tag, specifically includes the following steps:
  • the customer's customer label is set according to the product purchase intention recognition result, it is determined whether the customer's customer label is a high-intention customer.
  • the customer's customer tag After determining whether the customer's customer tag is a high-intention customer, if the customer's customer tag is a high-intention customer, it is determined to provide the customer with manual service and transfer the customer to a manual customer service to improve the customer's service experience.
  • the robot customer service communicates with the customer until the end of the conversation.
  • the customer’s customer tag is determined as a high-intention customer to determine whether to provide manual services according to the determination result. If it is determined that the customer’s customer tag is high-intention Customers are determined to provide manual services for customers.
  • the steps for determining whether to provide manual services for customers based on the customer’s customer tags have been further refined, the customer service strategy of different customer tags has been optimized, and the customer service experience of high-intention customers has been improved.
  • step S50 that is, updating the customer label of the customer according to the result of the manual service, specifically includes the following steps:
  • S51 Determine whether the product transaction is reached according to the result of the manual service.
  • the manual customer service can re-mark the customer's product purchase intention. Therefore, after determining whether the product transaction is completed according to the result of the manual service, if it is determined that the product transaction is not completed, the result of marking the customer's product purchase intention by the manual customer service can be obtained, so that the customer's customer label can be updated according to the marking result.
  • S53 Update the customer label of the customer according to the result of marking the customer by the manual customer service.
  • the customer's customer label is updated according to the manual customer service's marking result of the customer.
  • the customer's customer label is updated from a high-intention customer to a low-intention customer.
  • the result of the manual service is used to determine whether the transaction of the product is reached. If the transaction of the product is not reached, the manual customer service is obtained. Label the results, and then update the customer’s customer label according to the result of the manual customer service’s labeling of the customer, and refine the steps of updating the customer’s customer label according to the result of the manual service. After several services are pushed to high-intention customers, according to the transaction completion status The customer label is updated with the labeling result of manual customer service, which further improves the accuracy of the customer label, and facilitates the subsequent screening of high-quality customers based on the customer label.
  • the product transaction after determining whether the product transaction is completed according to the result of the manual service, if the product transaction is completed, it is inconvenient to keep the customer’s customer tag as a high-intention customer, and remark the customer as a customer that has completed the transaction to proceed in one step. Refine the customer's tags to facilitate subsequent screening of customers based on customer tags and remarks.
  • the preset multi-factor intention classifier before inputting the dialogue content into the preset multi-factor intention classifier, it is also necessary to perform product purchase intention classification and dialogue sentence intention classification training based on the customer dialogue data of multiple customer tags to obtain the preset multi-factor intention classification Device.
  • the preset multi-factor intention classifier is obtained in the following way:
  • Customer tags include high-intent customers, low-intention customers, and neutral customers.
  • the customer tags include high-intention customers who have purchase intentions for the product, low-intention customers who reject the product, and neutral customers who have not expressed their opinions on the product. That is, the customer conversation data includes high-intention customers. Conversation data of intended customers, conversation data of low-intention customers, and conversation data of neutral customers.
  • the customer dialogue data with the customer tag before acquiring the customer dialogue data with the customer tag, it is necessary to determine the customer's overall purchase intention for the product according to the historical conversation data of each customer, and then set the customer's customer tag according to the customer's overall purchase intention for the product , In order to obtain sufficient, multi-customer label customer conversation data for preset multi-factor intention classifier training.
  • the customer’s conversational statements such as: how to check the insurance policy, consult the insurance policy, how to purchase, etc.
  • the dialogue data is positive intention data; if the customer’s question is related to the product, but the intention of the customer dialogue statement (such as: no time, no need, etc.) is used to determine whether the customer’s attitude is questioning or even complaining or disgusting, then it is determined that the customer is against the product If the customer’s purchase intention is negative, that is, if the user refuses to understand and purchase the product, the customer’s customer tag is set as a low-intention customer, and the customer’s conversation data is negative intention data; if the customer’s conversation data is determined by the customer’s conversation statement If the intention is not related to the industry product (such as greetings, closing remarks, etc.), it will be determined that the customer has not expressed a position on the product, and the customer’s product purchase intention cannot be determined, and the customer’s customer label will be set as a neutral customer, and the customer’s dialogue data For neutral data.
  • the intention of the customer dialogue statement such as: no time, no need, etc.
  • S02 Use customer conversation data labeled as a high-intention customer as a positive intention data set.
  • the customer conversation data with the customer tags as high intention customers are taken as the positive intention data set, that is, the conversation data of customers who are interested in products and have product purchase intentions are taken as the positive intention data set.
  • the customer conversation data with the customer tags as low-intention customers are regarded as the negative intention data set, that is, the conversation data of the customers who refuse to know and refuse to purchase products are regarded as the negative intention data set.
  • the customer conversation data with the customer tag as a neutral customer is taken as the neutral data set, that is, the conversation data of the customer who expresses their opinions on the product is taken as the neutral data set.
  • S05 Collect the dialogue data of the positive intent data set, the negative intent data set and the neutral data set into intent data, and identify the intent of each sentence in the intent data to obtain the intent data set.
  • the customer conversation data of the positive intent data set, negative intent data set, and neutral data set are aggregated into intent data, and each of the intent data is identified.
  • the intent of a sentence sentence is used to obtain an intent data set.
  • the intent data set includes the intent data and the intent corresponding to each dialogue sentence in the intent data.
  • S06 Perform classifier training according to the positive intent data set, the negative intent data set, the neutral data set, and the intent data set to obtain a preset multi-factor intent classifier.
  • the preset multi-factor intention classifier After obtaining the positive intent data set, negative intent data set, neutral data set and intent data set, according to the positive intent data set, negative intent data set, neutral data set and intent data set to classify the product purchase intention and the dialogue statement intention Classification training to obtain a preset multi-factor intention classifier.
  • the preset multi-factor intention classifier not only the joint learning effect of the intention classification task of product purchase and the intention classification task of dialogue sentence is considered, but also the association between the intention classification task and the intention classification task is considered. In this way, the preset multi-factor intention classifier obtained by this method has higher accuracy in identifying the customer's intention and can more accurately predict the customer's purchase intention of the product.
  • the customer tags include high-intention customers, low-intention customers, and neutral customers.
  • the customer conversation data whose customers are tagged as high-intention customers are taken as the positive intention data set, and the customer tags are
  • the customer dialogue data for low-intention customers is regarded as the negative intention data set
  • the customer conversation data of customers labeled as neutral customers are regarded as the neutral data set
  • the conversation data of the positive intention data set, the negative intention data set and the neutral data set are summarized It is the intention data, and the intention of each sentence in the intention data is identified to obtain the intention data set
  • the classifier training is carried out according to the positive intention data set, the negative intention data set, the neutral data set and the intention data set to obtain the preset
