WO2019144773A1 - 机器人客服转人工客服的方法和装置 - Google Patents

机器人客服转人工客服的方法和装置 Download PDF

Info

Publication number
WO2019144773A1
WO2019144773A1 PCT/CN2018/125297 CN2018125297W WO2019144773A1 WO 2019144773 A1 WO2019144773 A1 WO 2019144773A1 CN 2018125297 W CN2018125297 W CN 2018125297W WO 2019144773 A1 WO2019144773 A1 WO 2019144773A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
feature
customer service
model
service
Prior art date
Application number
PCT/CN2018/125297
Other languages
English (en)
French (fr)
Inventor
杨明晖
Original Assignee
阿里巴巴集团控股有限公司
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 阿里巴巴集团控股有限公司 filed Critical 阿里巴巴集团控股有限公司
Priority to JP2020536672A priority Critical patent/JP6991341B2/ja
Priority to SG11202006127PA priority patent/SG11202006127PA/en
Priority to EP18902124.9A priority patent/EP3719732A1/en
Priority to KR1020207018960A priority patent/KR102445992B1/ko
Publication of WO2019144773A1 publication Critical patent/WO2019144773A1/zh
Priority to US16/888,801 priority patent/US10977664B2/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
    • G06Q30/016—After-sales
    • 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
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/40—Processing or translation of natural language
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound

