WO2023124215A1 - 用户问题的标注方法及装置 - Google Patents

用户问题的标注方法及装置 Download PDF

Info

Publication number
WO2023124215A1
WO2023124215A1 PCT/CN2022/117665 CN2022117665W WO2023124215A1 WO 2023124215 A1 WO2023124215 A1 WO 2023124215A1 CN 2022117665 W CN2022117665 W CN 2022117665W WO 2023124215 A1 WO2023124215 A1 WO 2023124215A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
target
customer service
statement
service system
Prior art date
Application number
PCT/CN2022/117665
Other languages
English (en)
French (fr)
Inventor
耿福明
吴海英
蒋宁
Original Assignee
马上消费金融股份有限公司
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 马上消费金融股份有限公司 filed Critical 马上消费金融股份有限公司
Priority to EP22913488.7A priority Critical patent/EP4345645A1/en
Publication of WO2023124215A1 publication Critical patent/WO2023124215A1/zh
Priority to US18/395,240 priority patent/US20240127259A1/en

Links

Images

Classifications

    • 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
    • G06Q30/01—Customer relationship services
    • G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
    • 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
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • 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
    • 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
    • G06Q30/01—Customer relationship services
    • G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
    • G06Q30/016—After-sales

Definitions

  • the present application relates to the technical field of artificial intelligence, and in particular to a method and device for labeling user questions.
  • machine learning has been widely used in many fields.
  • machine learning technology can be applied to the field of intelligent robots.
  • intelligent robots can provide users with high-quality services.
  • intelligent customer service can be used to assist manual customer service to solve the problem of shortage of human resources.
  • the ability of intelligent customer service to handle business is limited, and sometimes it is impossible to identify the real intention of the user from the user's question, causing the user The experience is poor. Therefore, it is necessary to provide a technical solution capable of improving the recognition accuracy of the user intention recognition model used by the intelligent customer service.
  • the purpose of the embodiments of the present application is to provide a method and device for marking user questions.
  • the embodiment of the present application provides a method for marking user questions, the method comprising:
  • the target quick answer statement includes: at least one quick answer statement in the preset quick statement set, the The quick answer sentence is a reply sentence for quick reply to the user's question preset in the manual customer service system;
  • the target intent label corresponding to the target quick response statement Based on the preset first correspondence, determine the target intent label corresponding to the target quick response statement; wherein, the first correspondence includes the target quick response statement and the target Correspondence between intent tags;
  • the embodiment of the present application provides a device for tagging user questions, the device comprising:
  • the first acquisition module is used to acquire the target questions submitted to the manual customer service system, and acquire the target quick answer sentences corresponding to the target question questions; wherein, the target quick answer sentences include: preset shortcut sentence sets At least one quick response statement, the quick response statement is a response statement preset in the manual customer service system for quick reply to user questions;
  • the first determination module is configured to determine the target intent label corresponding to the target quick response sentence based on a preset first correspondence; wherein the first correspondence includes the target pre-stored in the manual customer service system The corresponding relationship between the quick response sentence and the target intent label;
  • the first generation module is used to mark the target question based on the target intention label, and generate a user intention identification sample set; wherein, the user intention identification sample set is used to identify the user intention used by the intelligent customer service system
  • the recognition model is trained.
  • the embodiment of the present application provides a device for tagging user questions, the device comprising:
  • a processor configured to execute instructions configured to perform calculations and executes instructions.
  • the executable instructions comprising steps for performing a method as described in the first aspect .
  • an embodiment of the present application provides a storage medium, wherein the storage medium is used to store computer-executable instructions, and the executable instructions cause a computer to execute the steps in the method as described in the first aspect.
  • FIG. 1 is a schematic flow chart of the first method for labeling user questions provided by the embodiment of the present application
  • Fig. 2 is a second schematic flow chart of the method for labeling user questions provided by the embodiment of the present application
  • FIG. 3 is a schematic flowchart of a third method for labeling user questions provided in the embodiment of the present application.
  • Fig. 4 is the schematic diagram of the shortcut language setting interface of the agent client in the labeling method of the user question provided by the embodiment of the present application;
  • FIG. 5 is a schematic diagram of the training process of the user intention recognition model of the user question labeling method provided by the embodiment of the present application.
  • FIG. 6 is a schematic diagram of an application scenario of a method for labeling user questions provided by an embodiment of the present application
  • FIG. 7 is a schematic diagram of the module composition of the user question labeling device provided by the embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of a device for tagging user questions provided by an embodiment of the present application.
  • the embodiment of the present application provides a method and device for labeling user questions.
  • acquiring the dialogue data generated by the manual customer service system first determine the target quick answer sentence corresponding to the target question, and then combine the preset quick answer sentence with the user intention
  • the corresponding relationship between tags determine the target intent tag corresponding to the target quick answer sentence, and then get the corresponding relationship between the target question question and the target intent tag, so as to realize the user intent labeling of user questions, and then the generated user intent
  • the identification sample set is applied to the user intention recognition model in the intelligent customer service system, that is, fully utilizing the dialogue data generated in the artificial customer service system, the user intention training sample set required in the intelligent customer service system is automatically generated, and the real business of the artificial customer service system Traffic is closely related to the user intention recognition requirements of the intelligent customer service system, which can not only improve the efficiency of labeling questions asked by users, but also provide a large number of user intention recognition samples for the intelligent customer service system, and the target questions submitted by users to the artificial customer service system Questions include not only user questions that
  • the user intent recognition sample set can also improve the coverage of user problems that cannot be identified by the intelligent customer service system included in the user intent recognition sample set, thereby improving the recognition accuracy of the user intent recognition model and ensuring the service quality of the intelligent customer service system.
  • the intelligent customer service system in order to use the intelligent customer service system to provide services to users, it is necessary to collect a large number of user intention recognition samples in advance to train the user intention recognition model used by the intelligent customer service system.
  • the intelligent customer service system In the process of collecting user intention recognition samples for training the user intention recognition model, the intelligent customer service system usually needs to record the user questions that it cannot accurately answer, and then manually mark the intentions of these user questions; and considering that in some cases, the user The manual customer service system will be directly selected, resulting in the intelligent customer service system being unable to recognize these user questions, resulting in manual intention labeling of user questions, which may miss some user questions that the intelligent customer service system cannot accurately answer, so there will be user intent recognition
  • the problem of low sample acquisition efficiency and limited number of acquisitions leads to the low recognition accuracy of the trained user intention recognition model, which makes it impossible for the intelligent customer service system to accurately understand the real intention of users to ask questions when providing services to users.
  • the technical solution of this application makes full use of the dialogue data generated in the manual customer service system (that is, it can include the correspondence between user questions and customer service response sentences), and combines The correspondence between the preset quick response sentences and user intention labels automatically generates the user intention training sample set required in the intelligent customer service system, and closely matches the real business flow of the manual customer service system with the user intention recognition requirements of the intelligent customer service system.
  • the user questions answered by the human customer service system also include user questions that are directly answered by the human customer service but cannot be recognized by the intelligent customer service.
  • the user intent recognition sample set can be obtained, and the user intent recognition can also be improved.
  • the coverage of user problems that cannot be identified by the intelligent customer service system contained in the sample set can improve the recognition accuracy of the user intention recognition model, thereby ensuring the service quality of the intelligent customer service system.
  • Figure 1 is a schematic flow diagram of the first method for labeling user questions provided by the embodiment of the present application.
  • the method in Figure 1 is based on the real business traffic generated by the manual customer service system, and automatically obtains the user intention identification sample set from the manual customer service system. process, the user intention recognition sample set is used to train the user intention recognition model used by the intelligent customer service system, and the labeling method is applied to the background server for user question labeling, which can be the first server of the manual customer service system, It can also be a second server connected to the first server of the manual customer service system.
  • the method at least includes the following steps:
  • the above-mentioned target quick answer statement includes: at least one quick answer statement in the preset shortcut statement set, the quick answer statement is a response statement preset in the manual customer service system for quickly replying to user questions;
  • the above-mentioned target quick answer The statement may be a quick answer statement that the target agent selects from the preset shortcut statement set and directly replies, and correspondingly, the quick answer statement that the target agent responds to the user's target question is determined as the target quick answer statement; specifically Yes, the target agent can select the target quick answer sentence that matches the target question from the preset multiple quick answer sentences, and the target agent client used by the target agent detects the selection input operation for the target quick answer sentence Finally, send the target quick answer sentence to the server of the manual customer service system, correspondingly, the server sends the target quick answer sentence to the client that submitted the target question, so as to realize the one-click reply to the user's target question, and the agent will The step of manually inputting the answer sentence is omitted; in the specific implementation, the target agent can
  • the click operation of the shortcut answer statement can also be the input operation of the shortcut key, that is, the corresponding relationship between the shortcut key and the quick answer statement is set in advance, so that the agent can quickly call out the corresponding shortcut key through the agent client by inputting the corresponding shortcut key.
  • Shortcut answer statement can also be the input operation of the shortcut key, that is, the corresponding relationship between the shortcut key and the quick answer statement is set in advance, so that the agent can quickly call out the corresponding shortcut key through the agent client by inputting the corresponding shortcut key.
  • the above-mentioned target quick answer sentence can also be based on the regular answer sentence (that is, the answer sentence manually input by the manual customer service) that the target agent replies to the user's target question, and is automatically matched from the preset shortcut sentence set.
  • Response sentences specifically, the regular response sentences replied by the target agent can be matched with each shortcut response sentence in the preset shortcut sentence set, and the target corresponding to the target question question can be determined based on the shortcut response sentences whose similarity satisfies the preset condition Shortcut answer statement.
  • the above preset condition can be that the similarity is greater than a preset threshold, and correspondingly, at least one quick answer sentence is matched from the preset quick sentence set and determined as a recommended quick answer sentence, and the at least one quick answer sentence is the same as the target agent
  • the similarity between the conventional answer sentences replied to the user's target questions is greater than the preset threshold, and the target agent selects and replies from the above-mentioned recommended quick answer sentences as the target quick answer sentence; specifically Yes, the process of automatically matching the target quick answer statement from the preset shortcut statement set for the regular answer statement based on the target agent's reply (that is, the answer statement manually input by the human customer service), may be that the target agent client detects the During the process of the target agent inputting the regular answer sentence in the information input box (that is, when the manually entered regular answer sentence is not sent to the server of the manual customer service system), the target agent client automatically determines the corresponding answer based on the pre-stored shortcut sentence set The recommended quick answer sentences matched with the regular answer
  • the process of automatically matching the target quick answer statement from the preset shortcut statement set for the conventional answer statement based on the reply of the target agent may also be the target agent client to the
  • the server side of the manual customer service system uploads the normal response sentences entered by the target agent in the information input box (that is, when the manual input conventional response sentences have been sent to the server side of the manual customer service system), the server side of the manual customer service system automatically based on The pre-stored shortcut statement set determines the target quick answer statement that matches the regular answer statement uploaded by the target agent client.
  • the target intent label corresponding to the target quick response sentence determines the target intent label corresponding to the target quick response sentence; wherein, the first correspondence includes the correspondence between the target quick response sentence and the target intent label pre-stored in the manual customer service system;
  • the agent can determine the questions that the user may ask and the real intention of the user to ask the question based on business experience, and upload the quick answer sentence to the server of the manual customer service system through the agent client, and upload the corresponding The user intent label, so that the agents can use the preset quick answer sentences to quickly reply to the user's questions during the actual business process.
  • the server of the manual customer service system will pre-set
  • the corresponding relationship between the set quick answer statement and the user intent label is stored in the corresponding database, so that after obtaining the target question submitted by the user to the manual customer service system and the target quick answer statement corresponding to the target question, based on The above corresponding relationship determines the target intent tag corresponding to the target quick response sentence.
  • the target question submitted by the user to the artificial customer service system and the target quick response sentence corresponding to the target question can be obtained, that is, the relationship between the target question question and the target quick answer sentence can be obtained. Therefore, based on the preset corresponding relationship between the target quick response sentence and the target intent label, the target intent tag corresponding to the target question question can be determined, so that the target question question can be processed based on the target intent label. Labeling, that is, labeling the intention of the target question, and then generating user intention recognition samples, and after labeling a large number of target questions, generating a user intention recognition sample set.
  • the above-mentioned user intent recognition sample set since the above-mentioned user intent recognition sample set is generated based on the real dialogue data of the artificial customer service system, the above-mentioned user intent recognition sample set includes user problems that cannot be obtained by the intelligent customer service system, and thus cannot be labeled, for example, intelligent User questions that cannot be answered accurately by the customer service system and are instead answered by the manual customer service system, as well as user questions that are directly answered by the manual customer service but cannot be identified by the smart customer service, are obtained from the manual customer service system and marked. Generating a user intent recognition sample set can further improve the coverage of user problems that cannot be identified by the intelligent customer service system included in the user intent recognition sample set.
  • the model is trained, that is, the above-mentioned user intention recognition samples are input into the user intention recognition model used by the intelligent customer service system, so as to use machine learning methods to iteratively train the model parameters in the user intention recognition model, and obtain the user intention recognition after the model parameters are updated model, wherein the model parameters of the updated user intention recognition model are the model parameters determined when the objective function corresponding to the user intention recognition model converges; further, in the intelligent customer service system, the updated user intention recognition model is used to conduct Recognition can improve the recognition accuracy of the user intention recognition model, thereby improving the service quality of the intelligent customer service system.
  • the customer service scenario above can be a business consulting scenario or a collection scenario.
  • the agents need to determine the questions that the user may ask in the collection scenario based on business experience, and the user’s suggestions.
  • the real intention of the question for example, the agent determines that the user may raise a question about the situation in which the payment has been deducted in the collection scenario, but the collection text message is still received, that is, the question raised by the user may be "the payment has been deducted, why is there still a prompt I want to repay the money", so it is determined that the real intention of the user to ask this question is to inquire about the situation of "the money has been deducted”.
  • the agent encounters a user who inquires about the situation of "deducted money", for example, the target question submitted by the user to the manual customer service system is obtained as "All the money has been deducted. Why do you still remind me to repay the money?"
  • the target question submitted by the user to the manual customer service system is "The payment has been deducted, why did I receive a collection SMS?”
  • the agent can use the pre-set "Verified to The deduction has been successful, I am very sorry for causing trouble to you" to make a quick reply, that is, upload the target quick response statement to the server of the manual customer service system through the agent client client cause trouble", so that the server of the artificial customer service system sends the target quick response sentence to the corresponding user terminal; correspondingly, the background server used for user question markup obtains the user's target question and the target question
  • the corresponding target quick response sentence is "It has been verified that the deduction has been successful, and I am very sorry to cause you trouble", and the target intent tag
  • the user intention recognition model in the system is to make full use of the dialogue data generated in the artificial customer service system to automatically generate the user intention training sample set required in the intelligent customer service system, and combine the real business flow of the artificial customer service system with the user intention of the intelligent customer service system.
  • the identification needs are closely related, which can improve the efficiency of labeling questions asked by users, thereby providing a large number of user intention recognition samples for the intelligent customer service system; moreover, the target questions submitted by users to the artificial customer service system include not only the problems that the intelligent customer service system cannot accurately answer
  • the user questions answered by the manual customer service system also include user questions that are directly answered by the manual customer service but cannot be identified by the intelligent customer service.
  • a sample set of user intention identification is obtained, which improves the quality of the user experience.
  • the coverage of user problems that cannot be identified by the intelligent customer service system included in the intent recognition sample set thereby improving the recognition accuracy of the user intent recognition model and ensuring the service quality of the intelligent customer service system.
  • the questions submitted by the user to the manual customer service system are likely to be questions that the intelligent customer service system cannot answer.
  • Dialogue data with agents that is, to obtain the user’s target question and the target quick answer sentence that the agent replies to the target question, based on the target quick answer sentence and target intent label pre-stored in the manual customer service system
  • the above S102 obtains the target questions submitted to the manual customer service system, and obtains the target questions
  • the target quick answer statement corresponding to the question including:
  • the method further includes:
  • the method of marking the user's target questions in real time can be used. Questions for each target, and obtain the target quick answer sentences that the human customer service replies to each target question, and based on the correspondence between the target quick answer sentences and the target intent labels pre-stored in the manual customer service system (that is, the pre-stored The corresponding relationship between the quick answer sentence and the intent tag), determine the target intent tag corresponding to each target quick answer sentence, and based on the target intent tag, perform user intent labeling on each target question raised by the user in real time.
  • the dialog data set generated by the manual customer service system within a preset time period, and then mark the user intent on multiple target questions in the dialog data set.
  • the preset time interval to obtain the dialog data set generated by the manual customer service system within the preset time period, from the dialog data set, extract multiple target question questions and target quick answer sentences corresponding to each target question question, and based on The corresponding relationship between the target quick response sentences and the target intent labels pre-stored in the artificial customer service system is used to mark the user intent for multiple target questions in the dialogue data set.