  • the multi-factor intention classifier clarifies the source and type of training data, clarifies the process of obtaining the preset multi-factor intention classifier, and provides a way to identify the customer’s product purchase intention based on the preset multi-factor intention classifier in the subsequent dialogue
  • step S06 classifier training is performed according to the positive intent data set, the negative intent data set, the neutral data set, and the intent data set to obtain the preset multi-factor intent classifier , Specifically including the following steps:
  • S061 Perform intention classification learning according to the positive intention data set, the negative intention data set and the neutral data set to obtain the intention classification learning result.
  • the intention classification learning is performed according to the positive intention data set, the negative intention data set and the neutral data set to obtain the intention classification learning result.
  • the intention classification learning result is the customer's product purchase intention recognition result.
  • S062 Perform intent classification learning according to the intent data set to obtain an intent classification learning result.
  • the intent classification learning is performed according to the intent data set to obtain the intent classification learning result, where the intent classification learning result is the phased dialogue in the dialogue sentence or the intent recognition result in a single dialogue sentence.
  • S063 Adjust the intention classification learning according to the intention classification learning result and the intention classification learning result to obtain a preset multi-factor intention classifier.
  • the intention classification learning is adjusted and trained according to the intention classification learning results and the intention classification learning results to obtain a preset multi-factor intention classifier. That is, in the training process of the preset multi-factor intention classifier, not only the joint learning effect of the intention classification learning and the intention classification learning, but also the correlation between the intention classification learning and the intention classification learning is considered. Considering the relevance between intention classification learning and intention classification learning is to consider the consistency between the prediction results of intention classification learning and the prediction results of intention classification learning, and then consider the prediction results of intention classification learning and the real product purchase intention results. To improve the accuracy of the pre-trained multi-factor intention classifier.
  • the intention classification learning is performed according to the positive intention data set, the negative intention data set, and the neutral data set to obtain the intention classification learning result
  • the intention classification learning is performed according to the intention data set to obtain the intention classification learning result.
  • Intent classification learning results and intention classification learning results are adjusted to the intention classification learning to obtain a preset multi-factor intention classifier, which is further refined according to the positive intention data set, the negative intention data set, the neutral data set and the intention data set.
  • the classifier training to obtain the preset multi-factor intention classifier fully considers the influence of the customer's dialogue sentence data on the customer's product purchase intention recognition, and learns the intention classification according to the intention classification learning results and the intention classification learning results
  • the adjustment takes into account the relevance of the intention classification and the intention classification, thereby improving the accuracy of the preset multi-factor intention classifier, and providing a basis for accurately identifying the customer's product purchase intention in the subsequent dialogue with the customer.
  • a device for determining a customer label based on deep learning includes a first acquisition module 801, an input module 802, a setting module 803, a second acquisition module 804 and an update module 805.
  • the device for determining customer tags based on deep learning includes a first acquisition module 801, an input module 802, a setting module 803, a second acquisition module 804 and an update module 805.
  • the detailed description of each functional module is as follows:
  • the first obtaining module 801 is used to obtain the content of the conversation between the customer and the robot customer service;
  • the input module 802 is configured to input the dialogue content into a preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier An intention classifier obtained by conducting product purchase intention classification and dialog sentence intention classification training on customer conversation data of a customer label.
  • the customer label includes high intention customers who have purchase intentions for the product, low intention customers who reject the product, and non-compliance. Neutral customers of the said product;
  • the setting module 803 is configured to set the customer tag of the customer according to the result of the product purchase intention recognition, and determine whether to provide manual service for the customer according to the customer tag of the customer;
  • the second obtaining module 804 is configured to obtain the result of the manual service and the conversation data of the customer in the manual service if the manual service is provided for the customer;
  • the update module 805 is configured to update the customer tag of the customer according to the result of the manual service, and update the preset multi-factor intention classifier according to the conversation data of the customer.
  • setting module 803 is specifically configured to:
  • the customer tag of the customer is set as a high intention customer.
  • setting module 803 is specifically used for:
  • the customer tag of the customer is a high intention customer, it is determined to provide the manual service for the customer.
  • update module 805 is specifically configured to:
  • the device for determining customer tags based on deep learning further includes a third obtaining module 806, and the third obtaining module 806 is specifically configured to:
  • customer tags include the high-intentional customers, the low-intentional customers, and the neutral customers;
  • the third obtaining module 806 is specifically further configured to:
  • the intention classification learning is adjusted according to the intention classification learning result and the intention classification learning result to obtain the preset multi-factor intention classifier.
  • the various modules in the device for determining customer tags based on deep learning can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 9.
  • the computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and/or a volatile storage medium, and internal memory.
  • the non-volatile storage medium stores an operating system, computer readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium.
  • the database of the computer equipment is used to preset the data applied and generated in the multi-factor intention classifier and the method for determining customer labels based on deep learning.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer-readable instructions are executed by the processor, a method for determining customer tags based on deep learning is realized.
  • a computer device including a memory, a processor, and computer-readable instructions stored on the memory and capable of running on the processor, and the processor implements the following steps when the processor executes the computer-readable instructions:
  • the content of the conversation is input into a preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier is based on customer conversations with multiple customer tags
  • the data is the intention classifier obtained by product purchase intention classification and dialogue sentence intention classification training.
  • the customer tags include high-intention customers who have purchase intentions for the product, low-intention customers who reject the product, and medium-sized customers who have not expressed their opinions on the product.
  • a computer-readable storage medium on which computer-readable instructions are stored, and when the computer-readable instructions are executed by a processor, the following steps are implemented:
  • the content of the conversation is input into a preset multi-factor intention classifier to obtain the product purchase intention recognition result output by the preset multi-factor intention classifier, and the preset multi-factor intention classifier is based on customer conversations with multiple customer tags
  • the data is the intention classifier obtained by product purchase intention classification and dialogue sentence intention classification training.
  • the customer tags include high-intention customers who have purchase intentions for the product, low-intention customers who reject the product, and medium-sized customers who have not expressed their opinions on the product.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Mathematical Physics (AREA)
  • Human Computer Interaction (AREA)
  • Computational Linguistics (AREA)
  • Databases & Information Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种基于深度学习的客户标签确定方法、装置、设备及介质，涉及人工智能以及区块链领域。该方法通过获取客户与机器人客服的对话内容（S10），将对话内容输入预设多因子意向分类器，以获取预设多因子意向分类器输出的产品购买意向识别结果（S20），根据产品购买意向识别结果设置客户的客户标签，并根据客户的客户标签确定是否为客户提供人工服务（S30），若为客户提供人工服务，则获取人工服务的结果和人工服务中客户的对话数据（S40），最后根据人工服务的结果更新客户的客户标签，并根据客户的对话数据更新预设多因子意向分类器（S50）。该方法提高了预设多因子意向分类器对客户的产品购买意向识别的准确性，提高了客户标签的准确性，以便后续根据客户标签进行准确地客户筛选。