Definitions

  • the present specification relates to the field of data processing technology, and in particular, to a method and apparatus for a robot service to a manual customer service.
  • the robot customer service For some unconventional user problems, it is often difficult for the robot customer service to give a satisfactory response.
  • the most commonly used architecture of the customer service center is that the robot customer service and the manual customer service coexist.
  • the robot customer service first receives the user.
  • the manual customer service is transferred.
  • the artificial point that is, when the artificial customer service replaces the robot customer service to serve the user
  • the present specification provides a method for transferring a customer service to a manual customer service, including:
  • the confidence component assessment model is a machine learning model, using a session sample of the robot client and the user marked with the artificial point, and the user's State feature samples for training;
  • the specification also provides a device for transferring the customer service of the robot to the manual customer service, including:
  • a session feature obtaining unit configured to acquire a session feature from at least one round of conversation between the robot service and the user
  • a state feature acquiring unit configured to acquire a state feature of the user
  • a confidence component evaluation unit configured to input the conversation feature and the state feature into a confidence assessment model to obtain a current confidence score evaluation value;
  • the confidence component assessment model is a machine learning model, and the robot client and user marked with the artificial point are used The session sample and the user's state feature samples are trained;
  • the transfer judging unit is configured to transfer the user to the manual customer service when the current confidence score evaluation value satisfies the predetermined manual condition.
  • the computer device includes: a memory and a processor; the memory stores a computer program executable by the processor; and the method for executing the above-mentioned robot customer service to the manual customer service when the processor runs the computer program The steps described.
  • the present specification provides a computer readable storage medium having stored thereon a computer program, the computer program being executed by the processor, performing the steps described in the above method of robot service to manual customer service.
  • the session feature obtained from the conversation between the robot and the user and the state feature of the user are used as confidence input to evaluate the input of the machine learning model, and the current confidence score is obtained, and according to the current Confidence is used to determine whether it is necessary to transfer manual customer service. Since the user's status characteristics can often reflect its specific needs and urgency, applying the embodiments of this specification can improve the accuracy of the manual point and improve the service efficiency of the customer service center. At the same time, the user is more satisfied with the service.
  • FIG. 1 is a flow chart of a method for a robot service to a manual customer service in the embodiment of the present specification
  • FIG. 2 is a schematic structural view of a Wide and Deep model in an application example of the present specification
  • FIG. 3 is a flowchart of processing of a customer service in an application example of the present specification
  • FIG. 4 is a hardware structural diagram of a device running an embodiment of the present specification
  • FIG. 5 is a logical structural diagram of a device for changing a robot customer service to a manual customer service in the embodiment of the present specification
  • the interaction process between the user and the robot customer service can reflect the service effect of the robot customer service. For example, if the user repeats the same question in the interaction, or expresses dissatisfaction, it usually means that the robot customer service has poor ability to solve the user problem and needs manual customer service. Therefore, the user's conversation with the robot customer service is usually used as a basis for judging whether or not the manual customer service needs to be transferred.
  • the embodiment of the present specification proposes a new method of robot customer service to manual customer service, which uses the state feature of the user to describe the user's own factors, and is trained by using the session characteristics extracted by the robot customer service and the user session and the state characteristics of the user.
  • the confidence assessment model obtains the current confidence score, and transfers the manual customer service when the current confidence score meets the predetermined manual condition.
  • the confidence assessment model is based not only on the conversation process between the robot customer service and the user, but also based on the user state characteristics.
  • the user gives the current confidence assessment value to the degree and urgency of the manual customer service.
  • the embodiment of the present specification can give a more accurate judgment on the artificial point, which can not only improve the service efficiency of the customer service center, but also improve the service efficiency. User satisfaction with the service.
  • the embodiments of the present specification can be run on any computing and storage device, such as a mobile phone, a tablet, a PC (Personal Computer), a notebook, a server, etc., or can be operated by two or more.
  • the logical nodes of the device implement the functions in the embodiments of the present specification.
  • the machine learning model is established by using the session feature extracted from the session of the robot customer service and the user, and the state feature of the user, which is referred to as a confidence score evaluation model in this specification.
  • the session feature extracted from the session of the robot customer service and the user may be any feature that can be acquired by the NLP (Natural Language Processing) method based on the session of the robot customer service and the user, that is, the NLP feature.
  • NLP Natural Language Processing
  • it may be the degree of association between the user's question and the robot's answer, the number of questions and answers, the type of answer (the answer of the robot customer service is a statement or a question, the answer is the answer to a specific question, or the answer to the question, etc.), the number of repetitions of the answer, whether the user is Any number of features such as changing labor and whether the user is explaining his or her own problem is proposed.
  • the determination of the session feature and the specific manner of obtaining the session feature from the session can be implemented by referring to the prior art, and will not be described again.
  • the user's behavior record characteristics include access records, and/or function usage records of all products within the scope of the customer service center consulting service within a predetermined time period. For example, it can be which pages are opened in an App (application) in the past 72 hours, which function operations are performed, and the like.
  • the user's behavior record characteristics reflect the user's use of the product being consulted in a short period of time; if a user frequently uses the customer service function during that time period and queries the same knowledge point, the user is likely to encounter a robot customer service solution. If you can't solve the problem, you need manual service. If a user tries a function for many times, the user's demand for manual customer service is often more urgent.
  • the second example the user's business status characteristics.
  • the user's business status feature reflects the information of the account opened by the user on the consulted product, and may include the service opening status of the user account, the account authentication status, the account login status, and/or the account abnormal status.
  • the user's business status feature can often reflect the user's urgency to solve the problem. If the user's account is in a state of “frozen” (an account abnormal state) or “off-site login” (an account login status), the user is likely to In the case of stolen and cheated, there will be an urgent need for manual customer service.
  • the identity information of the user is information of the user as a natural person, and may include the gender, age, resident territory, and/or education level of the user. Users with different identity information characteristics usually accept different degrees of robot customer service. For example, young users and highly educated users are more accustomed to the question and answer mode of robot customer service. Older users and lower education users are more inclined to artificial customer service.
  • the input of the confidence assessment model includes obtaining session characteristics from the user's session with the user and the user's state characteristics, the output of which includes a confidence score.
  • the confidence assessment model is trained by the robot client and the user's session sample marked with the artificial point, and the user's state feature sample.
  • the manual point is the appropriate one to replace the robot customer service with the manual customer service during the conversation with the user. Time.