  • the dialogue data set Questions are asked, so before labeling user intentions for multiple target questions, similar target questions can be deduplicated in advance, so as to avoid repeated labeling of similar target questions and improve labeling efficiency.
  • the manual customer service when obtaining the dialogue data set generated by the manual customer service system, for the situation where the manual customer service directly uses the preset quick answer sentences to reply to the user's questions, based on the correspondence between the pre-stored quick answer sentences and the intent labels, Label the questions asked by users so that in the process of labeling user intentions on the questions asked by users, the coverage rate of labeling user questions that cannot be identified by the intelligent customer service system in the user intention identification sample is improved. Further, considering that there may be manual customer service in the When replying to the user's questions, the preset quick answer sentences are not directly used, but the regular answer sentences are used to reply to the user's questions (that is, the answer sentences are entered manually).
  • Similarity matching is performed on the quick answer sentences, and the quick answer sentences whose similarity meets the preset conditions are determined as the target quick answer sentences corresponding to the target questions. Automatically mark the user's intent on the question asked by the user, thereby further improving the coverage of the annotation of the user's problem that the intelligent customer service system cannot recognize in the user intent recognition sample.
  • the intelligent customer service system is precisely by closely linking the real business flow of the artificial customer service system with the user intention recognition requirements of the intelligent customer service system, that is, labeling the user intentions of multiple target questions in the dialogue data set to generate user intentions. After identifying the sample set, it is necessary to transmit the user intent identification sample set to the intelligent customer service system to increase the labeling coverage of user problems that the intelligent customer service system cannot identify.
  • the execution subject of the above-mentioned user problem labeling method is the manual customer service
  • a data transmission channel can be set between the manual customer service system and the intelligent customer service system, so as to automatically record the dialogue data between users and agents generated by real business traffic on the manual customer service system side (that is, including user Questions and quick answer sentences), and based on the correspondence between the preset quick answer sentences and user intent labels, user intentions are marked on the questions asked by users to generate user intent identification samples, and then the manual customer service system passes the preset
  • the data transmission channel automatically transmits the user intention identification sample to the intelligent customer service system
  • the execution subject of the labeling method for the above user problems is the second server connected to the first server of the manual customer service system, and the second The server obtains from the first server of the manual customer service system the dialog data set generated within the preset time period according to the preset time interval, and obtains the corresponding relationship between the preset quick answer sentence and the user intention label; from the dialog From the data set, multiple target
  • User intent recognition model which can improve the update timeliness of the user intent recognition sample set, thereby improving the training timeliness of the user intent recognition model in the intelligent customer service system, thereby improving the accuracy of the user intent recognition model for identifying the intent label of the user's question, Further improve the service quality of the intelligent customer service system.
  • the preset shortcut statement set and the user intention identification sample set can also be sent to the sample verification terminal in advance, so that the operator can check the shortcuts in the preset shortcut statement set on the page of the sample verification terminal. Verify the response statement and the user intent recognition sample set, for example, query, review, and modify the shortcut response statement and user intent recognition sample set in the preset shortcut statement set, and then based on the quick response that passes the verification The statement updates the shortcut statement set, and transmits the verified user intent recognition sample set to the intelligent customer service system.
  • the manual customer service and the intelligent customer service are activated at the same time, that is, the application scenario where the user switches between the intelligent customer service and the manual customer service during the process of asking questions
  • the intelligent customer service is If the question asked by the user cannot be identified during the process, the user will trigger the stage of manual customer service. Therefore, the user questions received by the manual customer service system will inevitably include a large number of user questions that the intelligent customer service system cannot identify.
  • the manual customer service system will The real business traffic is closely related to the user intent recognition requirements of the intelligent customer service system, which can not only automatically generate a large number of user intent recognition samples, but also improve the labeling coverage of user problems that the intelligent customer service system cannot identify.
  • the dialogue data generated by the manual customer service system based on real business traffic and the corresponding relationship between the pre-stored quick response sentence and the intent label (that is, the first corresponding relationship above), perform intent labeling on the user question, obtain the user intent recognition sample set, and then transmit the user intent recognition sample set to the customer service system
  • the above S1022 is to obtain the dialogue data set generated by the manual customer service system within a preset time period, specifically including:
  • Step 1 storing the questions submitted by the user to the manual customer service system in the second database of the manual customer service system;
  • Step 2 storing the quick response sentence replied by the target agent client to the user's question in the second database of the manual customer service system; wherein, the target agent client is the agent client assigned to answer the user's question;
  • Step 3 from the second database, based on the second corresponding relationship between the questions asked by the user and the quick answer sentences, the dialogue data set generated by the manual customer service system within a preset time period is obtained.
  • the above-mentioned second database can store the second corresponding relationship between the user question submitted by the user to the manual customer service system and the quick answer sentence replied by the target agent client, and the quick answer set by the agent in advance through the agent client
  • the first corresponding relationship between the statement and the user's intent label further, considering that the user's questions submitted by the user to the manual customer service system may have repetitive or irregular problems, therefore, not all the data stored in the second database are directly All user questions are regarded as target user questions, but after the dialogue data set is obtained, the user questions are preprocessed first to obtain multiple target question questions.
  • the preprocessing includes at least one of deduplication processing and standardization processing, Then, based on the above-mentioned second correspondence, determine the target quick response sentence corresponding to each target question; then, based on the above-mentioned first correspondence, determine the target intention label corresponding to each target question, and then based on the target intention label to the target question Perform intent labeling to obtain multiple user intent recognition samples, wherein each user intent recognition sample includes a third correspondence between the target question and the target intent label.
  • the above-mentioned second corresponding relationship may be based on the first message sequence number of the user's question submitted by the user to the manual customer service system, and the second message of the quick answer sentence replied by the target agent (that is, the manual customer service) to the user's question.
  • the sequence number is determined, specifically, the time of the first message sequence number is earlier than the time of the second message sequence number.
  • the target agent can select a shortcut answer sentence that matches the user's question from the preset shortcut statement set to reply, wherein, the above
  • the preset shortcut statement set may include a personal shortcut statement set and a shared shortcut statement set.
  • the candidate quick answer sentence is based on the answer sentence manually input by the target agent in the information input box, and automatically matched from the preset shortcut sentence set Response statement, in specific implementation, can first match in the personal shortcut statement set and the personal quick answer statement that the similarity between the response statement manually input by the target agent meets the preset condition, if there is no match from the personal shortcut statement set If an available quick answer statement is found, matching is performed from the shared quick answer statement set to determine the available shared quick answer statement.
  • the user questions submitted by the user to the manual customer service system, and the quick answer sentences replied by the target agent to the user's questions are stored in the second database of the manual customer service system, so as to extract the information to be marked later.
  • the target question and the corresponding target quick answer statement and then realize the user intention labeling of the target question, and obtain multiple user intention recognition samples.
  • the agent can set the user intention label corresponding to each quick answer sentence at the same time, so that the agent can use the quick answer sentence to quickly reply to the user's question in the subsequent process, so as to quickly respond to the user's questions.
  • the agent can set the user intention label corresponding to each quick answer sentence at the same time, so that the agent can use the quick answer sentence to quickly reply to the user's question in the subsequent process, so as to quickly respond to the user's questions.
  • each agent can determine the questions that the user may ask and the real intention of the user to ask the question based on business experience, and then set the corresponding quick answer statement through the agent client , and set a corresponding user intent label for each quick answer sentence, and pre-store it in the second database of the manual customer service system through the agent client, so that after obtaining the target questions submitted by the user to the manual customer service system, quickly respond to the Users ask questions to mark user intent.
  • the first agent client receives the first quick answer sentence input by the first agent, and receives the first intention label set by the first agent for the first quick answer sentence, and then, the first agent client The terminal uploads the first quick answer statement and the first intent label to the server of the manual customer service system, and the server of the manual customer service system stores the first corresponding relationship between the first quick answer statement and the first intention label in the manual customer service system in the second database.
  • receiving the first intention label corresponding to the first quick response sentence uploaded by the first agent client specifically includes:
  • Step 1 determining at least one candidate intent tag corresponding to the first quick response sentence
  • the above-mentioned candidate intent tags may be determined based on user intent tags corresponding to shared shortcut sentences or personal shortcut sentences in the preset shortcut sentence set. Specifically, the shortcut response sentences currently set by the first agent and the preset Similarity matching is performed on the shared shortcut sentences or personal shortcut sentences in the shortcut sentence set, and the intent tags corresponding to the shortcut sentences whose similarity meets the preset conditions are determined as candidate intent tags.
  • Step 2 receiving a first intent tag uploaded by the first agent client, wherein the first intent tag is a user intent tag in at least one candidate intent tag.
  • the above-mentioned at least one candidate intent tag is sent to the first agent client, and the first agent client displays the at least one candidate intent tag, and prompts the first agent to set the first quick response statement based on the at least one candidate intent tag.
  • the first intention label is to receive the user intention label selected by the first agent in at least one candidate intention label as the first intention label, and then upload the first intention label to the service end of the manual customer service system, so that the manual customer service system
  • the server stores the corresponding relationship between the first quick response sentence and the first intent label.
  • the above-mentioned shortcut statement set can include: Corresponding to the personal shortcut statement set and the shared shortcut statement set, the above S112 stores the correspondence between the first quick response statement and the first intent label, specifically including:
  • the first quick response statement is added, and the corresponding relationship between the first quick response statement and the first intent label is stored.
  • the first quick response statement and the corresponding first intent label set by the first agent through the first agent client can be directly stored in the corresponding personal shortcut statement set, and at the same time, the first agent can also request the The first quick answer sentence and the corresponding first intent label are stored in the shared shortcut sentence set, so that the first quick answer sentence can be used by other agents.
  • the first agent client can be distinguished according to the login account of the agent, specifically, the individual corresponding to the manual customer service account logged in on the first agent client In the shortcut statement set, the first quick response statement is added, and the corresponding relationship between the first quick response statement and the first intent label is stored.
  • the function of setting personal shortcuts as shared shortcuts can be provided for agents, so as to improve the selectivity of agents for quick response sentences, and thus improve the target questioning rate.
  • the intention of the question is marked comprehensively.
  • the set of shared shortcut sentences is updated.
  • the above-mentioned personal shortcut statement set may be corresponding to the login account of the agent. After the agent uses a certain agent client to log in to their account, they can enter the viewing page of the corresponding personal shortcut statement set; the above-mentioned shared shortcut statement set It can be corresponding to a login account group, which can include multiple login accounts under a certain customer service service category, and the agent uses a certain agent client to log in to any login account in the login account group After having an account, you can not only enter the view page of the personal shortcut statement set corresponding to the login account, but also enter the view page of the corresponding shared shortcut statement set.
  • the update process of the shared shortcut statement set it is possible to directly add the first quick response statement that the first agent client request is converted into the shared shortcut statement in the shared shortcut statement set, and store the first quick response statement and the first Correspondence between intent tags; further, in order to ensure the accuracy and standardization of the quick response sentences and user intent tags contained in the shared shortcut sentence set, it may also be first determined whether there is a relationship with the first agent in the shared shortcut sentence set
  • the client request is transformed into the second quick response sentence that matches the first quick response sentence of the shared shortcut; if it exists, then according to the similarity between the first intent tag and the second intent tag corresponding to the second quick response sentence , to return the corresponding feedback information to the first agent client, which can also achieve the effect of verifying the first intent tag in the personal shortcut statement set; if it does not exist, it will be converted into a shared shortcut based on the first agent client request
  • the first quick response sentence of the language and the corresponding first intent label are used to update the shared shortcut sentence set.
  • the agents log in to their respective login account aa through the agent client, and can enter the shortcut language setting interface.
  • the shortcut language setting interface they can view the set personal shortcut response sentences and the shared information corresponding to the agent group they belong to.
  • Quick answer statement and quickly search for the corresponding quick answer statement through the information input box of the search keyword; the agent can add a new quick answer statement and its corresponding user intention label by clicking the first control below; and by clicking the second
  • the control requests that a personal shortcut answer be set as a shared shortcut.
  • Step 1 judging whether there is a second quick response statement matching the first quick response statement in the shared shortcut statement set;
  • Step 2 if the result of the judgment is that it does not exist, then add the first quick response statement to the shared shortcut statement set; specifically, it is also possible to preprocess the first quick response statement first, and then add the preprocessed first quick response statement
  • the answer sentence is added to the set of shared shortcut sentences, and the preprocessing may include at least one of error correction processing and standardization processing;
  • Step 3 based on the first intent tag corresponding to the first quick response sentence, determine the real intent tag of the first quick response sentence; wherein, the real intent tag is based on the scoring result of the first intent tag uploaded by the second agent client Definite; specifically, the second agent client may include: at least one of the agent client of each login account in the login account group corresponding to the login sharing shortcut statement set, and the designated agent client. It is an agent client used by professionals;
  • Step 4 In the shared shortcut statement set, the corresponding relationship between the first quick response statement and the real intent label is stored.
  • the statement similarity between at least one quick response statement in the shared shortcut statement set and the first quick response statement is greater than the preset threshold, it is determined that there is a second quick response statement matching the first quick response statement; the corresponding If the sentence similarity between each quick response sentence in the shared shortcut sentence set and the first quick reply sentence is less than or equal to the preset threshold, it is determined that there is no second quick reply sentence matching the first quick reply sentence.
  • the real intention label can also be sent to the first agent client of the first agent, so that the first agent judges whether it is based on the real intention label, in the corresponding personal shortcut statement set, update the corresponding relationship between the first quick response statement and the first intent label.
  • the corresponding relationship between the first quick response sentence and the first intent tag can be sent to at least one second agent respectively
  • the second agent client wherein, the second agent can be each agent in the target agent group, wherein the first agent and the second agent belong to the target agent group; then, receive the second agent client Returned scoring results for the first intent tag, and based on the scoring results of multiple second agents for the first intent tag, determine whether the first intent tag corresponding to the first quick response statement satisfies the preset constraint conditions, for example, If the comprehensive score is greater than the preset score threshold, it is determined that the preset constraint condition is satisfied; if the comprehensive score is less than or equal to the preset score threshold, it is determined that the preset constraint condition is not satisfied.
  • the server determines the comprehensive score of the first intent tag based on the scoring results of multiple second agent clients, wherein the comprehensive score can be It is obtained by weighting the scores corresponding to the scoring results of multiple second agent clients; if the comprehensive score is greater than the preset score threshold, then determine that the first intent tag is the real intent tag of the first quick response statement; if If the comprehensive score is less than or equal to the preset score threshold, it will trigger the designated agent client to return the real intent label of the first quick response statement; then, store the difference between the first quick response statement and the real intention label
  • the designated agent client can be an agent client used by professionals.
  • the agent client has professional knowledge in a specific field. For questions in this field, it can more accurately determine the true intentions and corresponding truths of various answer sentences. intent label.
  • the first agent client and the second agent client may be different agent clients or the same agent client, that is, When different agents use the same agent client at different time nodes, at this time, the login account of the first agent client and the login account of the second agent client may be different.
  • the first agent client so that the first agent judges whether based on the real intention label, in the corresponding personal shortcut statement set, update the corresponding relationship between the first quick response statement and the first intention label, which can not only improve
  • the accuracy and standardization of the user intention label of the shared shortcut statement can also be improved, and at the same time, the accuracy and standardization of the user intention label of the personal shortcut statement can be improved.
  • step 1 after judging whether there is a second quick response statement matching the first quick response statement in the shared shortcut statement set, it also includes:
  • the similarity comparison between the first intent tag and the second intent tag can be performed first, and corresponding feedback information is returned to the first agent client according to the similarity comparison result, and in When the similarity between the first intent tag and the second intent tag is relatively low, there may be an inaccurate setting of the first intent tag.
  • the second intent tag can be returned to the first agent client, so that the first agent Personnel can decide whether to modify the first intent label of the first quick response statement in the personal shortcut statement set based on the second intent label, so as to improve the accuracy and specification of the user's intention label configured for the quick response statement in the personal shortcut statement set In this way, the accuracy of intention labeling of user questions can be improved.
  • the shortcut statement set includes a personal shortcut statement set and a shared shortcut statement set
  • determine the target corresponding to the target shortcut response statement Before the intent tag also include:
  • the target quick response statement belongs to the shared shortcut statement set, the first corresponding relationship is queried in the shared shortcut statement set.
  • the source information includes: the identification of the personal shortcut statement set, or the identification of the shared shortcut statement set, so as to quickly enter the personal shortcut statement set or the shared shortcut statement set , find the first corresponding relationship recorded for the target quick answer sentence, and then quickly determine the target intent label corresponding to the target question.
  • the above-mentioned target question is marked, and after the user intention recognition sample set is generated, it also includes:
  • the intent recognition is performed on the user questions received by the intelligent customer service system, and the corresponding user intent recognition results are output.