Description

基于深度学习的客户标签确定方法、装置、设备及介质
本申请要求于2020年8月06日提交中国专利局、申请号为202010783681.9，发明名称“基于深度学习的客户标签确定方法、装置、设备及介质”的中国发明专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及智能决策技术领域，尤其涉及一种基于深度学习的客户标签确定方法、装置、设备及介质。
背景技术
随着人工智能技术的发展，特别是自然语音处理技术的快速发展，人机对话技术越来越受到各界人士的关注和研究，人机对话产品也如雨后春笋不断涌现。在客户服务技术领域，人机对话系统能全年24小时无间歇的为客户提供咨询、销售等相关服务，能大量节省人力和成本支出，因此，为客户服务的智能客服机器人是最具商业价值、使用场景最多的人机对话产品之一。
技术问题
但发明人发现，由于现有人工智能技术的限制，现有的智能客服机器人虽然能低成本全天候地为客户提供服务，但无法为客户提供高质量的个性化服务。尤其是在需要推销产品的售前服务中，需要根据与客户的对话交流过程中对客户进行筛选，以提高产品的销售率。但在售前服务中，智能客服机器人一般根据简单的意向识别方法进行客户意向识别，根据意向识别结果进行机械应答，如关键字识别触发回应预先编辑的对应资料。该类方法对客户的意向识别不准确，只能回答较为少的问题且语句生硬，客户使用感觉差，导致客户交流服务提前中断，后台无法根据少量的客户对话数据准确地判断客户的产品购买意向，导致客户筛选的准确性不高，无法准确地筛选出高价值的客户。
技术解决方案
本申请提供一种基于深度学习的客户标签确定方法、装置、设备及介质，以解决现有技术中，对客户的意向的识别不准确，导致客户筛选的准确性不高的问题。
一种基于深度学习的客户标签确定方法，包括：
获取客户与机器人客服的对话内容；
将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
一种基于深度学习的客户标签确定装置，包括：
第一获取模块，用于获取客户与机器人客服的对话内容；
输入模块，用于将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
设置模块，用于根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
第二获取模块，用于若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
更新模块，用于根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
获取客户与机器人客服的对话内容；
将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如下步骤：
获取客户与机器人客服的对话内容；
将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
有益效果
本申请中，在预设多因子意向分类器训练过程考虑了产品购买意向分类和对话语句意图分类之间的关联性，提高了预设多因子意向分类器对客户的产品购买意向识别的准确性，进而根据较高的识别结果设置客户的客户标签，以根据客户标签提供工人服务，根据人工 服务的结果更新客户标签，提高了客户标签的准确性，以便后续根据客户标签进行准确地客户筛选，进而提高了客户服务的质量，在此基础上，还根据人工服务过程中客户的对话数据对预设多因子意向分类器进行不断地更新，进一步提高了预设多因子意向分类器对客户意向识别的准确性，从而进一步提高对客户筛选的准确性。
本申请的一个或多个实施例的细节在下面的附图和描述中提出，本申请的其他特征和优点将从说明书、附图以及权利要求变得明显。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例的描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例中基于深度学习的客户标签确定方法的一应用环境示意图；
图2是本申请一实施例中基于深度学习的客户标签确定方法的一流程示意图；
图3是图2中步骤S30的一实现流程示意图；
图4是图2中步骤S30的另一实现流程示意图；
图5是图2中步骤S50的一实现流程示意图；
图6是本申请一实施例中预设多因子意向分类器的一获取流程示意图；
图7是图6中步骤S06的一实现流程示意图；
图8是本申请一实施例中基于深度学习的客户标签确定装置的一结构示意图；
图9是本申请一实施例中计算机设备的一结构示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请实施例提供的基于深度学习的客户标签确定方法，可应用在如图1的应用环境中，其中，客户端通过网络与服务器进行通信。服务器通过获取客户端中客户与机器人客服的对话内容，并将对话内容输入预设多因子意向分类器，以获取预设多因子意向分类器输出的产品购买意向识别结果，预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，客户标签包括对产品有购买意向的高意向客户、拒绝产品的低意向客户和未对产品表态的中性客户，然后根据产品购买意向识别结果设置客户的客户标签，并根据客户的客户标签确定是否为客户提供人工服务，若为客户提供人工服务，则获取人工服务的结果和人工服务中客户的对话数据，最后根据人工服务的结果更新客户的客户标签，并根据客户的对话数据更新预设多因子意向分类器。
其中，客户端可以但不限于各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备。服务器可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在一实施例中，如图2所示，提供一种基于深度学习的客户标签确定方法，以该方法应用在图1中的服务器为例进行说明，包括如下步骤：
S10：获取客户与机器人客服的对话内容。
在客户通过客户端与与机器人客服进行对话沟通时，获取客户与机器人客服的对话内容，以便后续根据客户输入的对话内容识别出客户对产品的购买意向。
S20：将对话内容输入预设多因子意向分类器，以获取预设多因子意向分类器输出的 产品购买意向识别结果，预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，客户标签包括对产品有购买意向的高意向客户、拒绝产品的低意向客户和未对产品表态的中性客户。
在获取客户与机器人客服的对话内容之后，将获取的对话内容输入存储至区块链数据库中的预设多因子意向分类器，以获得预设多因子意向分类器输出的客户的产品购买意向识别结果，以确定客户是否具有产品购买意向。
其中，预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，多种客户标签包括对产品有购买意向的高意向客户、拒绝产品的低意向客户和未对产品表态的中性客户，进一步提高预设多因子意向分类器的训练数据的多样性，增加预设多因子意向分类器的准确性。与传统的意向分类器不同，预设多因子意向分类器考虑了客户的产品购买意向与客户在不同对话阶段的意图之间的关联性，对客户的产品购买意向识别准确性更高。在本实施例中，通过预设多因子意向分类器对客户在对话过程中的产品购买意向进行识别，实现了人工智能+意向识别的自动化处理，无需人工参与即可快递而准确地获得意向识别结果，提高了识别效率和准确性。
此外，在根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练的过程中，所获取的相关数据信息和生成的预设多因子意向分类器，均可保存在区块链网络中。