  • the user intention recognition model inputs the user intention recognition sample set based on the dialogue data generated by the artificial customer service system into the user intention recognition model to be trained, use machine learning methods and based on the above user intention recognition sample set, the user intention recognition model
  • the model parameters are iteratively trained to obtain the trained user intent recognition model; wherein, the model parameters of the trained user intent recognition model are the model parameters determined when the target function corresponding to the user intent recognition model converges; further, in the intelligent customer service system
  • the trained user intention recognition model is used to identify the intention of the user's questions received by the intelligent customer service system, and output the corresponding user intention recognition results, which can improve the recognition accuracy and improve the service quality of the intelligent customer service system.
  • the dialogue data set generated by the manual customer service system within the preset time period is obtained, and the dialogue data set is extracted from the dialogue data set.
  • the target question 1 to the target question n submitted by the user to the artificial customer service system, and the target quick answer sentences corresponding to the target question 1 to the target question n respectively ie, the target quick answer sentence 1 to the target quick answer sentence n
  • Based on the correspondence between the pre-stored target quick response sentences and target intent tags in the manual customer service system determine the target intent tags corresponding to the target question 1 to the target question n (ie, target intent tag 1 to target intent tag n)
  • the target question 1 to the target question n are marked, and the user intention recognition sample set is generated (for example, the target question 1-target intention label 1 to the target question n-target Intent label n), further, input the above-mentioned user intention
  • the user intention training sample set required in the intelligent customer service system is automatically generated, and the real business flow of the artificial customer service system is combined with the user intention recognition requirements of the intelligent customer service system.
  • Carrying out close correlation can improve the labeling efficiency of user questions, thereby providing a large number of user intention recognition samples for the intelligent customer service system; and, the target questions submitted by users to the manual customer service system include not only the intelligent customer service system cannot accurately answer but turn to
  • the user questions answered by the manual customer service system also include user questions that are directly answered by the manual customer service but cannot be recognized by the intelligent customer service.
  • a user intention recognition sample set is obtained, which improves user intention recognition.
  • the coverage of user problems that cannot be identified by the intelligent customer service system contained in the sample set can improve the recognition accuracy of the user intention recognition model, thereby ensuring the service quality of the intelligent customer service system.
  • the specific application scenario of the method for labeling user questions is shown in Figure 6.
  • the manual customer service system Based on business experience, the agents determine the questions that users may ask in a specific business scenario, as well as the real intentions of users to ask questions, and the agent backend of the artificial customer service system receives the agent's pre-recorded information on the agent front-end page of the artificial customer service system.
  • the manual customer service system will automatically assign agents to the user;
  • the connection channel between the user and the agent is set in advance through the first module.
  • the first module is determined based on the port that the user accesses the manual customer service system.
  • the port of the manual customer service system may include WeChat applet port, For any of the webpage port and application port, if the user accesses the manual customer service system through the WeChat applet port (that is, the first module is the WeChat applet port), the user will be assigned the corresponding agent of the WeChat applet, and through the long-term
  • the connection module establishes the connection between the agent and the user, so that when there are many users, the agent can maintain the conversation with the user through the long connection module, and then can switch between different users;
  • the questions asked by the user and the response sentences replied by the agent are stored in the second database through the long connection module and the second module in turn.
  • the user and the agent When conducting a dialogue between them, the user sends a question message to the artificial customer service system through the front end of the agent, the back end of the agent and the long connection module (that is, the user asks a question); Send a reply message (that is, a customer service answer statement, which can be a conventional answer statement or a quick answer statement); wherein, the long connection module is used to check whether the format of the question message and the reply message is correct and deliver the message to The second module (comprising the kafka module and the message module in turn), the second module is used to preprocess the message (such as saving, recording the message sequence number, buffering, etc.), and then storing it in the second database;
  • the data processing module obtains from the second database the dialog data sets generated by the manual customer service system within a preset time period, and extracts from the dialog data sets multiple target questions submitted by the user to the manual customer service system, And based on the first message sequence number of the question question and the second message sequence number of the quick answer statement, determine the target quick answer statement that the agent replies to each target question question, and then based on the pre-stored in the second database of the manual customer service system The corresponding relationship between the quick answer sentence and the intention label, determine the intention label corresponding to the target quick answer statement answered by the agent for each target question, and then determine the corresponding relationship between the target question question and the intention label, and then Realize the user intention labeling of the target question, and generate the user intention recognition sample set.
  • the above-mentioned first message serial number based on the question question and the second message serial number of the quick answer sentence, determining the target quick answer sentence that the agent replies to each target question question includes: according to the keywords in the quick answer sentence and at least The matching degree of keywords in a question determines the specific question corresponding to the quick answer sentence; or, according to the matching degree between the quick answer sentence and at least one question, determine one or more specific question questions corresponding to the quick answer sentence .
  • the agent After determining one or more specific question questions corresponding to the quick answer sentence, based on the correspondence between the quick answer sentence and the intention label pre-stored in the second database of the manual customer service system, it is determined that the agent asks questions for each target The intent label corresponding to the target quick answer sentence answered by the question, and then determine the corresponding relationship between the target question and the intent label, and then realize the user intention labeling of the target question, and generate a user intention recognition sample set.
  • a long connection is established between the artificial customer service system and the intelligent customer service system in advance, and the above-mentioned user intention recognition sample set generated in the artificial customer service system is transmitted to the intelligent customer service system through the channel of the long connection, so that the intelligent customer service system will send the user
  • the intent recognition sample set is stored in the first database, and the user intent recognition model is trained using a machine learning method based on the above user intent recognition sample set to obtain a trained user intent recognition model, and using the trained user intent recognition model, Carry out intent recognition on user questions received by the intelligent customer service system, and output corresponding user intent recognition results.
  • the method for labeling user questions in the embodiment of the present application by obtaining the dialog data generated by the manual customer service system, first determine the target quick answer sentence corresponding to the target question question, and then combine the preset quick answer sentence with the user intention label Correspondence, determine the target intent label corresponding to the target quick answer sentence, and then get the corresponding relationship between the target question question and the target intent label, so as to realize the user intent labeling of the user question, and then apply the generated user intent recognition sample set to
  • the user intention recognition model in the intelligent customer service system is to make full use of the dialogue data generated in the artificial customer service system to automatically generate the user intention training sample set required in the intelligent customer service system, and combine the real business flow of the artificial customer service system with the intelligent customer service flow.
  • the user intention recognition requirements of the system are closely related, which can improve the efficiency of labeling questions asked by users, thereby providing a large number of user intention recognition samples for the intelligent customer service system; moreover, the target questions submitted by users to the artificial customer service system include not only intelligent customer service User questions that the system cannot answer accurately and are then answered by the manual customer service system also include user questions that are directly answered by the manual customer service but cannot be recognized by the intelligent customer service.
  • the user intention identification sample set is obtained by automatically marking these target questions. , which improves the coverage of user problems that cannot be identified by the intelligent customer service system contained in the user intention recognition sample set, thereby improving the recognition accuracy of the user intention recognition model and ensuring the service quality of the intelligent customer service system.
  • the embodiment of the present application also provides a labeling device for user questions
  • Figure 7 is the labeling device for user questions provided by the embodiment of the present application
  • a schematic diagram of the module composition, the device is used to implement the labeling method for user questions described in Figures 1 to 6, as shown in Figure 7, the device includes:
  • the first acquiring module 702 is used to acquire the target question submitted to the manual customer service system, and acquire the target quick answer sentence corresponding to the target question question; wherein, the target quick answer sentence includes: in the preset shortcut sentence set At least one quick answer statement, the quick answer statement is a response statement preset in the manual customer service system for quick reply to user questions;
  • the first determining module 704 is configured to determine the target intent label corresponding to the target quick response sentence based on the preset first correspondence; wherein the first correspondence includes the pre-stored in the manual customer service system The corresponding relationship between the target quick response sentence and the target intent label;
  • the first generating module 706 is configured to mark the target question based on the target intention label, and generate a user intention identification sample set; wherein, the user intention identification sample set is used to identify users used by the intelligent customer service system
  • the intent recognition model is trained.
  • the user problem labeling device in the embodiment of the present application first determines the target quick answer sentence corresponding to the target question by acquiring the dialogue data generated by the manual customer service system, and then combines the preset quick answer sentence with the user intention label. Correspondence, determine the target intent label corresponding to the target quick answer sentence, and then get the corresponding relationship between the target question question and the target intent label, so as to realize the user intent labeling of the user question, and then apply the generated user intent recognition sample set to
  • the user intention recognition model in the intelligent customer service system is to make full use of the dialogue data generated in the artificial customer service system to automatically generate the user intention training sample set required in the intelligent customer service system, and combine the real business flow of the artificial customer service system with the intelligent customer service flow.
  • the user intention recognition requirements of the system are closely related, which can improve the efficiency of labeling questions asked by users, thereby providing a large number of user intention recognition samples for the intelligent customer service system; moreover, the target questions submitted by users to the artificial customer service system include not only intelligent customer service User questions that the system cannot answer accurately and are then answered by the manual customer service system also include user questions that are directly answered by the manual customer service but cannot be recognized by the intelligent customer service.
  • the user intention identification sample set is obtained by automatically marking these target questions. , which improves the coverage of user problems that cannot be identified by the intelligent customer service system contained in the user intention recognition sample set, thereby improving the recognition accuracy of the user intention recognition model and ensuring the service quality of the intelligent customer service system.
  • an embodiment of the present application also provides a user question labeling device, which is used to perform the above user question labeling method, as shown in Figure 8 shows.
  • the devices for marking user questions may vary greatly due to different configurations or performances, and may include one or more processors 801 and memory 802, and one or more application programs or data may be stored in the memory 802.
  • the storage 802 may be a short-term storage or a persistent storage.
  • the application program stored in the memory 802 may include one or more modules (not shown), and each module may include a series of computer-executable instructions.
  • the processor 801 may be configured to communicate with the memory 802, so as to execute a series of computer-executable instructions in the memory 802 on the user question labeling device.
  • the labeling device for user questions may also include one or more power sources 803, one or more wired or wireless network interfaces 804, one or more input and output interfaces 805, one or more keyboards 806, and the like.
  • the labeling device for user questions includes a memory, wherein one or more programs are stored in the memory, and one or more programs may include one or more modules, and each module may include a series of Computer-executable instructions configured to be executed by one or more processors; when the computer-executable instructions included in the one or more programs are executed by the processors, they are used to implement the following processes:
  • the target quick answer statement includes: at least one quick answer statement in the preset quick statement set, the The quick answer sentence is a reply sentence for quick reply to the user's question preset in the manual customer service system;
  • the target intent label corresponding to the target quick response statement Based on the preset first correspondence, determine the target intent label corresponding to the target quick response statement; wherein, the first correspondence includes the target quick response statement and the target Correspondence between intent tags;
  • the user question labeling device in the embodiment of the present application first determines the target quick answer sentence corresponding to the target question by acquiring the dialogue data generated by the manual customer service system, and then combines the preset quick answer sentence with the user intention label. Correspondence, determine the target intent label corresponding to the target quick answer sentence, and then get the corresponding relationship between the target question question and the target intent label, so as to realize the user intent labeling of the user question, and then apply the generated user intent recognition sample set to
  • the user intention recognition model in the intelligent customer service system is to make full use of the dialogue data generated in the artificial customer service system to automatically generate the user intention training sample set required in the intelligent customer service system, and combine the real business flow of the artificial customer service system with the intelligent customer service flow.
  • the user intention recognition requirements of the system are closely related, which can improve the efficiency of labeling questions asked by users, thereby providing a large number of user intention recognition samples for the intelligent customer service system; moreover, the target questions submitted by users to the artificial customer service system include not only intelligent customer service User questions that the system cannot answer accurately and are then answered by the manual customer service system also include user questions that are directly answered by the manual customer service but cannot be recognized by the intelligent customer service.
  • the user intention identification sample set is obtained by automatically marking these target questions. , which improves the coverage of user problems that cannot be identified by the intelligent customer service system contained in the user intention recognition sample set, thereby improving the recognition accuracy of the user intention recognition model and ensuring the service quality of the intelligent customer service system.
  • this embodiment of the present application also provides a storage medium for storing computer-executable instructions.
  • the The storage medium can be a USB flash drive, an optical disc, a hard disk, etc., and the computer-executable instructions stored in the storage medium can realize the following process when executed by the processor:
  • the target quick answer statement includes: at least one quick answer statement in the preset quick statement set, the The quick answer sentence is a reply sentence for quick reply to the user's question preset in the manual customer service system;
  • the target intent label corresponding to the target quick response statement Based on the preset first correspondence, determine the target intent label corresponding to the target quick response statement; wherein, the first correspondence includes the target quick response statement and the target Correspondence between intent tags;
  • the processor executes by obtaining the dialog data generated by the manual customer service system, first determine the target shortcut response sentence corresponding to the target question, and then combine the preset shortcut The corresponding relationship between the answer sentence and the user intention label, determine the target intention label corresponding to the target quick answer statement, and then get the corresponding relationship between the target question question and the target intention label, so as to realize the user intention labeling of the user question, and then Apply the generated user intent recognition sample set to the user intent recognition model in the intelligent customer service system, that is, fully utilize the dialogue data generated in the manual customer service system to automatically generate the user intent training sample set required in the intelligent customer service system, and artificially
  • the real business flow of the customer service system is closely related to the user intention recognition requirements of the intelligent customer service system, which can improve the efficiency of labeling questions asked by users, thereby providing a large number of user intention identification samples for the intelligent customer service system;
  • the submitted target questions include not only user questions that the intelligent customer service system cannot answer accurately and
  • the user intention recognition sample set is obtained, which improves the coverage of user problems that cannot be identified by the intelligent customer service system contained in the user intention recognition sample set, thereby improving the recognition accuracy of the user intention recognition model and ensuring the service of the intelligent customer service system quality.
  • a Programmable Logic Device such as a Field Programmable Gate Array (FPGA)
  • FPGA Field Programmable Gate Array
  • HDL Hardware Description Language
  • the controller may be implemented in any suitable way, for example the controller may take the form of a microprocessor or processor and a computer readable medium storing computer readable program code (such as software or firmware) executable by the (micro)processor , logic gates, switches, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable logic controllers and embedded microcontrollers, examples of controllers include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the controller can also be implemented as part of the control logic of the memory.
  • controller in addition to realizing the controller in a purely computer-readable program code mode, it is entirely possible to make the controller use logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded The same function can be realized in the form of a microcontroller or the like. Therefore, such a controller can be regarded as a hardware component, and the devices included in it for realizing various functions can also be regarded as structures within the hardware component. Or even, means for realizing various functions can be regarded as a structure within both a software module realizing a method and a hardware component.
  • a typical implementing device is a computer.
  • the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or Combinations of any of these devices.
  • the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on a computer-usable storage medium (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • a computer-usable storage medium including but not limited to disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions
  • the means realize the function specified in one or more procedures of the flowcharts and/or one or more blocks of the block diagrams.
  • a computing device includes a processor (CPU), input/output interfaces, network interfaces, and memory.
  • Memory may include non-permanent storage in computer readable media, in the form of random access memory (RAM) and/or nonvolatile memory such as read-only memory (ROM) or flash RAM. Memory is an example of computer readable media.
  • RAM random access memory
  • ROM read-only memory
  • flash RAM flash random access memory
  • Computer-readable media includes both volatile and non-volatile, removable and non-removable media, and can be implemented by any method or technology for storage of information.
  • Information may be computer readable instructions, data structures, modules of a program, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Flash memory or other memory technology, Compact Disc Read-Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridge, tape disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
  • computer-readable media excludes transitory computer-readable media, such as modulated data signals and carrier waves.
  • the embodiments of the present application may be provided as methods, systems or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on a computer-usable storage medium (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • a computer-usable storage medium including but not limited to disk storage, CD-ROM, optical storage, etc.
  • program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types.
  • the application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network.
  • program modules may be located in both local and remote computer storage media including storage devices.
  • each embodiment in the present application is described in a progressive manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments.
  • the description is relatively simple, and for relevant parts, refer to part of the description of the method embodiment.

Abstract

本申请实施例提供了一种用户问题的标注方法及装置，通过获取人工客服系统产生的对话数据，确定目标提问问题对应的目标快捷应答语句，再结合预设的快捷应答语句与用户意图标签之间的对应关系，确定目标快捷应答语句对应的目标意图标签，即可得到目标提问问题与目标意图标签之间的对应关系，从而实现对用户问题进行用户意图标注，进而将生成的用户意图识别样本集应用于智能客服系统中的用户意图识别模型，不仅能够提高用户问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本，还能够提高用户意图识别样本集中包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。

Description

用户问题的标注方法及装置
本申请要求于2021年12月31日提交的申请号为202111678653.1、名称为“用户问题的标注方法及装置”的中国专利申请的优先权，上述申请的内容通过引用并入本文。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种用户问题的标注方法及装置。
背景技术
目前，随着机器学习技术的快速发展，机器学习在许多领域中有了十分广泛的应用。例如，机器学习技术可以应用于智能机器人领域，通过采集真实的业务场景数据训练智能机器人所使用的机器学习模型，使得智能机器人能够为用户提供优质的服务。
其中，针对客户服务场景，可以利用智能客服协助人工客服工作，从而解决人力资源紧张的问题，但是目前智能客服处理业务的能力有限，有时无法从用户的问题中识别出用户的真正意图，导致用户体验差，因此，需要提供一种能够提高智能客服所使用的用户意图识别模型的识别准确度的技术方案。
发明内容
本申请实施例的目的是提供一种用户问题的标注方法及装置。
第一方面，本申请实施例提供了一种用户问题的标注方法，所述方法包括：
获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句；其中，所述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，所述快捷应答语句为在所述人工客服系统中预先设置的用于快捷回复用户问题的应答语句；
基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签；其中，所述第一对应关系包括在所述人工客服系统中预存的所述目标快捷应答语句与所述目标意图标签之间的对应关系；
基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集；其中，所述用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练。
第二方面，本申请实施例提供了一种用户问题的标注装置，所述装置包括：
第一获取模块，用于获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句；其中，所述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，所述快捷应答语句为在所述人工客服系统中预先设置的用于快捷回复用户问题的应答语句；
第一确定模块，用于基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签；其中，所述第一对应关系包括在所述人工客服系统中预存的所述目标快捷应答语句与所述目标意图标签之间的对应关系；
第一生成模块，用于基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集；其中，所述用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练。
第三方面，本申请实施例提供了一种用户问题的标注设备，所述设备包括：
处理器；以及被安排成存储计算机可执行指令的存储器，所述可执行指令被配置由所述处理器执行，所述可执行指令包括用于执行如第一方面中所述的方法中的步骤。
第四方面，本申请实施例提供了一种存储介质，其中，所述存储介质用于存储计算机可执行指令，所述可执行指令使得计算机执行如第一方面中所述的方法中的步骤。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术的描述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对应于本申请中记载的一些实施例，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为本申请实施例提供的用户问题的标注方法的第一种流程示意图；
图2为本申请实施例提供的用户问题的标注方法的第二种流程示意图；
图3为本申请实施例提供的用户问题的标注方法的第三种流程示意图；
图4为本申请实施例提供的用户问题的标注方法中坐席客户端的快捷语设置界面示意图；
图5为本申请实施例提供的用户问题的标注方法的用户意图识别模型的训练过程示意图；
图6为本申请实施例提供的用户问题的标注方法的应用场景示意图；
图7为本申请实施例提供的用户问题的标注装置的模块组成示意图；
图8为本申请实施例提供的用户问题的标注设备的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本申请中的技术方案，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，所描述的实施例仅仅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，都应当属于本申请的保护范围。
需要说明的是，在不冲突的情况下，本申请中的实施例以及实施例中的特征可以相互组合。下面将参考附图来详细说明本申请实施例。
本申请实施例提供了一种用户问题的标注方法及装置，通过获取人工客服系统所产生的对话数据，先确定目标提问问题对应的目标快捷应答语句，再结合预设的快捷应答语句与用户意图标签之间的对应关系，确定目标快捷应答语句对应的目标意图标签，即可得到目标提问问题与目标意图标签之间的对应关系，从而实现对用户问题进行用户意图标注，进而将生成的用户意图识别样本集应用于智能客服系统中的用户意图识别模型，即充分借助人工客服系统中所产生的对话数据，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样不仅能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本，并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，还能够提高用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
需要说明的是，在客户服务的场景中，针对利用智能客服系统为用户提供服务的情况，需要预先采集大量的用户意图识别样本对智能客服系统所使用的用户意图识别模型进行训练，在现有的采集用于训练用户意图识别模型的用户意图识别样本的过程中，通常需要智能客服系统记录其无法准确应答的用户问题，再由人工对这些用户问题进行意图标注；并且考虑到有些情况下用户会直接选择人工客服系统，导致智能客服系统无法识别到这部分用户问题，从而导致人工对用户问题进行意图标注时，可能会错过一些智能客服系统无法准确应答的用户问题，因此会存在用户意图识别样本获取效率低、获取数量有限的问题，进而导致训练后的用户意图识别模型的识别准确度低的问题，使得智能客服系统在为用户提供服务时，无法准确地对用户提问问题的真正意图进行识别，造成用户体验感差的问题，基于上述问题，本申请技术方案通过充分借助人工客服系统中所产生的对话数据(即可以包括用户提问问题与客服应答语句之间的对应关系)，并结合预设的快捷应答语句与用户意图标签之间的对应关系，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样不仅能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本，并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，还能够提高用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
图1为本申请实施例提供的用户问题的标注方法的第一种流程示意图，图1中的方法是基于人工客服系统所产生的真实业务流量，自动从人工客服系统获取用户意图识别样本集的过程，该用户意图识别样本集用于训练智能客服系统所使用的用户意图识别模型，该标注方法应用于用户问题标注的后台服务端，该后台服务端可以是人工客服系统的第一服务端，也可以是与人工客服系统的第一服务端通信连接的第二服务端，如图1所示，该方法至少包括以下步骤：
S102，获取向人工客服系统提交的目标提问问题，以及获取目标提问问题对应的目标快捷应答语句；
其中，上述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，该快捷应答语句为在人工客服系统中预先设置的用于快捷回复用户问题的应答语句；上述目标快捷应答语句可以是目标坐席人员从预设的快捷语句集合中选择并直接回复的快捷应答语句，对应的，将目标坐席人 员针对用户的目标提问问题所回复的快捷应答语句确定为目标快捷应答语句；具体的，目标坐席人员可以在预设的多个快捷应答语句中选择与目标提问问题匹配的目标快捷应答语句，目标坐席人员所使用的目标坐席客户端在检测到针对目标快捷应答语句的选择输入操作后，向人工客服系统的服务端发送该目标快捷应答语句，对应的，服务端将该目标快捷应答语句发送至提交目标提问问题的客户端，从而实现一键回复用户的目标提问问题，坐席人员省去了手动输入应答语句的步骤；在具体实施时，目标坐席人员针对目标快捷应答语句的选择输入操作可以是通过鼠标对目标快捷应答语句的点击操作，还可以是在触控屏上对目标快捷应答语句的点击操作，也可以是快捷键的输入操作，即预先设置快捷键与快捷应答语句之间的对应关系，这样坐席人员通过坐席客户端输入相应的快捷键即可快速调出相应的快捷应答语句。
其中，上述目标快捷应答语句也可以是基于目标坐席人员针对用户的目标提问问题所回复的常规应答语句(即人工客服手动输入的应答语句)，自动从预设的快捷语句集合中匹配出来的快捷应答语句，具体的，可以将目标坐席人员回复的常规应答语句与预设的快捷语句集合中的各快捷应答语句进行匹配，基于相似度满足预设条件的快捷应答语句确定目标提问问题对应的目标快捷应答语句。
例如，上述预设条件可以为相似度大于预设阈值，对应的，从预设的快捷语句集合中匹配出至少一个快捷应答语句确定为推荐快捷应答语句，该至少一个快捷应答语句与目标坐席人员针对用户的目标提问问题所回复的常规应答语句之间的相似度大于预设阈值，并将目标坐席人员从上述推荐快捷应答语句中选择并进行回复的快捷应答语句确定为目标快捷应答语句；具体的，针对基于目标坐席人员所回复的常规应答语句(即人工客服手动输入的应答语句)，自动从预设的快捷语句集合中匹配目标快捷应答语句的过程，可以是目标坐席客户端在检测到目标坐席人员在信息输入框中输入常规应答语句的过程中(即人工输入的常规应答语句未发送至人工客服系统的服务端的情况下)，目标坐席客户端自动基于预存的快捷语句集合，确定与在信息输入框中输入的常规应答语句匹配的推荐快捷应答语句，并提示目标客服人员从推荐快捷应答语句中选择目标快捷应答语句；又如，上述预设条件还可以为相似度最大，对应的，服务端自动从预设的快捷语句集合中匹配出与常规应答语句的相似度最大的快捷应答语句确定为目标快捷应答语句。具体的，针对基于目标坐席人员所回复的常规应答语句(即人工客服手动输入的应答语句)，自动从预设的快捷语句集合中匹配目标快捷应答语句的过程，也可以是目标坐席客户端向人工客服系统的服务端上传目标坐席人员在信息输入框中输入的常规应答语句(即人工输入的常规应答语句已发送至人工客服系统的服务端的情况下)，由人工客服系统的服务端自动基于预存的快捷语句集合，确定与目标坐席客户端上传的常规应答语句匹配的目标快捷应答语句。
S104，基于预设的第一对应关系，确定目标快捷应答语句对应的目标意图标签；其中，第一对应关系包括在人工客服系统中预存的目标快捷应答语句与目标意图标签之间的对应关系；
其中，考虑到坐席人员可以根据业务经验确定用户可能会提出的问题、以及用户提出问题的真实意图，并通过坐席客户端向人工客服系统的服务端上传快捷应答语句，以及上传该快捷应答语句对应的用户意图标签，以使坐席人员在实际处理业务的过程中，可以利用预先设置的快捷应答语句快捷回复用户的问题。
具体的，坐席人员通过坐席客户端向人工客服系统的服务端上传设置的快捷应答语句的同时，针对每个快捷应答语句均设置相应的用户意图标签，对应的，人工客服系统的服务端将预先设置的快捷应答语句与用户意图标签之间的对应关系存储至相应的数据库中，以便后续在获取到用户向人工客服系统提交的目标提问问题，以及目标提问问题对应的目标快捷应答语句之后，基于上述对应关系，确定目标快捷应答语句对应的目标意图标签。
S106，基于上述目标意图标签，对上述目标提问问题进行标注，生成用户意图识别样本集；其中，用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练。
具体的，由于基于人工客服系统产生的对话数据，可以获取用户向人工客服系统提交的目标提问问题，以及目标提问问题对应的目标快捷应答语句，即能够获取目标提问问题与目标快捷应答语句之间的对应关系，因此，再基于预设的目标快捷应答语句与目标意图标签之间的对应关系，即可确定目标提问问题对应的目标意图标签，从而可以基于该目标意图标签，对目标提问问题进行标注，即对目标提问问题的意图进行标注，进而生成用户意图识别样本，以及在对大量的目标提问问题进行标注后，生成用户意图识别样本集。
其中，由于上述用户意图识别样本集是基于人工客服系统的真实对话数据生成的，因此，上述用户意图识别样本集中包括了智能客服系统无法获取，进而无法对其进行标注的用户问题，例如，智能客服系统无法准确应答而转由人工客服系统回复的用户问题，以及直接由人工客服回复但智能客服无法识别的用户问题，从人工客服系统获取到智能客服系统无法获取的用户问题并进行标注，从而生成用户意图识别样本集，进而能够提高用户意图识别样本集中所包含的智能客服系统无法识别的用户问 题的覆盖率，进一步的，利用上述用户意图识别样本集对智能客服系统所使用的用户意图识别模型进行训练，即将上述用户意图识别样本输入至智能客服系统所使用的用户意图识别模型，以利用机器学习方法对用户意图识别模型中的模型参数进行迭代训练，得到模型参数更新后的用户意图识别模型，其中，更新后的用户意图识别模型的模型参数为用户意图识别模型对应的目标函数收敛时确定的模型参数；进一步的，在智能客服系统中利用更新后的用户意图识别模型对用户问题进行识别，能够提高用户意图识别模型的识别准确度，进而提高智能客服系统的服务质量。
在具体实施时，上述客户服务场景可以是业务咨询场景，也可以是催收场景，以催收场景为例，首先坐席人员需要基于业务经验，确定在催收场景下用户可能会提出的问题，以及用户提出问题的真实意图，例如，坐席人员确定用户可能会针对催收场景中存在的已扣款，但还是接收到了催收短信的情况提出疑问，即用户提出的问题可能为“款都扣完了，为什么还提示我要还款”，因此确定用户提出该问题的真实意图是针对“已扣款”的情况进行询问，基于此，坐席人员可以预先设置相应的快捷应答语句为“核实到扣款已经成功，十分抱歉给您造成困扰”，用于快捷回复用户针对已扣款，但还是接收到了催收短信的情况提出的问题，并将“已扣款”设置为该快捷应答语句的用户意图标签，即快捷应答语句“核实到扣款已经成功，十分抱歉给您造成困扰”与用户意图标签“已扣款”之间具有对应关系，并将快捷应答语句“核实到扣款已经成功，十分抱歉给您造成困扰”、用户意图标签“已扣款”、及其对应关系存储至人工客服系统后端的数据库中。