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。本实施例中将预设多因子意向分类器和相关数据保存在区块链网络，便于对目标分类器和数据进行快速查询，提高处理速度。
S30：根据产品购买意向识别结果设置客户的客户标签，并根据客户的客户标签确定是否为客户提供人工服务。
在获取预设多因子意向分类器输出的产品购买意向识别结果之后，根据产品购买意向识别结果设置客户的客户标签，并根据设置的客户标签确定是否为客户提供人工服务。
例如，若根据客户的客户标签为中性客户，则确定不为客户提供人工服务，继续控制机器人客服与客户对话，直至对话结束。
S40：若为客户提供人工服务，则获取人工服务的结果和人工服务中客户的对话数据。
若根据客户的客户标签确定为客户提供人工服务，则将客户转接至人工客户处，以让经验丰富的客服人员为客户提供个性化的人工服务，在为客户提供人工服务的过程中，记录人工服务中客户的对话数据和人工服务的结果，以便后续对人工服务的结果和人工服务中客户的对话数据进行查询、提取。
其中，人工服务可以是产品推销服务，在其他实施例中，人工服务还可以是其他类型的服务。本实施例中，以人工服务为产品推销服务为例进行说明。
S50：根据人工服务的结果更新客户的客户标签，并根据客户的对话数据更新预设多因子意向分类器。
在获取人工服务的结果和人工服务中客户的对话数据之后，根据人工服务的结果更新客户的客户标签，以提高客户标签的准确性，进而为后续为不同客户标签的客户提供不同服务策略提供基础；此外，还会根据人工服务中客户的对话数据更新预设多因子意向分类器，即根据人工服务过程中的对话数据重新训练预设多因子意向分类器，以提高预设多因子意向分类器对客户的产品购买意向识别的准确性，使得随着服务器的不断运行，预设多因子意向分类器预测出的结果的准确性越来越高。
本实施例中，通过获取客户与机器人客服的对话内容，并将对话内容输入预设多因子意向分类器，以获取预设多因子意向分类器输出的产品购买意向识别结果，再根据产品购买意向识别结果设置客户的客户标签，并根据客户的客户标签确定是否为客户提供人工服务，若为客户提供人工服务，则获取人工服务的结果和人工服务中客户的对话数据，最后根据人工服务的结果更新客户的客户标签，并根据客户的对话数据更新预设多因子意向分类器；在预设多因子意向分类器训练过程考虑了产品购买意向分类和对话语句意图分类之间的关联性，提高了预设多因子意向分类器对客户的产品购买意向识别的准确性，进而根据识别结果设置客户的客户标签，以根据客户标签提供工人服务，根据人工服务的结果更新客户标签，提高了客户标签的准确性，以便后续根据客户标签进行准确地客户筛选，进而提高了客户服务的质量，在此基础上，还根据人工服务过程中客户的对话数据对预设多因子意向分类器进行不断地更新，进一步提高了预设多因子意向分类器对客户意向识别的准确性，从而进一步提高对客户筛选的准确性。
在一实施例中，在获取预设多因子意向分类器输出的产品购买意向识别结果之后，如图3所示，步骤S30中，即根据产品购买意向识别结果设置客户的客户标签，具体包括如下步骤：
S31：确定产品购买意向识别结果是否为高购买意向。
在获取预设多因子意向分类器输出的产品购买意向识别结果之后，确定产品购买意向识别结果是否为高购买意向，以根据确定结果设置客户标签。
S32：若确定产品购买意向识别结果为高购买意向，则在对话过程中，确定预设多因子意向分类器输出高购买意向的次数是否大于预设次数。
在确定产品购买意向识别结果是否为高购买意向之后，若确定产品购买意向识别结果为高购买意向，则在对话过程中，对预设多因子意向分类器输出高购买意向的次数进行统计，并确定预设多因子意向分类器输出高购买意向的次数是否大于预设次数。
其中，在本次对话过程中，若预设多因子意向分类器输出高购买意向的次数大于0且不大于预设次数，则将客户的客户标签设为低意向客户。
S33：若预设多因子意向分类器输出高购买意向的次数大于预设次数，则将客户的客户标签设为高意向客户。
在确定预设多因子意向分类器输出高购买意向的次数是否大于预设次数之后，在本次对话过程中，若预设多因子意向分类器输出高购买意向的次数大于预设次数，则将客户的客户标签设为高意向客户，以便后续根据客户标签向高意向客户提供个性化的人工服务，从而提高客户的服务体验。
本实施例中，在获取预设多因子意向分类器输出的产品购买意向识别结果之后，确定产品购买意向识别结果是否为高购买意向，确定产品购买意向识别结果为高购买意向，则在对话过程中，确定预设多因子意向分类器输出高购买意向的次数是否大于预设次数，若预设多因子意向分类器输出高购买意向的次数大于预设次数，则将客户的客户标签设为高意向客户，细化了根据产品购买意向识别结果设置客户的客户标签的步骤，明确了根据预设多因子意向分类器输出高购买意向的次数确定高意向客户的过程，为后续为不同客户标签的客户提供不同的服务策略提供了基础。
在一实施例，若确定产品购买意向识别结果为低购买意向，且预设多因子意向分类器输出低购买意向的次数大于第一预设阈值，则将客户的客户标签设为低意向客户；确定产品购买意向识别结果为未对产品进行表态的中性意向，且预设多因子意向分类器输出中性意向的次数大于第二预设阈值，则将客户的客户标签设为中性客户，以便后续根据客户的客户标签进行客户群体筛选，进而提供不同的客户服务。
在一实施例中，如图4所示，步骤S30中，即根据客户的客户标签确定是否为客户提供人工服务，具体包括如下步骤：
S34：确定客户的客户标签是否为高意向客户。
在根据产品购买意向识别结果设置客户的客户标签之后，确定客户的客户标签是否为高意向客户。
S35：若确定客户的客户标签为高意向客户，则确定为客户提供人工服务。
在确定客户的客户标签是否为高意向客户之后，若客户的客户标签为高意向客户，则确定为客户提供人工服务，将客户转接给人工客服，以提高客户的服务体验。
其中，在确定客户的客户标签是否为高意向客户之后，若确定客户的客户标签不为高意向客户，即客户标签为低意向客户或者中性客户，则确定不为客户提供人工服务，继续由机器人客服与客户进行对话沟通，直至对话结束。
本实施例中，在根据产品购买意向识别结果设置客户的客户标签之后，通过确定客户的客户标签是否为高意向客户，以根据确定结果确定是否提供人工服务，若确定客户的客户标签为高意向客户，则确定为客户提供人工服务，进一步细化了根据客户的客户标签确定是否为客户提供人工服务的步骤，优化了不同客户标签的客户服务策略，提高了高意向客户的客户服务体验。
在一实施例中，在获取人工服务的结果之后，如图5所示，步骤S50中，即根据人工服务的结果更新客户的客户标签，具体包括如下步骤：
S51：根据人工服务的结果确定产品的交易是否达成。
在为客户提供产品推销的人工服务，并获取人工服务的结果之后，根据人工服务的结果确定产品的交易是否达成，以根据交易达成的情况确定是否需要更新客户标签。
S52：若产品的交易未达成，则获取人工客服对客户的标注结果。
在向高意向客户提供产品推荐服务的过程中，若产品交易未达成，人工客服可以对客户的产品购买意向进行重新标注。因此，在根据人工服务的结果确定产品的交易是否达成之后，若确定产品的交易未达成，则可获取人工客服对客户的产品购买意向的标注结果，以便后续根据标注结果更新客户的客户标签。
S53：根据人工客服对客户的标注结果更新客户的客户标签。
在确定产品的交易未达成，并获取人工客服对客户的标注结果之后，根据人工客服对客户的标注结果更新客户的客户标签。
例如，若确定产品的交易未达成，人工客服对客户的标注结果为低意向客户，则将客户的客户标签从高意向客户更新为低意向客户。
本实施例中，在为客户提供产品推销的人工服务，并获取人工服务的结果之后，通过根据人工服务的结果确定产品的交易是否达成，若产品的交易未达成，则获取人工客服对客户的标注结果，进而根据人工客服对所述客户的标注结果更新客户的客户标签，细化了根据人工服务的结果更新客户的客户标签的步骤，当向高意向客户推送若干服务后，根据交易达成情况和人工客服的标注结果对客户标签进行更新，进一步提高了客户标签的准确性，便于后续根据客户标签筛选出高质量客户。