进一步的，坐席人员在实际业务处理的过程中，若遇到针对“已扣款”的情况进行询问的用户，例如，获取到用户向人工客服系统提交的目标提问问题为“款都扣完了，为什么还提示我要还款”，又如，获取到用户向人工客服系统提交的目标提问问题为“款都扣完了，我为什么收到了催收短信”，坐席人员均可使用预先设置的“核实到扣款已经成功，十分抱歉给您造成困扰”这一快捷应答语句进行快捷回复，即通过坐席客户端向人工客服系统的服务端上传目标快捷应答语句“核实到扣款已经成功，十分抱歉给您造成困扰”，以使人工客服系统的服务端将该目标快捷应答语句发送至相应的用户终端；对应的，用于用户问题标注的后台服务端，获取用户的目标提问问题，以及该目标提问问题对应的目标快捷应答语句为“核实到扣款已经成功，十分抱歉给您造成困扰”，并确定出该目标快捷应答语句对应的目标意图标签为“已扣款”，从而可以基于该目标意图标签，对目标提问问题进行标注，即使用“已扣款”这一目标意图标签，对目标提问问题“款都扣完了，为什么还提示我要还款”，以及对目标提问问题“款都扣完了，我为什么收到了催收短信”进行标注，进而生成用于对智能客服系统所使用的用户意图识别模型进行训练的用户意图识别样本。
本申请提供的实施例中，通过获取人工客服系统所产生的对话数据，先确定目标提问问题对应的目标快捷应答语句，再结合预设的快捷应答语句与用户意图标签之间的对应关系，确定目标快捷应答语句对应的目标意图标签，即可得到目标提问问题与目标意图标签之间的对应关系，从而实现对用户问题进行用户意图标注，进而将生成的用户意图识别样本集应用于智能客服系统中的用户意图识别模型，即充分借助人工客服系统中所产生的对话数据，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本；并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，提高了用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
其中，针对客服服务场景而言，由于智能客服服务和人工客服服务通常是同步处于启动状态的，因此，存在用户在提出问题的过程中可能会在智能客服服务与人工客服服务之间进行切换的情况，其中，针对用户在提问过程中由智能客服切换为人工客服的情况，用户向人工客服系统提交的问题很可能为智能客服系统无法解答的问题，因此可以通过获取人工客服系统所产生的用户与坐席人员之间的对话数据，即获取用户的目标提问问题，以及坐席人员针对该目标提问问题所回复的目标快捷应答语句，并基于在人工客服系统中预存的目标快捷应答语句与目标意图标签之间的对应关系，从而实现对目标快捷应答语句对应的目标提问问题进行自动标注，进而将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样不仅能够实现自动生成大量的用户意图识别样本，还能够提高智能客服系统无法识别的用户问题的标注覆盖率，基于此，如图2所示，上述S102，获取向人工客服系统提交的目标提问问题，以及获取目标提问问题对应的目标快捷应答语句，具体包括：
S1022，获取人工客服系统在预设时间段内所产生的对话数据集合；
S1024，从上述对话数据集合中，提取多个目标提问问题和目标提问问题对应的目标快捷应答语句；
对应的，在上述S106，基于上述目标意图标签，对上述目标提问问题进行标注，生成用户意图识别样本集之后，还包括：
S108，将上述用户意图识别样本集传输至智能客服系统，以使智能客服系统将用户意图识别样本集存储至第一数据库中；其中，第一数据库为用于存储用户意图识别模型的训练样本集的数据库。
具体的，在获取到人工客服系统在预设时间段内所产生的对话数据集合之后，从对话数据集合中，提取出用户所提出的多个目标提问问题，以及人工客服针对每个目标提问问题所回复的目标快捷应答语句，并基于人工客服系统中预存的目标快捷应答语句与目标意图标签之间的对应关系，确定每个目标快捷应答语句对应的目标意图标签，并基于对应的目标意图标签，对用户所提出的每个目标提问问题进行用户意图标注，进而得到多个用户意图识别样本。
其中，在对提交至人工客服系统的多个目标提问问题进行用户意图标注的过程中，可以采用实时对用户的目标提问问题进行用户意图标注的方式，具体的，实时获取用户向人工客服系统提交的每个目标提问问题，以及获取人工客服针对每个目标提问问题所回复的目标快捷应答语句，并基于人工客服系统中预存的目标快捷应答语句与目标意图标签之间的对应关系(即预存的快捷应答语句与意图标签之间的对应关系)，确定每个目标快捷应答语句对应的目标意图标签，并基于该目标意图标签，实时对用户所提出的每个目标提问问题进行用户意图标注。
具体的，也可以采用先获取人工客服系统在预设时间段内所产生的对话数据集合，再对该对话数据集合中的多个目标提问问题进行用户意图标注，在具体实施时，按照预设时间间隔，获取人工客服系统在预设时间段内所产生的对话数据集合，从该对话数据集合中，提取出多个目标提问问题和每个目标提问问题所对应的目标快捷应答语句，并基于人工客服系统中预存的目标快捷应答语句与目标意图标签之间的对应关系，对对话数据集合中的多个目标提问问题进行用户意图标注，其中，考虑到对话数据集合中可能会存在相似的目标提问问题，因此在对多个目标提问问题进行用户意图标注之前，可以预先对相似的目标提问问题进行去重处理，从而避免对相似的目标提问问题进行重复标注，进而提高标注效率。
具体的，在获取人工客服系统所产生的对话数据集合时，针对人工客服直接使用预设的快捷应答语句回复用户提问问题的情况，可以基于预存的快捷应答语句与意图标签之间的对应关系，对用户提问问题进行标注，以便在对用户提问问题进行用户意图标注的过程中，提高用户意图识别样本中智能客服系统无法识别的用户问题的标注覆盖率，进一步的，考虑到可能存在人工客服在回复用户提问问题时并未直接使用预设的快捷应答语句，而是使用常规应答语句回复用户提问问题(即手动输入应答语句)，此时可以将常规应答语句与预设的快捷语句集合中的快捷应答语句进行相似度匹配，将相似度满足预设条件的快捷应答语句确定为目标提问问题对应的目标快捷应答语句，这样针对人工客服采用手动输入应答语句来回复用户提问问题的情况，也能够自动对该用户提问问题进行用户意图标注，从而进一步提高用户意图识别样本中智能客服系统无法识别的用户问题的标注覆盖率。
在具体实施时，恰恰是通过将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，即在对对话数据集合中的多个目标提问问题进行用户意图标注，生成用户意图识别样本集之后，还需将用户意图识别样本集传输至智能客服系统，以提高智能客服系统无法识别的用户问题的标注覆盖率，具体的，针对上述用户问题的标注方法的执行主体为人工客服系统的第一服务端的情况，可以在人工客服系统与智能客服系统之间设置数据传输通道，从而实现在人工客服系统侧自动记录真实业务流量所产生的用户与坐席之间对话数据(即包括用户提问问题和快捷应答语句)，并基于预设的快捷应答语句与用户意图标签之间的对应关系，对用户提问问题进行用户意图标注，生成用户意图识别样本，然后，由人工客服系统通过预设的数据传输通道将该用户意图识别样本自动传输至智能客服系统；另外，针对上述用户问题的标注方法的执行主体为与人工客服系统的第一服务端通信连接的第二服务端的情况，第二服务端按照预设时间间隔从人工客服系统的第一服务端获取在预设时间段内所产生的对话数据集合，以及获取预设的快捷应答语句与用户意图标签之间的对应关系；从对话数据集合中，提取多个目标提问问题和目标提问问题对应的目标快捷应答语句，再基于预设的快捷应答语句与用户意图标签之间的对应关系，对目标提问问题进行用户意图标注，生成用户意图识别样本，然后，由第二服务端将该用户意图识别样本自动传输至智能客服系统；智能客服系统自动基于获取到的用户意图识别样本，对用户意图识别模型进行训练，得到模型参数更新的用户意图识别模型，这样能够提高用户意图识别样本集的更新及时性，从而提高智能客服系统中用户意图识别模型的训练时效性，进而提高用户意图识别模型对用户提问问题的意图标签识别准确度，进一步提高智能客服系统的服务质量。
在具体实施时，还可以预先将预设的快捷语句集合和用户意图识别样本集发送至样本校验终端，以使运营人员在样本校验终端的页面上对预设的快捷语句集合中的快捷应答语句和用户意图识别样本 集进行校验，例如，对预设的快捷语句集合中的快捷应答语句和用户意图识别样本集进行查询、审核、以及修改等操作，再基于校验通过的快捷应答语句更新快捷语句集合、以及将校验通过的用户意图识别样本集传输至智能客服系统。
本申请提供的实施例中，针对同时启动人工客服服务和智能客服服务的情况，即用户在提出问题的过程中在智能客服服务与人工客服服务之间进行切换的应用场景，若智能客服在服务过程中无法识别用户提问的问题，用户将触发进入人工客服服务阶段，因此，人工客服系统接收到的用户提问问题势必会包含大量智能客服系统无法识别的用户问题，基于此，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，不仅能够实现自动生成大量的用户意图识别样本，还能够提高智能客服系统无法识别的用户问题的标注覆盖率。
其中，为了简化用户问题的标注的整体系统架构，充分利用已有的客服服务系统架构(即包括人工客服系统和智能客服系统)的情况下，通过人工客服系统基于真实业务流量所产生的对话数据和预存的快捷应答语句与意图标签之间的对应关系(即上述第一对应关系)，对用户问题进行意图标注，得到用户意图识别样本集，再将用户意图识别样本集传输至能客服系统，具体的，针对人工客服系统的对话数据集合的生成过程，上述S1022，获取人工客服系统在预设时间段内所产生的对话数据集合，具体包括：
步骤一，将向人工客服系统提交的用户提问问题存储至人工客服系统的第二数据库中；以及，
步骤二，将目标坐席客户端针对用户提问问题所回复的快捷应答语句存储至人工客服系统的第二数据库中；其中，目标坐席客户端即为分配回答用户提问问题的坐席客户端；
步骤三，从第二数据库中，基于用户提问问题和快捷应答语句之间的第二对应关系，获取人工客服系统在预设时间段内所产生的对话数据集合。
其中，上述第二数据库中可以存储用户向人工客服系统提交的用户提问问题与目标坐席客户端所回复的快捷应答语句之间的第二对应关系、以及坐席人员预先通过坐席客户端设置的快捷应答语句与用户意图标签之间的第一对应关系；进一步的，考虑到用户向人工客服系统提交的用户提问问题可能存在重复或者不规范的问题，因此，并不是直接将第二数据库中存储的所有用户问题均作为目标用户问题，而是在获取到对话数据集合后，先对用户提问问题进行预处理，得到多个目标提问问题，该预处理包括去重处理、标准化处理中的至少一项，再基于上述第二对应关系，确定各目标提问问题对应的目标快捷应答语句；然后，再基于上述第一对应关系，确定各目标提问问题对应的目标意图标签，进而基于目标意图标签对目标提问问题进行意图标注，得到多个用户意图识别样本，其中，每个用户意图识别样本包括目标提问问题与目标意图标签之间的第三对应关系。
其中，上述第二对应关系可以是基于用户向人工客服系统提交的用户提问问题的第一消息序列号，以及目标坐席人员(即人工客服)针对用户提问问题所回复的快捷应答语句的第二消息序列号所确定的，具体的，第一消息序列号的时间要早于第二消息序列号的时间。
其中，在目标坐席人员通过目标坐席客户端针对用户提问问题进行回复的过程中，目标坐席人员可以从预设的快捷语句集合中选择与用户提问问题相匹配的快捷应答语句进行回复，其中，上述预设的快捷语句集合中可以包括个人快捷语句集合、以及共享快捷语句集合。
具体的，也可以为目标坐席人员推荐的候选快捷应答语句，该候选快捷应答语句是基于目标坐席人员在信息输入框中手动输入的应答语句，自动从预设的快捷语句集合中匹配出来的快捷应答语句，在具体实施时，可以先在个人快捷语句集合中匹配与目标坐席人员手动输入的应答语句之间的相似度满足预设条件的个人快捷应答语句，若从个人快捷语句集合中未匹配出可用的快捷应答语句，则从共享快捷语句集合中进行匹配，确定可用的共享快捷应答语句。
本申请实施例中，通过将用户向人工客服系统提交的用户提问问题，以及目标坐席人员针对用户提问问题所回复的快捷应答语句存储至人工客服系统的第二数据库中，以便后续从中提取待标注的目标提问问题和对应的目标快捷应答语句，进而实现对目标提问问题进行用户意图标注，得到多个用户意图识别样本。
其中，坐席人员在预先设置快捷应答语句的过程中，可以同时设置各快捷应答语句对应的用户意图标签，以便后续在坐席人员使用快捷应答语句对用户提问问题进行快捷回复的过程，实现快速地对用户提问问题进行用户意图标注，基于此，如图3所示，在上述S102，获取向人工客服系统提交的目标提问问题之前，还包括：
S110，接收第一坐席客户端上传的第一快捷应答语句；以及，接收第一坐席客户端上传的第一快捷应答语句对应的第一意图标签；
S112，存储第一快捷应答语句与第一意图标签之间的第一对应关系。
具体的，考虑到人工客服系统中包括多个坐席人员，每个坐席人员可以根据业务经验确定用户可能会提出的问题、以及用户提出问题的真实意图，进而通过坐席客户端设置相应的快捷应答语句，并针对每个快捷应答语句均设置相应的用户意图标签，并通过坐席客户端预存至人工客服系统的第二数据库中，以便在获取用户向人工客服系统提交的目标提问问题之后，快速地对用户提问问题进行用户意图标注。
在具体实施时，第一坐席客户端接收第一坐席人员输入的第一快捷应答语句，以及接收第一坐席人员针对该第一快捷应答语句所设置的第一意图标签，然后，第一坐席客户端将第一快捷应答语句和第一意图标签上传至人工客服系统的服务端，人工客服系统的服务端将第一快捷应答语句与第一意图标签之间的第一对应关系存储至人工客服系统的第二数据库中。
进一步的，考虑到在为快捷应答语句设置相应的用户意图标签时，可以自动为坐席人员推荐多个备选的用户意图标签，以供坐席人员进行快速选择，提高坐席人员针对快捷应答语句对应的意图标签的设置效率、设置准确度和规范性，基于此，上述S110中，接收第一坐席客户端上传的第一快捷应答语句对应的第一意图标签，具体包括：
步骤一，确定第一快捷应答语句对应的至少一个候选意图标签；
其中，上述候选意图标签可以是基于预设的快捷语句集合中的共享快捷语句或者个人快捷语句对应的用户意图标签所确定的，具体的，将第一坐席人员当前设置的快捷应答语句与预设的快捷语句集合中的共享快捷语句或者个人快捷语句进行相似度匹配，将相似度满足预设条件的快捷语句对应的意图标签确定为候选意图标签。
步骤二，接收第一坐席客户端上传的第一意图标签，其中，该第一意图标签为至少一个候选意图标签中的用户意图标签。
具体的，向第一坐席客户端发送上述至少一个候选意图标签，第一坐席客户端显示该至少一个候选意图标签，并提示第一坐席人员基于该至少一个候选意图标签为第一快捷应答语句设置第一意图标签，接收第一坐席人员在至少一个候选意图标签中选择的用户意图标签作为第一意图标签，然后将该第一意图标签上传至人工客服系统的服务端，以使人工客服系统的服务端存储第一快捷应答语句与第一意图标签之间的对应关系。
其中，为了确保快捷应答语句的私有性，同时，还能够提高多个坐席人员所设置的快捷应答语句的互通性，从而提高坐席人员对快捷应答语句的可选择性，上述快捷语句集合可以包括：个人快捷语句集合和共享快捷语句集合，对应的，上述S112，存储上述第一快捷应答语句与第一意图标签之间的对应关系，具体包括：
在第一坐席客户端对应的个人快捷语句集合中，添加第一快捷应答语句，以及存储第一快捷应答语句与第一意图标签之间的对应关系。
具体的，针对第一坐席人员通过第一坐席客户端设置的第一快捷应答语句和对应的第一意图标签可以直接存储至对应的个人快捷语句集合中，同时，第一坐席人员还可以请求将第一快捷应答语句和对应的第一意图标签存储至共享快捷语句集合，这样该第一快捷应答语句可供其他坐席人员所使用。
其中，考虑到可能存在多个坐席人员使用同一坐席客户端的情况，第一坐席客户端可以按照坐席人员的登录账号进行区分，具体的，在第一坐席客户端所登录的人工客服账号对应的个人快捷语句集合中，添加第一快捷应答语句，以及存储第一快捷应答语句与第一意图标签之间的对应关系。
进一步的，为了提高快捷应答语句的共享性和利用率，可以为坐席人员提供将个人快捷语设置为共享快捷语的功能，以提高坐席人员对快捷应答语句的可选择性，进而能够提高目标提问问题的意图标注全面性，基于此，上述在第一坐席客户端对应的个人快捷语句集合中，添加第一快捷应答语句之后，还包括：
接收第一坐席客户端发送的共享快捷语设置请求；其中，共享快捷语设置请求中包含第一快捷应答语句；
基于上述共享快捷语设置请求中的第一快捷应答语句、以及该第一快捷应答语句对应的第一意图标签，更新共享快捷语句集合。
其中，上述个人快捷语句集合可以是与坐席人员的登录账号相对应的，坐席人员使用某一坐席客户端登录各自的账号后，可以进入对应的个人快捷语句集合的查看页面；上述共享快捷语句集合可以是与某一登录账号群组相对应，该登录账号群组可以包括某一客服业务类别下的多个登录账号，坐席人员使用某一坐席客户端登录该登录账号群组中的任一登录账号后，不仅可以进入该登录账号对应的个人快捷语句集合的查看页面，还可以进入对应的共享快捷语句集合的查看页面。
具体的，针对共享快捷语句集合的更新过程，可以直接在共享快捷语句集合中，添加第一坐席客户端请求转化为共享快捷语的第一快捷应答语句，以及存储第一快捷应答语句与第一意图标签之间的 对应关系；进一步的，为了确保共享快捷语句集合中包含的快捷应答语句和用户意图标签的准确性、规范性，也可以先确定在共享快捷语句集合中是否存在与第一坐席客户端请求转化为共享快捷语的第一快捷应答语句相匹配的第二快捷应答语句；若存在，则根据第一意图标签与第二快捷应答语句对应的第二意图标签之间的相似度大小，向第一坐席客户端返回相应的反馈信息，这样也能够实现对个人快捷语句集合中的第一意图标签进行校验的效果；若不存在，则基于第一坐席客户端请求转化为共享快捷语的第一快捷应答语句、以及对应的第一意图标签，更新共享快捷语句集合。
如图4所示，坐席人员通过坐席客户端登录各自的登录账号aa，可以进入快捷语设置界面，在快捷语设置界面中可以查看已设置的个人快捷应答语句，以及所属坐席群组对应的共享快捷应答语句，并且通过搜索关键词的信息输入框快速搜索相应的快捷应答语句；坐席人员通过点击下方的第一控件可以添加新的快捷应答语句及其对应的用户意图标签；以及通过点击第二控件请求将某一个人快捷应答语句设置为共享快捷语句。
其中，在提高快捷应答语句的共享性和利用率的同时，还需要进一步提高添加至共享快捷语句集合中的快捷应答语句和用户意图标签的准确性、规范性，因此，并不是直接将第一快捷应答语句和对应的第一意图标签添加至共享快捷语句集合，而是通过触发第二坐席客户端对第一意图标签进行打分，并基于打分结果确定第一快捷应答语句的真实意图标签，具体的，上述基于上述共享快捷语设置请求中的第一快捷应答语句、以及该第一快捷应答语句对应的第一意图标签，更新共享快捷语句集合，具体包括：
步骤一，判断共享快捷语句集合中是否存在与第一快捷应答语句匹配的第二快捷应答语句；
步骤二，若判断结果为不存在，则在共享快捷语句集合中，添加第一快捷应答语句；具体的，还可以先对第一快捷应答语句进行预处理，再将预处理后的第一快捷应答语句添加至共享快捷语句集合中，该预处理可以包括纠错处理、标准化处理中的至少一项；
步骤三，基于上述第一快捷应答语句对应的第一意图标签，确定第一快捷应答语句的真实意图标签；其中，真实意图标签为基于第二坐席客户端上传的针对第一意图标签的打分结果确定的；具体的，第二坐席客户端可以包括：登录共享快捷语句集合对应的登录账号群组中各登录账号的坐席客户端、指定坐席客户端中的至少一项，该指定坐席客户端可以是专业人员所使用的坐席客户端；
步骤四，在共享快捷语句集合中，存储第一快捷应答语句与真实意图标签之间的对应关系。
具体的，若共享快捷语句集合中至少一个快捷应答语句与第一快捷应答语句之间的语句相似度大于预设阈值，则确定存在与第一快捷应答语句匹配的第二快捷应答语句；对应的，若共享快捷语句集合中各快捷应答语句与第一快捷应答语句的语句相似度均小于或等于预设阈值，则确定不存在与第一快捷应答语句匹配的第二快捷应答语句。
其中，为了提高共享快捷语句的用户意图标签的准确性和规范性，需要先验证第一快捷应答语句对应的第一意图标签是否满足预设约束条件，若满足，则在共享快捷语句集合中，添加第一快捷应答语句，以及存储第一快捷应答语句与第一意图标签之间的对应关系；若不满足，则确定第一快捷应答语句对应的真实意图标签，在共享快捷语句集合中，存储第一快捷应答语句与真实意图标签之间的对应关系；进一步的，还可以将该真实意图标签发送给第一坐席人员的第一坐席客户端，以使第一坐席人员判断是否基于该真实意图标签，在对应的个人快捷语句集合中，更新第一快捷应答语句与第一意图标签之间的对应关系。
其中，针对验证第一快捷应答语句对应的第一意图标签是否满足预设约束条件的过程，可以将第一快捷应答语句与第一意图标签之间的对应关系分别发送至至少一个第二坐席人员的第二坐席客户端，其中，该第二坐席人员可以是目标坐席群组中各坐席人员，其中第一坐席人员与第二坐席人员同属于目标坐席群组；然后，接收第二坐席客户端返回的针对第一意图标签的打分结果，并基于多个第二坐席人员针对该第一意图标签的打分结果，确定第一快捷应答语句对应的第一意图标签是否满足预设约束条件，例如，若综合分值大于预设分值阈值，则确定满足预设约束条件；若综合分值小于或等于预设分值阈值，则确定不满足预设约束条件。
也就是说，针对共享快捷语句集合中不存在与第一快捷应答语句匹配的第二快捷应答语句的情况，并不是直接在共享快捷语句集合中存储第一快捷应答语句与第一意图标签之间的对应关系，而是先将第一快捷应答语句与第一意图标签之间的对应关系发送至第二坐席客户端，第二坐席客户端返回针对该第一意图标签的打分结果，其中，该打分结果中的分值大小与第一意图标签的准确度成正比；对应的，服务端基于多个第二坐席客户端的打分结果确定第一意图标签的综合分值，其中，该综合分值可以是对多个第二坐席客户端的打分结果对应的分值进行加权平均得到的；若综合分值大于预设分值阈值，则确定第一意图标签为第一快捷应答语句的真实意图标签；若综合分值小于或等于预设分值阈值， 则触发指定坐席客户端返回第一快捷应答语句的真实意图标签；然后，再在共享快捷语句集合中，存储第一快捷应答语句与真实意图标签之间的对应关系。
指定坐席客户端可以是专业人员所使用的坐席客户端，该坐席客户端具备特定领域的专业知识，针对该领域的提问问题，能够更准确的确定各种应答语句对应的真实意图及对应的真实意图标签。
需要说明的是，考虑到可能存在多个坐席人员使用同一坐席客户端的情况，因此，第一坐席客户端和第二坐席客户端可以是不同的坐席客户端，也可以是同一坐席客户端，即不同的坐席人员在不同时间节点使用同一坐席客户端的情况，此时，第一坐席客户端的登录账号和第二坐席客户端的登录账号可以是不同的。
本申请实施例中，先验证第一快捷应答语句对应的第一意图标签是否满足预设约束条件，若满足，则在共享快捷语句集合中，添加第一快捷应答语句，以及存储第一快捷应答语句与第一意图标签之间的对应关系；若不满足，则确定第一快捷应答语句对应的真实意图标签，以及将最终确定的第一快捷应答语句的真实意图标签发送给第一坐席人员的第一坐席客户端，以使第一坐席人员判断是否基于该真实意图标签，在对应的个人快捷语句集合中，更新第一快捷应答语句与第一意图标签之间的对应关系，这样不仅能够提高共享快捷语句的用户意图标签的准确性和规范性，同时，还能够提高个人快捷语句的用户意图标签的准确性和规范性。
其中，在上述步骤一，判断共享快捷语句集合中是否存在与第一快捷应答语句匹配的第二快捷应答语句之后，还包括：
若判断结果为存在，则判断第一意图标签与第二快捷应答语句对应的第二意图标签的相似度是否大于预设阈值；
若大于，则向第一坐席客户端发送第一反馈信息；其中，第一反馈信息指示共享快捷语句集合已包含第一快捷应答语句；
若不大于，则向第一坐席客户端发送第二反馈信息；其中，第二反馈信息包括推荐使用的第二意图标签。
具体的，针对共享快捷语句集合中存在与第一快捷应答语句匹配的第二快捷应答语句的情况，并不是在共享快捷语句集合中添加重复的快捷应答语句，也不是不做任何处理，而是将第一意图标签与对应的第二意图标签进行相似度比对，考虑到由于共享快捷语句集合中的第二快捷应答语句对应的第二意图标签均是已完成用户意图标签验证且验证通过的，即第二意图标签是可信的，因此，可以先将第一意图标签与第二意图标签进行相似度对比，并根据相似度对比结果向第一坐席客户端返回相应的反馈信息，并且在第一意图标签与第二意图标签的相似度比较低的情况下，可能存在第一意图标签设置不准确的问题，此时可以将第二意图标签返回给第一坐席客户端，这样第一坐席人员可以基于第二意图标签来决定是否修改个人快捷语句集合中第一快捷应答语句的第一意图标签，从而达到提高个人快捷语句集合中针对快捷应答语句所配置的用户意图标签的准确度和规范性，进而提升用户问题的意图标注准确度。
其中，为了进一步提高用户问题的意图标注效率，针对快捷语句集合包括个人快捷语句集合和共享快捷语句集合的情况，在上述S104，基于预设的第一对应关系，确定目标快捷应答语句对应的目标意图标签之前，还包括：
若目标快捷应答语句属于个人快捷语句集合，则在个人快捷语句集合中，查询第一对应关系；
若目标快捷应答语句属于共享快捷语句集合，则在共享快捷语句集合中，查询第一对应关系。
本申请实施例中，通过记录目标快捷应答语句对应的来源信息，该来源信息包括：个人快捷语句集合的标识、或者共享快捷语句集合的标识，以便快速在个人快捷语句集合或者共享快捷语句集合中，查找到针对目标快捷应答语句所记录的第一对应关系，进而快速确定目标提问问题对应的目标意图标签。
其中，在上述S106，基于上述目标意图标签，对上述目标提问问题进行标注，生成用户意图识别样本集之后，还包括：
利用机器学习方法基于上述用户意图识别样本集，对用户意图识别模型进行训练，得到训练的用户意图识别模型；
利用用户意图识别模型，对智能客服系统接收到的用户提问问题进行意图识别，输出相应的用户意图识别结果。
具体的，将基于人工客服系统所产生的对话数据得到的用户意图识别样本集输入至待训练的用户意图识别模型中，利用机器学习方法并基于上述用户意图识别样本集，对用户意图识别模型的模型参数进行迭代训练，得到训练后的用户意图识别模型；其中，训练后的用户意图识别模型的模型参数为用户意图识别模型对应的目标函数收敛时确定的模型参数；进一步的，在智能客服系统中利用训练后 的用户意图识别模型对智能客服系统接收到的用户提问问题进行意图识别，并输出相应的用户意图识别结果，能够提高识别准确度，进而提高智能客服系统的服务质量。
在具体实施时，针对智能客服系统的用户意图识别模型的训练过程，如图5所示，获取人工客服系统在预设时间段内所产生的对话数据集合，并从该对话数据集合中提取出用户向人工客服系统提交的目标提问问题1至目标提问问题n，以及目标提问问题1至目标提问问题n分别对应的目标快捷应答语句(即目标快捷应答语句1至目标快捷应答语句n)；再基于人工客服系统中预存的目标快捷应答语句与目标意图标签之间的对应关系，确定目标提问问题1至目标提问问题n分别对应的目标意图标签(即目标意图标签1至目标意图标签n)，并基于目标意图标签1至目标意图标签n，对目标提问问题1至目标提问问题n进行标注，生成用户意图识别样本集(例如，目标提问问题1-目标意图标签1至目标提问问题n-目标意图标签n)，进一步的，将上述从人工客服系统获取到的用户意图识别样本集，输入至智能客服系统所使用的待训练的用户意图识别模型中，利用机器学习方法并基于上述用户意图识别样本集，对用户意图识别模型的模型参数进行迭代训练，得到训练后的用户意图识别模型。
本申请实施例中，充分借助人工客服系统中所产生的对话数据，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本；并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，提高了用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
在一个具体的实施例中，针对上述用户问题的标注方法的执行主体为人工客服系统的第一服务端的情况，用户问题的标注方法的具体应用场景如图6所示，首先，由人工客服系统的坐席人员基于业务经验确定在特定的业务场景下，用户可能会提出的问题、以及用户提出问题的真实意图，并由人工客服系统的坐席后端接收坐席人员预先在人工客服系统的坐席前端页面上针对用户可能会提出的问题，设置相应的用于快捷回复用户问题的快捷应答语句，以及快捷应答语句对应的意图标签，并存储至人工客服系统的第二数据库中；然后，在接收到用户通过用户前端进入人工客服系统后，由人工客服系统自动为用户分配坐席人员；
具体的，预先通过第一模块设置用户与坐席人员之间的连接渠道，该第一模块是基于用户接入人工客服系统的端口确定的，例如，人工客服系统的端口可以包括微信小程序端口、网页端口、以及应用端口中任一项，若用户从微信小程序端口(即第一模块为微信小程序端口)接入人工客服系统，则为用户分配微信小程序对应的坐席人员，并通过长连接模块建立坐席人员与用户之间的连接，以使在用户很多的情况下，使得坐席人员可以通过长连接模块保持与用户之间的会话，进而可以切换不同的用户；
进一步的，在坐席人员与用户之间进行对话的过程中，依次通过长连接模块、第二模块将用户提问问题和坐席人员回复的应答语句存储至第二数据库中，具体的，用户与坐席人员之间进行对话时，用户通过坐席前端、坐席后端和长连接模块向人工客服系统发送问题消息(即用户提问问题)；以及坐席通过坐席前端、坐席后端和长连接模块向用户的客户端发送回复消息(即客服应答语句，该客服应答语句可以是常规应答语句，也可以是快捷应答语句)；其中，该长连接模块用于校验问题消息和回复消息的格式是否正确并投递消息到第二模块(依次包括kafka模块、message模块)，该第二模块用于对消息进行预处理(例如保存、记录消息序列号、缓存等处理)，进而将其存储至第二数据库中；
进一步的，由数据处理模块从第二数据库中获取人工客服系统在预设时间段内所产生的对话数据集合，并从对话数据集合中提取出用户向人工客服系统提交的多个目标提问问题，以及基于提问问题的第一消息序列号和快捷应答语句的第二消息序列号，确定坐席人员针对每个目标提问问题所回复的目标快捷应答语句，再基于人工客服系统的第二数据库中预存的快捷应答语句与意图标签之间的对应关系，确定出坐席人员针对每个目标提问问题所回复的目标快捷应答语句对应的意图标签，进而确定出目标提问问题与意图标签之间的对应关系，进而实现对目标提问问题进行用户意图标注，生成用户意图识别样本集。上述基于提问问题的第一消息序列号和快捷应答语句的第二消息序列号，确定坐席人员针对每个目标提问问题所回复的目标快捷应答语句包括：根据该快捷应答语句中的关键词以及至少一个提问问题中关键词的匹配度确定该快捷应答语句对应的具体提问问题；或者，根据根据该快捷应答语句与至少一个提问问题的匹配度确定该快捷应答语句对应的一个或多个具体提问问题。在确定该快捷应答语句对应的一个或多个具体提问问题之后，再基于人工客服系统的第二数据库中预存的快捷应答语句与意图标签之间的对应关系，确定出坐席人员针对每个目标提问问题所回复的目标快捷应 答语句对应的意图标签，进而确定出目标提问问题与意图标签之间的对应关系，进而实现对目标提问问题进行用户意图标注，生成用户意图识别样本集。
进一步的，预先在人工客服系统与智能客服系统之间建立长连接，并将上述在人工客服系统生成的用户意图识别样本集通过长连接的渠道传输至智能客服系统，以使智能客服系统将用户意图识别样本集存储至第一数据库中，并利用机器学习方法基于上述用户意图识别样本集，对用户意图识别模型进行训练，得到训练的用户意图识别模型，以及利用训练后的用户意图识别模型，对智能客服系统接收到的用户提问问题进行意图识别，并输出相应的用户意图识别结果。