在一实施例中，在根据人工服务的结果确定产品的交易是否达成之后，若产品的交易达成，则保持客户的客户标签为高意向客户不便，并将客户备注为交易达成客户，以进行一步细化客户的标签，便于后续根据客户标签和备注结果筛选客户。
在一实施例中，将对话内容输入预设多因子意向分类器之前，还需要根据多个客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练，以获得预设多因子意向分类器。如图6所示，预设多因子意向分类器通过如下方式获取：
S01：获取不同客户标签的客户对话数据，客户标签包括高意向客户、低意向客户和中性客户。
获取历史存储的、不同客户标签的客户对话数据，其中，客户标签包括对产品有购买意向的高意向客户、拒绝产品的低意向客户和未对产品表态的中性客户，即客户对话数据 包括高意向客户的对话数据、低意向客户的对话数据和中性客户的对话数据。
在一实施例中，在获取同客户标签的客户对话数据之前，需要根据每个客户的历史对话数据确定客户对产品的总体购买意向，再根据客户对产品的总体购买意向设定客户的客户标签，以获取足够的、多客户标签的客户对话数据进行预设多因子意向分类器训练。
例如，可以通过查询历史存储的客户对话数据，确定客户的对话数据是否跟产品相关：若通过客户对话语句(如：如何查看保单、咨询保单保障内容、如何购买等语句)的意图发现客户对产品问题比较感兴趣，且客户对产品进行了正面的询问或了解，则确定该客户对产品的购买意向为正，即用户具有产品购买意向，则将客户的客户标签设置为高意向客户，该客户的对话数据为正意向数据；若客户的问题与产品相关，但通过客户对话语句(如：没时间、不需要等语句)的意图确定客户态度是质疑甚至是抱怨反感的，则确定客户对产品的购买意向负，即用户拒绝了解、购买产品，则将客户的客户标签设置为低意向客户，该客户的对话数据为负意向数据；若客户的对话数据中，通过客户的对话语句确定客户的意图与业产品无关的(如打招呼、结束语等)，则将确定该客户未对产品表态，无法确定客户的产品购买意向，则将客户的客户标签设定为中性客户，该客户的对话数据为中性数据。
S02：将客户标签为高意向客户的客户对话数据作为正意向数据集。
在获取不同客户标签的客户对话数据之后，将客户标签为高意向客户的客户对话数据作为正意向数据集，即将对产品有兴趣、具有产品购买意向的客户的对话数据作为正意向数据集。
S03：将客户标签为低意向客户的客户对话数据作为负意向数据集。
在获取不同客户标签的客户对话数据之后，将客户标签为低意向客户的客户对话数据作为负意向数据集，即将拒绝了解、拒绝购买产品的客户的对话数据作为负意向数据集。
S04：将客户标签为中性客户的客户对话数据作为中性数据集。
在获取不同客户标签的客户对话数据之后，将客户标签为中性客户的客户对话数据作为中性数据集，即将为对产品表态的客户的对话数据作为中性数据集。
S05：将正意向数据集、负意向数据集和中性数据集的对话数据汇总为意图数据，并识别意图数据中每一话语句的意图，以获得意图数据集。
在获得将正意向数据集、负意向数据集和中性数据集之后，将正意向数据集、负意向数据集和中性数据集的客户对话数据汇总为意图数据，并识别出意图数据中每一话语句的意图，以获得意图数据集，意图数据集包括意图数据和意图数据中每一对话语句对应的意图。
S06：根据正意向数据集、负意向数据集、中性数据集和意图数据集进行分类器训练，以获取预设多因子意向分类器。
在获取正意向数据集、负意向数据集、中性数据集和意图数据集之后，根据正意向数据集、负意向数据集、中性数据集和意图数据集进行产品购买意向分类和对话语句意图分类训练，以获得预设多因子意向分类器。在训练预设多因子意向分类器的过程中，不仅考虑产品购买的意向分类任务和对话语句的意图分类任务这两个任务的联合学习效果，还考虑意向分类任务和意图分类任务之间的关联性，通过该方式获得的预设多因子意向分类器，对客户的意向识别准确性更高，可以更准确地预测出客户对产品购买意向。
本实施例中，通过获取不同客户标签的客户对话数据，客户标签包括高意向客户、低意向客户和中性客户，将客户标签为高意向客户的客户对话数据作为正意向数据集，将客户标签为低意向客户的客户对话数据作为负意向数据集，将客户标签为中性客户的客户对话数据作为中性数据集，将正意向数据集、负意向数据集和中性数据集的对话数据汇总为意图数据，并识别意图数据中每一话语句的意图，以获得意图数据集，根据正意向数据集、负意向数据集、中性数据集和意图数据集进行分类器训练，以获取预设多因子意向分类器， 明确了训练数据的来源和类型，明确了获取预设多因子意向分类器的过程，为后续在对话过程中根据预设多因子意向分类器识别客户的产品购买意向提供了基础，从而提高了识别出客户的产品购买意向的准确性。
在一实施例中，如图7所示，步骤S06中，即根据正意向数据集、负意向数据集、中性数据集和意图数据集进行分类器训练，以获取预设多因子意向分类器，具体包括如下步骤：
S061：根据正意向数据集、负意向数据集和中性数据集进行意向分类学习，以获取意向分类学习结果。
在获取意向数据集、负意向数据集和中性数据集之后，根据正意向数据集、负意向数据集和中性数据集进行意向分类学习，以获取意向分类学习结果。其中，意向分类学习结果为客户的产品购买意向识别结果。
S062：根据意图数据集进行意图分类学习，以获取意图分类学习结果。
在获取意图数据集之后，根据意图数据集进行意图分类学习，以获取意图分类学习结果，其中，意图分类学习结果为对话语句中的阶段性对话或者单一对话语句中的意图识别结果。
S063：根据意图分类学习结果和意向分类学习结果对意向分类学习进行调整，以获得预设多因子意向分类器。
在获取意图分类学习结果和意向分类学习结果之后，根据意图分类学习结果和意向分类学习结果对意向分类学习进行调整训练，以获得预设多因子意向分类器。即在预设多因子意向分类器训练过程中，不仅考虑意图分类学习与意向分类学习的联合学习效果，同时还考虑意图分类学习与意向分类学习之间的关联性。而考虑意图分类学习与意向分类学习之间的关联性，是通过考虑意图分类学习的预测结果与意向分类学习的预测结果的一致性，进而考虑意图分类学习的预测结果与真实的产品购买意向结果的一致性，以提高训练的预设多因子意向分类器的准确性。
本实施例中，通过根据正意向数据集、负意向数据集和中性数据集进行意向分类学习，以获取意向分类学习结果，根据意图数据集进行意图分类学习，以获取意图分类学习结果，根据意图分类学习结果和意向分类学习结果对意向分类学习进行调整，以获得预设多因子意向分类器，进一步细化了根据正意向数据集、负意向数据集、中性数据集和意图数据集进行分类器训练，以获取预设多因子意向分类器的过程，充分考虑了客户对话语句数据的意图对客户的产品购买意向识别的影响，通过根据意图分类学习结果和意向分类学习结果对意向分类学习进行调整，考虑了意图分类和意向分类的关联性，从而提高了预设多因子意向分类器的准确性，为后续在与客户对话过程中准确地识别客户的产品购买意向提供了基础。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
在一实施例中，提供一种基于深度学习的客户标签确定装置，该基于深度学习的客户标签确定装置与上述实施例中基于深度学习的客户标签确定方法一一对应。如图8所示，该基于深度学习的客户标签确定装置包括第一获取模块801、输入模块802、设置模块803、第二获取模块804和更新模块805。各功能模块详细说明如下：
第一获取模块801，用于获取客户与机器人客服的对话内容；
输入模块802，用于将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
设置模块803，用于根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