本申请实施例中的用户问题的标注方法，通过获取人工客服系统所产生的对话数据，先确定目标提问问题对应的目标快捷应答语句，再结合预设的快捷应答语句与用户意图标签之间的对应关系，确定目标快捷应答语句对应的目标意图标签，即可得到目标提问问题与目标意图标签之间的对应关系，从而实现对用户问题进行用户意图标注，进而将生成的用户意图识别样本集应用于智能客服系统中的用户意图识别模型，即充分借助人工客服系统中所产生的对话数据，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本；并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，提高了用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
对应上述图1至图6描述的用户问题的标注方法，基于相同的技术构思，本申请实施例还提供了一种用户问题的标注装置，图7为本申请实施例提供的用户问题的标注装置的模块组成示意图，该装置用于执行图1至图6描述的用户问题的标注方法，如图7所示，该装置包括：
第一获取模块702，用于获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句；其中，所述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，所述快捷应答语句为在所述人工客服系统中预先设置的用于快捷回复用户问题的应答语句；
第一确定模块704，用于基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签；其中，所述第一对应关系包括在所述人工客服系统中预存的所述目标快捷应答语句与所述目标意图标签之间的对应关系；
第一生成模块706，用于基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集；其中，所述用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练。
本申请实施例中的用户问题的标注装置，通过获取人工客服系统所产生的对话数据，先确定目标提问问题对应的目标快捷应答语句，再结合预设的快捷应答语句与用户意图标签之间的对应关系，确定目标快捷应答语句对应的目标意图标签，即可得到目标提问问题与目标意图标签之间的对应关系，从而实现对用户问题进行用户意图标注，进而将生成的用户意图识别样本集应用于智能客服系统中的用户意图识别模型，即充分借助人工客服系统中所产生的对话数据，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本；并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，提高了用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
需要说明的是，本申请中关于用户问题的标注装置的实施例与本申请中关于用户问题的标注方法的实施例基于同一发明构思，因此该实施例的具体实施可以参见前述对应的用户问题的标注方法的实施，重复之处不再赘述。
进一步地，对应上述图1至图6所示的方法，基于相同的技术构思，本申请实施例还提供了一种用户问题的标注设备，该设备用于执行上述的用户问题的标注方法，如图8所示。
用户问题的标注设备可因配置或性能不同而产生比较大的差异，可以包括一个或一个以上的处理器801和存储器802，存储器802中可以存储有一个或一个以上应用程序或数据。其中，存储器802可以是短暂存储或持久存储。存储在存储器802的应用程序可以包括一个或一个以上模块(图示未示出)，每个模块可以包括一系列计算机可执行指令。更进一步地，处理器801可以设置为与存储器802通信， 以在用户问题的标注设备上执行存储器802中的一系列计算机可执行指令。用户问题的标注设备还可以包括一个或一个以上电源803，一个或一个以上有线或无线网络接口804，一个或一个以上输入输出接口805，一个或一个以上键盘806等。
在一个具体的实施例中，用户问题的标注设备包括有存储器，其中一个或者一个以上程序存储于存储器中，且一个或者一个以上程序可以包括一个或一个以上模块，且每个模块可以包括一系列计算机可执行指令，且经配置以由一个或者一个以上处理器执行；该一个或者一个以上程序包含的计算机可执行指令被该处理器执行时，用于实现以下流程：
获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句；其中，所述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，所述快捷应答语句为在所述人工客服系统中预先设置的用于快捷回复用户问题的应答语句；
基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签；其中，所述第一对应关系包括在所述人工客服系统中预存的所述目标快捷应答语句与所述目标意图标签之间的对应关系；
基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集；其中，所述用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练。
本申请实施例中的用户问题的标注设备，通过获取人工客服系统所产生的对话数据，先确定目标提问问题对应的目标快捷应答语句，再结合预设的快捷应答语句与用户意图标签之间的对应关系，确定目标快捷应答语句对应的目标意图标签，即可得到目标提问问题与目标意图标签之间的对应关系，从而实现对用户问题进行用户意图标注，进而将生成的用户意图识别样本集应用于智能客服系统中的用户意图识别模型，即充分借助人工客服系统中所产生的对话数据，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本；并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，提高了用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
需要说明的是，本申请中关于用户问题的标注设备的实施例与本申请中关于用户问题的标注方法的实施例基于同一发明构思，因此该实施例的具体实施可以参见前述对应的用户问题的标注方法的实施，重复之处不再赘述。
进一步地，对应上述图1至图6所示的方法，基于相同的技术构思，本申请实施例还提供了一种存储介质，用于存储计算机可执行指令，一种具体的实施例中，该存储介质可以为U盘、光盘、硬盘等，该存储介质存储的计算机可执行指令在被处理器执行时，能实现以下流程：
获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句；其中，所述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，所述快捷应答语句为在所述人工客服系统中预先设置的用于快捷回复用户问题的应答语句；
基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签；其中，所述第一对应关系包括在所述人工客服系统中预存的所述目标快捷应答语句与所述目标意图标签之间的对应关系；
基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集；其中，所述用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练。
本申请实施例中的存储介质存储的计算机可执行指令在被处理器执行时，通过获取人工客服系统所产生的对话数据，先确定目标提问问题对应的目标快捷应答语句，再结合预设的快捷应答语句与用户意图标签之间的对应关系，确定目标快捷应答语句对应的目标意图标签，即可得到目标提问问题与目标意图标签之间的对应关系，从而实现对用户问题进行用户意图标注，进而将生成的用户意图识别样本集应用于智能客服系统中的用户意图识别模型，即充分借助人工客服系统中所产生的对话数据，自动生成智能客服系统中所需的用户意图训练样本集，将人工客服系统的真实业务流量与智能客服系统的用户意图识别需求进行紧密关联，这样能够提高用户提问问题的标注效率，从而为智能客服系统提供大量的用户意图识别样本；并且，用户向人工客服系统所提交的目标提问问题不仅包括智能客服系统无法准确应答而转由人工客服系统回复的用户问题，还包括直接由人工客服回复但智能客服无法识别的用户问题，这样通过自动对这些目标提问问题进行用户意图标注，得到用户意图识别样本集，提高了用户意图识别样本集中所包含的智能客服系统无法识别的用户问题的覆盖率，从而提高用户意图识别模型的识别准确度，进而确保智能客服系统的服务质量。
需要说明的是，本申请中关于存储介质的实施例与本申请中关于用户问题的标注方法的实施例基于同一发明构思，因此该实施例的具体实施可以参见前述对应的用户问题的标注方法的实施，重复之处不再赘述。
上述对本申请特定实施例进行了描述。其它实施例在所附权利要求书的范围内。在一些情况下，在权利要求书中记载的动作或步骤可以按照不同于实施例中的顺序来执行并且仍然可以实现期望的结果。另外，在附图中描绘的过程不一定按照示出的特定顺序或者连续顺序才能实现期望的结果。在某些实施方式中，多任务处理和并行处理也是可以的或者可能是有利的。
在20世纪90年代，对于一个技术的改进可以很明显地区分是硬件上的改进(例如，对二极管、晶体管、开关等电路结构的改进)还是软件上的改进(对于方法流程的改进)。然而，随着技术的发展，当今的很多方法流程的改进已经可以视为硬件电路结构的直接改进。设计人员几乎都通过将改进的方法流程编程到硬件电路中来得到相应的硬件电路结构。因此，不能说一个方法流程的改进就不能用硬件实体模块来实现。例如，可编程逻辑器件(Programmable Logic Device，PLD)(例如现场可编程门阵列(Field Programmable Gate Array，FPGA))就是这样一种集成电路，其逻辑功能由用户对器件编程来确定。由设计人员自行编程来把一个数字系统“集成”在一片PLD上，而不需要请芯片制造厂商来设计和制作专用的集成电路芯片。而且，如今，取代手工地制作集成电路芯片，这种编程也多半改用“逻辑编译器(logic compiler)”软件来实现，它与程序开发撰写时所用的软件编译器相类似，而要编译之前的原始代码也得用特定的编程语言来撰写，此称之为硬件描述语言(Hardware Description Language，HDL)，而HDL也并非仅有一种，而是有许多种，如ABEL(Advanced Boolean Expression Language)、AHDL(Altera Hardware Description Language)、Confluence、CUPL(Cornell University Programming Language)、HD Cal、JHDL(Java Hardware Description Language)、Lava、Lola、My HDL、PALASM、RHDL(Ruby Hardware Description Language)等，目前最普遍使用的是VHDL(Very-High-Speed Integrated Circuit Hardware Description Language)与Verilog。本领域技术人员也应该清楚，只需要将方法流程用上述几种硬件描述语言稍作逻辑编程并编程到集成电路中，就可以很容易得到实现该逻辑方法流程的硬件电路。
控制器可以按任何适当的方式实现，例如，控制器可以采取例如微处理器或处理器以及存储可由该(微)处理器执行的计算机可读程序代码(例如软件或固件)的计算机可读介质、逻辑门、开关、专用集成电路(Application Specific Integrated Circuit，ASIC)、可编程逻辑控制器和嵌入微控制器的形式，控制器的例子包括但不限于以下微控制器：ARC 625D、Atmel AT91SAM、Microchip PIC18F26K20以及Silicone Labs C8051F320，控制器还可以被实现为存储器的控制逻辑的一部分。本领域技术人员也知道，除了以纯计算机可读程序代码方式实现控制器以外，完全可以通过将方法步骤进行逻辑编程来使得控制器以逻辑门、开关、专用集成电路、可编程逻辑控制器和嵌入微控制器等的形式来实现相同功能。因此这种控制器可以被认为是一种硬件部件，而对其内包括的用于实现各种功能的装置也可以视为硬件部件内的结构。或者甚至，可以将用于实现各种功能的装置视为既可以是实现方法的软件模块又可以是硬件部件内的结构。
上述实施例阐明的系统、装置、模块或单元，具体可以由计算机芯片或实体实现，或者由具有某种功能的产品来实现。一种典型的实现设备为计算机。具体的，计算机例如可以为个人计算机、膝上型计算机、蜂窝电话、相机电话、智能电话、个人数字助理、媒体播放器、导航设备、电子邮件设备、游戏控制台、平板计算机、可穿戴设备或者这些设备中的任何设备的组合。
为了描述的方便，描述以上装置时以功能分为各种单元分别描述。当然，在实施本申请时可以把各单元的功能在同软件和/或硬件中实现。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图的一个流程或多个流程和/或方框图的一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图的一个流程或多个流程和/或方框图的一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图的一个流程或多个流程和/或方框图的一个方框或多个方框中指定的功能的步骤。
在一个典型的配置中，计算设备包括处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体，可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
还需要说明的是，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、商品或者设备不仅包括那些要素，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、商品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白，本申请的实施例可提供为方法、系统或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本申请可采用在其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请可以在由计算机执行的计算机可执行指令的一般上下文中描述，例如程序模块。一般地，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中实践本申请，在这些分布式计算环境中，由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中，程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
本申请中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于系统实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
以上所述仅为本申请的实施例而已，并不用于限制本申请。对于本领域技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本申请的权利要求范围之内。

Claims (14)

  1. 一种用户问题的标注方法，应用于人工客服系统的后台服务端，所述方法包括：
    获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句；其中，所述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，所述快捷应答语句为在所述人工客服系统中预先设置的用于快捷回复用户问题的应答语句；
    基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签；其中，所述第一对应关系包括在所述人工客服系统中预存的所述目标快捷应答语句与所述目标意图标签之间的对应关系；
    基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集；其中，所述用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练，所述用户意图识别模型用于对智能客服系统接收到的用户提问问题进行意图识别，输出相应的用户意图识别结果。
  2. 根据权利要求1所述的方法，其中，所述获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句，包括：
    获取所述人工客服系统在预设时间段内所产生的对话数据集合；
    从所述对话数据集合中，提取多个所述目标提问问题和所述目标提问问题对应的所述目标快捷应答语句；
    在所述基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集之后，还包括：
    将所述用户意图识别样本集传输至智能客服系统，以使所述智能客服系统将所述用户意图识别样本集存储至第一数据库中；其中，所述第一数据库为用于存储所述用户意图识别模型的训练样本集的数据库。
  3. 根据权利要求2所述的方法，其中，所述获取所述人工客服系统在预设时间段内所产生的对话数据集合，包括：
    将向所述人工客服系统提交的用户提问问题存储至所述人工客服系统的第二数据库中；以及，将目标坐席客户端针对所述用户提问问题所回复的快捷应答语句存储至所述第二数据库中；
    从所述第二数据库中，基于所述用户提问问题和所述快捷应答语句之间的第二对应关系，获取所述人工客服系统在所述预设时间段内所产生的对话数据集合。
  4. 根据权利要求1-3任一项所述的方法，其中，在所述获取向人工客服系统提交的目标提问问题之前，还包括：
    接收第一坐席客户端上传的第一快捷应答语句；
    接收所述第一坐席客户端上传的所述第一快捷应答语句对应的第一意图标签；以及，
    存储所述第一快捷应答语句与所述第一意图标签之间的对应关系。
  5. 根据权利要求4所述的方法，其中，所述接收所述第一坐席客户端上传的所述第一快捷应答语句对应的第一意图标签，包括：
    确定所述第一快捷应答语句对应的至少一个候选意图标签；
    接收所述第一坐席客户端上传的所述第一意图标签，其中，所述第一意图标签为所述至少一个候选意图标签中的用户意图标签。
  6. 根据权利要求4或5所述的方法，其中，所述快捷语句集合包括：个人快捷语句集合和共享快捷语句集合；
    所述存储所述第一快捷应答语句与所述第一意图标签之间的对应关系，包括：
    在所述第一坐席客户端对应的所述个人快捷语句集合中，添加所述第一快捷应答语句，以及存储所述第一快捷应答语句与所述第一意图标签之间的对应关系。
  7. 根据权利要求6所述的方法，其中，在所述第一坐席客户端对应的所述个人快捷语句集合中，添加所述第一快捷应答语句之后，还包括：
    接收所述第一坐席客户端发送的共享快捷语设置请求；其中，所述共享快捷语设置请求中包含所述第一快捷应答语句；
    基于所述第一快捷应答语句和所述第一意图标签，更新所述共享快捷语句集合。
  8. 根据权利要求7所述的方法，其中，所述基于所述第一快捷应答语句和所述第一意图标签，更新所述共享快捷语句集合，包括：
    判断所述共享快捷语句集合中是否存在与所述第一快捷应答语句匹配的第二快捷应答语句；
    若不存在，则在所述共享快捷语句集合中，添加所述第一快捷应答语句；
    基于所述第一意图标签，确定所述第一快捷应答语句的真实意图标签；其中，所述真实意图标签为基于第二坐席客户端上传的针对所述第一意图标签的打分结果确定的；以及，
    在所述共享快捷语句集合中，存储所述第一快捷应答语句与所述真实意图标签之间的对应关系。
  9. 根据权利要求8所述的方法，其中，在判断所述共享快捷语句集合中是否存在与所述第一快捷应答语句匹配的第二快捷应答语句之后，还包括：
    若存在，则判断所述第一意图标签与所述第二快捷应答语句对应的第二意图标签的相似度是否大于预设阈值；
    若大于，则向所述第一坐席客户端发送第一反馈信息；其中，所述第一反馈信息指示所述共享快捷语句集合已包含所述第一快捷应答语句；
    若不大于，则向所述第一坐席客户端发送第二反馈信息；其中，所述第二反馈信息包括推荐使用的所述第二意图标签。
  10. 根据权利要求6-9任一项所述的方法，其中，在基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签之前，还包括：
    若所述目标快捷应答语句属于所述个人快捷语句集合，则在所述个人快捷语句集合中，查询所述第一对应关系；
    若所述目标快捷应答语句属于所述共享快捷语句集合，则在所述共享快捷语句集合中，查询所述第一对应关系。
  11. 根据权利要求1-10任一项所述的方法，其中，在基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集之后，还包括：
    利用机器学习方法基于所述用户意图识别样本集，对用户意图识别模型进行训练，得到训练的用户意图识别模型；
    利用所述用户意图识别模型，对所述智能客服系统接收到的用户提问问题进行意图识别，输出所述相应的用户意图识别结果。
  12. 一种用户问题的标注装置，包括：
    第一获取模块，用于获取向人工客服系统提交的目标提问问题，以及获取所述目标提问问题对应的目标快捷应答语句；其中，所述目标快捷应答语句包括：预设的快捷语句集合中的至少一个快捷应答语句，所述快捷应答语句为在所述人工客服系统中预先设置的用于快捷回复用户问题的应答语句；
    第一确定模块，用于基于预设的第一对应关系，确定所述目标快捷应答语句对应的目标意图标签；其中，所述第一对应关系包括在所述人工客服系统中预存的所述目标快捷应答语句与所述目标意图标签之间的对应关系；
    第一生成模块，用于基于所述目标意图标签，对所述目标提问问题进行标注，生成用户意图识别样本集；其中，所述用户意图识别样本集用于对智能客服系统所使用的用户意图识别模型进行训练，所述用户意图识别模型用于对智能客服系统接收到的用户提问问题进行意图识别，输出相应的用户意图识别结果。
  13. 一种用户问题的标注设备，包括：
    处理器；以及
    被安排成存储计算机可执行指令的存储器，所述可执行指令被配置由所述处理器执行，所述可执行指令被所述处理器执行时用于实现如权利要求1-11任一项所述的方法中的步骤。
  14. 一种存储介质，其中，所述存储介质用于存储计算机可执行指令，所述可执行指令使得计算机执行如权利要求1-11任一项所述的方法。
PCT/CN2022/117665 2021-12-31 2022-09-07 用户问题的标注方法及装置 WO2023124215A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP22913488.7A EP4345645A1 (en) 2021-12-31 2022-09-07 User question labeling method and device
US18/395,240 US20240127259A1 (en) 2021-12-31 2023-12-22 User question labeling method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111678653.1A CN114490975B (zh) 2021-12-31 2021-12-31 用户问题的标注方法及装置
CN202111678653.1 2021-12-31

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/395,240 Continuation US20240127259A1 (en) 2021-12-31 2023-12-22 User question labeling method and apparatus

Publications (1)

Publication Number Publication Date
WO2023124215A1 true WO2023124215A1 (zh) 2023-07-06

Family

ID=81510945

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/117665 WO2023124215A1 (zh) 2021-12-31 2022-09-07 用户问题的标注方法及装置

Country Status (4)

Country Link
US (1) US20240127259A1 (zh)
EP (1) EP4345645A1 (zh)
CN (1) CN114490975B (zh)
WO (1) WO2023124215A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114490975B (zh) * 2021-12-31 2023-02-07 马上消费金融股份有限公司 用户问题的标注方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110619051A (zh) * 2019-08-16 2019-12-27 科大讯飞(苏州)科技有限公司 问题语句分类方法、装置、电子设备及存储介质
CN112328761A (zh) * 2020-11-03 2021-02-05 中国平安财产保险股份有限公司 一种意图标签设置方法、装置、计算机设备及存储介质
CN112347760A (zh) * 2020-11-16 2021-02-09 北京京东尚科信息技术有限公司 意图识别模型的训练方法及装置、意图识别方法及装置
US20210256969A1 (en) * 2020-02-19 2021-08-19 Baidu Online Network Technology (Beijing) Co., Ltd. Method and apparatus for recognizing user intention, device, and readable storage medium
CN114490975A (zh) * 2021-12-31 2022-05-13 马上消费金融股份有限公司 用户问题的标注方法及装置

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6264215B2 (ja) * 2014-07-18 2018-01-24 トヨタ自動車株式会社 質疑応答装置および質疑応答装置の制御方法
CN108446286B (zh) * 2017-02-16 2023-04-25 阿里巴巴集团控股有限公司 一种自然语言问句答案的生成方法、装置及服务器
JP6694910B2 (ja) * 2018-04-03 2020-05-20 株式会社Ｌ ｉｓ Ｂ Ｆａｑ支援装置、ｆａｑ支援システム、ｆａｑ支援方法およびｆａｑ支援プログラム
CN109460451A (zh) * 2018-09-29 2019-03-12 厦门快商通信息技术有限公司 主动获取用户信息的方法、智能客服方法及系统
CN109783704B (zh) * 2019-01-03 2021-02-02 中国科学院自动化研究所 人机混合的应答方法、系统、装置
CN110162611B (zh) * 2019-04-23 2021-03-26 苏宁金融科技(南京)有限公司 一种智能客服应答方法及系统
CN111429157A (zh) * 2020-03-27 2020-07-17 上海东普信息科技有限公司 投诉工单的评价处理方法、装置、设备及存储介质
CN111400482B (zh) * 2020-06-05 2020-09-01 支付宝(杭州)信息技术有限公司 一种建立问答对的方法及装置
CN113722452B (zh) * 2021-07-16 2024-01-19 上海通办信息服务有限公司 一种问答系统中基于语义的快速知识命中方法及装置
CN113486182A (zh) * 2021-08-02 2021-10-08 上海明略人工智能(集团)有限公司 一种意图识别的方法、系统、电子设备及可读存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110619051A (zh) * 2019-08-16 2019-12-27 科大讯飞(苏州)科技有限公司 问题语句分类方法、装置、电子设备及存储介质
US20210256969A1 (en) * 2020-02-19 2021-08-19 Baidu Online Network Technology (Beijing) Co., Ltd. Method and apparatus for recognizing user intention, device, and readable storage medium
CN112328761A (zh) * 2020-11-03 2021-02-05 中国平安财产保险股份有限公司 一种意图标签设置方法、装置、计算机设备及存储介质
CN112347760A (zh) * 2020-11-16 2021-02-09 北京京东尚科信息技术有限公司 意图识别模型的训练方法及装置、意图识别方法及装置
CN114490975A (zh) * 2021-12-31 2022-05-13 马上消费金融股份有限公司 用户问题的标注方法及装置

Also Published As

Publication number Publication date
US20240127259A1 (en) 2024-04-18
EP4345645A1 (en) 2024-04-03
CN114490975B (zh) 2023-02-07
CN114490975A (zh) 2022-05-13

Similar Documents

Publication Publication Date Title
CN108897867B (zh) 用于知识问答的数据处理方法、装置、服务器和介质
US11948058B2 (en) Utilizing recurrent neural networks to recognize and extract open intent from text inputs
US10937413B2 (en) Techniques for model training for voice features
US10733197B2 (en) Method and apparatus for providing information based on artificial intelligence
US11087094B2 (en) System and method for generation of conversation graphs
CN107832433B (zh) 基于对话交互的信息推荐方法、装置、服务器和存储介质
CN109002510B (zh) 一种对话处理方法、装置、设备和介质
US20190311036A1 (en) System and method for chatbot conversation construction and management
US11651250B2 (en) Automatically generated conversation output
CN109299320B (zh) 一种信息交互方法、装置、计算机设备和存储介质
US20220164683A1 (en) Generating a domain-specific knowledge graph from unstructured computer text
CN107210035A (zh) 语言理解系统和方法的生成
US11163778B2 (en) Integrating virtual and human agents in a multi-channel support system for complex software applications
WO2018214895A1 (zh) 数据处理方法、数据处理装置、存储设备及网络设备
JP2020521210A (ja) 情報処理方法及び端末、コンピュータ記憶媒体
US11004449B2 (en) Vocal utterance based item inventory actions
US10854189B2 (en) Techniques for model training for voice features
US10643601B2 (en) Detection mechanism for automated dialog systems
US20240127259A1 (en) User question labeling method and apparatus
US11423219B2 (en) Generation and population of new application document utilizing historical application documents
WO2020233381A1 (zh) 基于语音识别的服务请求方法、装置及计算机设备
CN116521841A (zh) 用于生成回复信息的方法、装置、设备及介质
US11144727B2 (en) Evaluation framework for intent authoring processes
US20200159824A1 (en) Dynamic Contextual Response Formulation
CN111241236B (zh) 任务导向性问答方法、系统、电子设备、可读存储介质

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: 22913488

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2022913488

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2022913488

Country of ref document: EP

Effective date: 20231227