第二获取模块804，用于若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
更新模块805，用于根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
进一步地，所述设置模块803具体用于：
确定所述产品购买意向识别结果是否为高购买意向；
若确定所述产品购买意向识别结果为所述高购买意向，则在对话过程中，确定所述预设多因子意向分类器输出所述高购买意向的次数是否大于预设次数；
若所述预设多因子意向分类器输出所述高购买意向的次数大于所述预设次数，则将所述客户的客户标签设为高意向客户。
进一步地，所述设置模块803具体还用于：
若所述客户的客户标签为高意向客户，则确定为所述客户提供所述人工服务。
进一步地，所述更新模块805具体用于：
根据所述人工服务的结果确定所述产品的交易是否达成；
若所述产品的交易未达成，则获取人工客服对所述客户的标注结果；
根据所述人工客服对所述客户的标注结果更新所述客户的客户标签。
进一步地，所述基于深度学习的客户标签确定装置还包括第三获取模块806，所述第三获取模块806具体用于：
获取不同客户标签的客户对话数据，所述客户标签包括所述高意向客户、所述低意向客户和所述中性客户；
将所述客户标签为高意向客户的客户对话数据作为正意向数据集；
将所述客户标签为低意向客户的客户对话数据作为负意向数据集；
将所述客户标签为中性客户的客户对话数据作为中性数据集；
将所述正意向数据集、所述负意向数据集和所述中性数据集的对话数据汇总为意图数据，并识别所述意图数据中每一话语句的意图，以获得意图数据集；
根据所述正意向数据集、所述负意向数据集、所述中性数据集和意图数据集进行分类器训练，以获取所述预设多因子意向分类器。
进一步地，所述第三获取模块806具体还用于：
根据所述正意向数据集、所述负意向数据集和所述中性数据集进行意向分类学习，以获取意向分类学习结果；
根据所述意图数据集进行意图分类学习，以获取意图分类学习结果；
根据所述意图分类学习结果和所述意向分类学习结果对所述意向分类学习进行调整，以获得所述预设多因子意向分类器。
关于基于深度学习的客户标签确定装置的具体限定可以参见上文中对于基于深度学习的客户标签确定方法的限定，在此不再赘述。上述基于深度学习的客户标签确定装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图9所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质和/或易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作系统和计算 机可读指令的运行提供环境。该计算机设备的数据库用于预设多因子意向分类器和基于深度学习的客户标签确定方法中所应用、生成的数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种基于深度学习的客户标签确定方法。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机可读指令，处理器执行计算机可读指令时实现以下步骤：
获取客户与机器人客服的对话内容；
将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机可读指令，计算机可读指令被处理器执行时实现以下步骤：
获取客户与机器人客服的对话内容；
将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功 能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种基于深度学习的客户标签确定方法，其中，包括：
    获取客户与机器人客服的对话内容；
    将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
    根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
    若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
    根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
  2. 如权利要求1所述的基于深度学习的客户标签确定方法，其中，所述根据所述产品购买意向识别结果设置所述客户的客户标签，包括：
    确定所述产品购买意向识别结果是否为高购买意向；
    若确定所述产品购买意向识别结果为所述高购买意向，则在对话过程中，确定所述预设多因子意向分类器输出所述高购买意向的次数是否大于预设次数；
    若所述预设多因子意向分类器输出所述高购买意向的次数大于所述预设次数，则将所述客户的客户标签设为高意向客户。
  3. 如权利要求1所述的基于深度学习的客户标签确定方法，其中，所述根据所述客户的客户标签确定是否为所述客户提供人工服务，包括：
    若所述客户的客户标签为高意向客户，则确定为所述客户提供所述人工服务。
  4. 如权利要求1所述的基于深度学习的客户标签确定方法，其中，所述根据所述人工服务的结果更新所述客户的客户标签，包括：
    根据所述人工服务的结果确定所述产品的交易是否达成；
    若所述产品的交易未达成，则获取人工客服对所述客户的标注结果；
    根据所述人工客服对所述客户的标注结果更新所述客户的客户标签。
  5. 如权利要求1-4任一项所述的基于深度学习的客户标签确定方法，其中，所述预设多因子意向分类器通过如下方式获取：
    获取不同客户标签的客户对话数据，所述客户标签包括所述高意向客户、所述低意向客户和所述中性客户；
    将所述客户标签为高意向客户的客户对话数据作为正意向数据集；
    将所述客户标签为低意向客户的客户对话数据作为负意向数据集；
    将所述客户标签为中性客户的客户对话数据作为中性数据集；
    将所述正意向数据集、所述负意向数据集和所述中性数据集的对话数据汇总为意图数据，并识别所述意图数据中每一话语句的意图，以获得意图数据集；
    根据所述正意向数据集、所述负意向数据集、所述中性数据集和意图数据集进行分类器训练，以获取所述预设多因子意向分类器。
  6. 如权利要求5所述的基于深度学习的客户标签确定方法，其中，所述根据所述正意向数据集、所述负意向数据集、所述中性数据集和意图数据集进行分类器训练，以获取所述预设多因子意向分类器，包括：
    根据所述正意向数据集、所述负意向数据集和所述中性数据集进行意向分类学习，以获取意向分类学习结果；
    根据所述意图数据集进行意图分类学习，以获取意图分类学习结果；
    根据所述意图分类学习结果和所述意向分类学习结果对所述意向分类学习进行调整，以获得所述预设多因子意向分类器。
  7. 一种基于深度学习的客户标签确定装置，其中，包括：
    第一获取模块，用于获取客户与机器人客服的对话内容；
    输入模块，用于将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
    设置模块，用于根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
    第二获取模块，用于若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
    更新模块，用于根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
  8. 如权利要求7所述的基于深度学习的客户标签确定装置，其中，所述设置模块具体用于：
    确定所述产品购买意向识别结果是否为高购买意向；
    若确定所述产品购买意向识别结果为所述高购买意向，则确定在对话过程中，所述预设多因子意向分类器输出所述高购买意向的次数是否大于预设次数；
    若所述预设多因子意向分类器输出所述高购买意向的次数大于所述预设次数，则将所述客户的客户标签设为高意向客户。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，其中，所述处理器执行所述计算机可读指令时实现如下步骤：
    获取客户与机器人客服的对话内容；
    将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
    根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
    若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
    根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
  10. 如权利要求9所述的计算机设备，其中，所述根据所述产品购买意向识别结果设置所述客户的客户标签，包括：
    确定所述产品购买意向识别结果是否为高购买意向；
    若确定所述产品购买意向识别结果为所述高购买意向，则在对话过程中，确定所述预设多因子意向分类器输出所述高购买意向的次数是否大于预设次数；
    若所述预设多因子意向分类器输出所述高购买意向的次数大于所述预设次数，则将所述客户的客户标签设为高意向客户。
  11. 如权利要求9所述的计算机设备，其中，所述根据所述客户的客户标签确定是否为所述客户提供人工服务，包括：
    若所述客户的客户标签为高意向客户，则确定为所述客户提供所述人工服务。
  12. 如权利要求9所述的计算机设备，其中，所述根据所述人工服务的结果更新所述客户的客户标签，包括：
    根据所述人工服务的结果确定所述产品的交易是否达成；
    若所述产品的交易未达成，则获取人工客服对所述客户的标注结果；
    根据所述人工客服对所述客户的标注结果更新所述客户的客户标签。
  13. 如权利要求9-12任一项所述的计算机设备，其中，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取不同客户标签的客户对话数据，所述客户标签包括所述高意向客户、所述低意向客户和所述中性客户；
    将所述客户标签为高意向客户的客户对话数据作为正意向数据集；
    将所述客户标签为低意向客户的客户对话数据作为负意向数据集；
    将所述客户标签为中性客户的客户对话数据作为中性数据集；
    将所述正意向数据集、所述负意向数据集和所述中性数据集的对话数据汇总为意图数据，并识别所述意图数据中每一话语句的意图，以获得意图数据集；
    根据所述正意向数据集、所述负意向数据集、所述中性数据集和意图数据集进行分类器训练，以获取所述预设多因子意向分类器。
  14. 如权利要求13所述的计算机设备，其中，所所述根据所述正意向数据集、所述负意向数据集、所述中性数据集和意图数据集进行分类器训练，以获取所述预设多因子意向分类器，包括：
    根据所述正意向数据集、所述负意向数据集和所述中性数据集进行意向分类学习，以获取意向分类学习结果；
    根据所述意图数据集进行意图分类学习，以获取意图分类学习结果；
    根据所述意图分类学习结果和所述意向分类学习结果对所述意向分类学习进行调整，以获得所述预设多因子意向分类器。
  15. 一个或多个存储有计算机可读指令的可读存储介质，其中，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
    获取客户与机器人客服的对话内容；
    将所述对话内容输入预设多因子意向分类器，以获取所述预设多因子意向分类器输出的产品购买意向识别结果，所述预设多因子意向分类器根据多种客户标签的客户对话数据进行产品购买意向分类和对话语句意图分类训练获得的意向分类器，所述客户标签包括对产品有购买意向的高意向客户、拒绝所述产品的低意向客户和未对所述产品表态的中性客户；
    根据所述产品购买意向识别结果设置所述客户的客户标签，并根据所述客户的客户标签确定是否为所述客户提供人工服务；
    若为所述客户提供人工服务，则获取所述人工服务的结果和所述人工服务中所述客户的对话数据；
    根据所述人工服务的结果更新所述客户的客户标签，并根据所述客户的对话数据更新所述预设多因子意向分类器。
  16. 如权利要求15所述的可读存储介质，其中，所述根据所述产品购买意向识别结果设置所述客户的客户标签，包括：
    确定所述产品购买意向识别结果是否为高购买意向；
    若确定所述产品购买意向识别结果为所述高购买意向，则在对话过程中，确定所述预设多因子意向分类器输出所述高购买意向的次数是否大于预设次数；
    若所述预设多因子意向分类器输出所述高购买意向的次数大于所述预设次数，则将所述客户的客户标签设为高意向客户。
  17. 如权利要求15所述的可读存储介质，其中，所述根据所述客户的客户标签确定是否为所述客户提供人工服务，包括：
    若所述客户的客户标签为高意向客户，则确定为所述客户提供所述人工服务。
  18. 如权利要求15所述的可读存储介质，其中，所述根据所述人工服务的结果更新所述客户的客户标签，包括：
    根据所述人工服务的结果确定所述产品的交易是否达成；
    若所述产品的交易未达成，则获取人工客服对所述客户的标注结果；
    根据所述人工客服对所述客户的标注结果更新所述客户的客户标签。
  19. 如权利要求15-18任一项所述的可读存储介质，其中，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器还执行如下步骤：
    获取不同客户标签的客户对话数据，所述客户标签包括所述高意向客户、所述低意向客户和所述中性客户；
    将所述客户标签为高意向客户的客户对话数据作为正意向数据集；
    将所述客户标签为低意向客户的客户对话数据作为负意向数据集；
    将所述客户标签为中性客户的客户对话数据作为中性数据集；
    将所述正意向数据集、所述负意向数据集和所述中性数据集的对话数据汇总为意图数据，并识别所述意图数据中每一话语句的意图，以获得意图数据集；
    根据所述正意向数据集、所述负意向数据集、所述中性数据集和意图数据集进行分类器训练，以获取所述预设多因子意向分类器。
  20. 如权利要求19所述的可读存储介质，其中，所述根据所述正意向数据集、所述负意向数据集、所述中性数据集和意图数据集进行分类器训练，以获取所述预设多因子意向分类器，包括：
    根据所述正意向数据集、所述负意向数据集和所述中性数据集进行意向分类学习，以获取意向分类学习结果；
    根据所述意图数据集进行意图分类学习，以获取意图分类学习结果；
    根据所述意图分类学习结果和所述意向分类学习结果对所述意向分类学习进行调整，以获得所述预设多因子意向分类器。
PCT/CN2020/117495 2020-08-06 2020-09-24 基于深度学习的客户标签确定方法、装置、设备及介质 WO2021135441A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/620,736 US20220414687A1 (en) 2020-08-06 2020-09-24 Method, device, equipment and medium for determining customer tabs based on deep learning

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010783681.9A CN111914075A (zh) 2020-08-06 2020-08-06 基于深度学习的客户标签确定方法、装置、设备及介质
CN202010783681.9 2020-08-06

Publications (1)

Publication Number Publication Date
WO2021135441A1 true WO2021135441A1 (zh) 2021-07-08

Family

ID=73287268

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/117495 WO2021135441A1 (zh) 2020-08-06 2020-09-24 基于深度学习的客户标签确定方法、装置、设备及介质

Country Status (3)

Country Link
US (1) US20220414687A1 (zh)
CN (1) CN111914075A (zh)
WO (1) WO2021135441A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112488750A (zh) * 2020-11-27 2021-03-12 上海容大数字技术有限公司 一种保险场景智能推荐续保系统
CN112734467A (zh) * 2020-12-31 2021-04-30 北京明略软件系统有限公司 线下服务场景的客流量预测方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109146610A (zh) * 2018-07-16 2019-01-04 众安在线财产保险股份有限公司 一种智能保险推荐方法、装置及智能保险机器人设备
US20190139538A1 (en) * 2015-10-21 2019-05-09 Google Llc Parameter collection and automatic dialog generation in dialog systems
CN110110058A (zh) * 2019-04-22 2019-08-09 南京硅基智能科技有限公司 一种外呼机器人的意向打分方法及系统
CN111324727A (zh) * 2020-02-19 2020-06-23 百度在线网络技术（北京）有限公司 用户意图识别方法、装置、设备和可读存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108521525A (zh) * 2018-04-03 2018-09-11 南京甄视智能科技有限公司 基于用户标签体系的智能机器人客服营销方法和系统
CN109151008A (zh) * 2018-08-09 2019-01-04 北京中关村科金技术有限公司 一种无感知的人机对话控流方法和系统
CN109688281A (zh) * 2018-12-03 2019-04-26 复旦大学 一种智能语音交互方法及系统
US10915905B1 (en) * 2018-12-13 2021-02-09 Square, Inc. Batch-processing transactions in response to an event
CN110738545A (zh) * 2019-08-30 2020-01-31 深圳壹账通智能科技有限公司 基于用户意向识别的产品推荐方法、装置、计算机设备和存储介质
US20220101371A1 (en) * 2020-06-16 2022-03-31 Kaci Anne Beerbower System for online advertising analytics

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190139538A1 (en) * 2015-10-21 2019-05-09 Google Llc Parameter collection and automatic dialog generation in dialog systems
CN109146610A (zh) * 2018-07-16 2019-01-04 众安在线财产保险股份有限公司 一种智能保险推荐方法、装置及智能保险机器人设备
CN110110058A (zh) * 2019-04-22 2019-08-09 南京硅基智能科技有限公司 一种外呼机器人的意向打分方法及系统
CN111324727A (zh) * 2020-02-19 2020-06-23 百度在线网络技术（北京）有限公司 用户意图识别方法、装置、设备和可读存储介质

Also Published As

Publication number Publication date
CN111914075A (zh) 2020-11-10
US20220414687A1 (en) 2022-12-29

Similar Documents

Publication Publication Date Title
US11663409B2 (en) Systems and methods for training machine learning models using active learning
US10997258B2 (en) Bot networks
CA3069936C (en) System and method for identifying and providing personalized self-help content with artificial intelligence in a customer self-help system
CN109871446B (zh) 意图识别中的拒识方法、电子装置及存储介质
US20190180196A1 (en) Systems and methods for generating and updating machine hybrid deep learning models
US11144827B1 (en) Cooperatively operating a network of supervised learning processors to concurrently distribute supervised learning processor training and provide predictive responses to input data
WO2021243828A1 (zh) 基于机器学习的文本处理方法、装置、计算机设备及介质
WO2020147395A1 (zh) 基于情感的文本分类处理方法、装置和计算机设备
WO2021068321A1 (zh) 基于人机交互的信息推送方法、装置和计算机设备
CN110070391B (zh) 数据处理方法、装置、计算机可读介质及电子设备
WO2019113122A1 (en) Systems and methods for improved machine learning for conversations
US20190179903A1 (en) Systems and methods for multi language automated action response
WO2022142006A1 (zh) 基于语义识别的话术推荐方法、装置、设备及存储介质
WO2020119031A1 (zh) 基于深度学习的问答反馈方法、装置、设备及存储介质
WO2021135441A1 (zh) 基于深度学习的客户标签确定方法、装置、设备及介质
CN109829629A (zh) 风险分析报告的生成方法、装置、计算机设备和存储介质
CN114528844A (zh) 意图识别方法、装置、计算机设备及存储介质
CN114399396A (zh) 保险产品推荐方法、装置、计算机设备及存储介质
CN113919437A (zh) 生成客户画像的方法、装置、设备及存储介质
CN113868391B (zh) 基于知识图谱的法律文书生成方法、装置、设备及介质
US10896034B2 (en) Methods and systems for automated screen display generation and configuration
CN116628163A (zh) 客服服务处理方法、装置、设备及存储介质
CN117093682A (zh) 意图识别方法、装置、计算机设备及存储介质
CN113255368B (zh) 针对文本数据进行情感分析的方法、装置及相关设备
US20220292393A1 (en) Utilizing machine learning models to generate initiative plans

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: 20910784

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: 20910784

Country of ref document: EP

Kind code of ref document: A1