  • the manual point and session features may be marked in the session sample and the confidence score evaluation model may be input for training; or the artificial point may be marked only in the session sample, and the session sample is automatically performed by the computer program.
  • the session features are obtained, and then the confidence score evaluation model is input for training.
  • the machine learning algorithm used in the confidence assessment model can be selected according to the characteristics of the actual application scenario, and is not limited.
  • the confidence assessment model can be a machine learning model based on support vector machine, such as SVC (Support Vector Machine), etc.; it can be a tree-based machine learning model, such as GBDT (Gradient Boosting Decision Tree). ); can be a linear model, such as LR (Logistic Regression), etc.; can also be a neural network model, such as DNN (Deep Neural Networks), RNN (Recurrent Neural Networks), CNN (Convolutional Neural Networks, Convolutional Neural Networks), etc.
  • the Wide and Deep model is used to build a confidence assessment model.
  • the Wide and Deep model includes a linear submodel and a deep neural network submodel, using a training model that combines a deep neural network submodel with a shallow linear submodel.
  • the training error of the overall model is simultaneously fed back to the linear submodel and the deep neural network.
  • the parameter update is performed in the sub-model, and the parameters of the two sub-models are optimized at the same time, so that the prediction ability of the overall Wide and Deep model is optimal.
  • the conversation feature directly reflects the user's demand for manual customer service, which is a strong correlation feature; and the user's state feature is not directly related to the question and answer process, but only indirectly affects the user's demand for manual customer service. , is a weak correlation feature.
  • different sub-models in the Wide and Deep learning model can be used to deal with these two types of features. Specifically, the session feature is used as the input of the linear sub-model, and the output vector of the linear sub-model is obtained by the linear sub-model; the state feature of the user is used as the input of the deep neural network sub-model, and processed by the deep neural network sub-model. The output vector of the deep neural network submodel is obtained; then the output vectors of the two submodels are put together and the output is calculated through a neuron.
  • the output can be regarded as the time point.
  • the possibility of a manual point the confidence value is evaluated as a floating point number between 0 and 1).
  • the flow of the method of the robot customer service to the manual customer service is as shown in FIG. 1 .
  • Step 110 Acquire a session feature from at least one round of conversation between the robot service and the user.
  • the confidence assessment model can be used in the real-time conversation between the robot customer service and the user to determine whether the user needs to switch to the manual customer service after the user speaks in the conversation.
  • one user speaks, or one robot customer service speaks and one user speaks one round.
  • the first round of the robot customer service and the user's conversation is the user's speech
  • the second and subsequent rounds are a robot customer service speech and a user speech.
  • Each round ends with a user's speech, which is the point in time at which the robotic customer service can be replaced by manual customer service, or a possible artificial point.
  • One or more rounds of the robotic customer service session with the user can be used as the basis for acquiring session features. For example, the destination session number of the predetermined number of times before the current time point is used as the basis for acquiring the session feature. If the session round that has been performed at the current time point is less than the predetermined number of rounds, the entire session is used as the basis for acquiring the session feature; for example, it will always be already The entire session is performed as the basis for acquiring session characteristics.
  • Step 120 Acquire a state feature of the user.
  • the state characteristics of the user can be read from a predetermined network location, a predetermined database table, and the like that save the user state feature, and are not limited.
  • step 110 there is no timing relationship between step 110 and step 120.
  • the session of the robot customer service and the user is usually updated continuously as the session process continues, and the user's status feature generally does not change during a customer service process, so step 110 may be performed multiple times during a customer service process, and the steps are performed. 120 is usually executed once.
  • step 130 the conversation feature and the state feature are input into the confidence assessment model to obtain the current confidence score.
  • Step 140 When the current confidence score evaluation value meets the predetermined manual condition, the user is transferred to the manual customer service.
  • the predetermined manual condition may be that the current confidence score is greater than or less than a predetermined confidence threshold, depending on whether the current confidence score in the actual application scenario is higher, which represents a stronger demand for manual customer service, or a weaker manual service.
  • the predetermined confidence threshold can be determined by comprehensively considering the degree of fitting of the confidence assessment sub-model and the session sample after training, and the ratio of the number of users and manual customer service in the actual application scenario. It is also possible to set a certain standard, and the program automatically determines the predetermined confidence threshold according to the set criteria.
  • the session sample may be input into the trained confidence assessment model to obtain a sample confidence score corresponding to the artificial point in the session sample; a specific series of different predetermined confidence threshold values are set, and the calculation is different when the selection is different.
  • the predetermined confidence value of the value is divided into thresholds, the coverage and accuracy of the sample confidence score are evaluated, and the criteria for coverage and accuracy are set. The best coverage and accuracy corresponding to the evaluation criteria are used as the predetermined confidence.
  • the threshold is divided.
  • the user's state feature is used to describe the user's own factors
  • the session feature acquired from the robot and the user's session and the user's state feature are used as confidence to evaluate the input of the machine learning model, and the current Confidence assessment value, when the current confidence assessment value meets the predetermined manual condition, the manual customer service is transferred.
  • the confidence assessment model is based on the degree and urgency of the user's demand for manual customer service reflected by the user state characteristics, the current confidence assessment is given. Values, after applying the embodiments of the present specification, the accuracy of the manual point can be increased, which not only improves the service efficiency of the customer service center, but also improves the user's satisfaction with the service.
  • a customer service center of a third party payment platform provides online services to users who use their client applications.
  • the customer center technicians use the conversation features extracted from the robot customer service and the user's conversation, as well as the user's state characteristics, to establish the Wide and Deep model as a confidence assessment model.
  • the user's status characteristics include the user's behavior record characteristics, business status characteristics, and identity information characteristics.
  • the structure of the Wide and Deep model is shown in Figure 2.
  • the session feature is used as the input to the linear submodel, and the user's state feature is used as the input to the deep neural network submodel.
  • the neural layers of the deep neural network sub-model in the Wide and Deep model are Dense (tight) neural layers, that is, the neural layer uses several Dense neurons for data processing.
  • the deep neural network sub-model uses the Dense neural layer to process the business state features, the Dense neural layer to process the identity information features, and the LSTM (Long Short-Term Memory) neural layer processing behavior record characteristics;
  • the output of the three neural layers is integrated into the output vector of a deep neural network submodel by the Dense neural layer, and the output vector of the two submodels is integrated into the output of the entire model by a Dense neural layer. , that is, the confidence assessment value.
  • the inventors of the present application found in the experiment that the nerve layer structure shown in Fig. 2 can achieve a better effect.
  • the technician points out the artificial points in the history records of several robot customer service and user sessions, uses them as session samples, and uses the state features of these users as state feature samples.
  • the program automatically extracts the session features from the session samples
  • the program automatically extracts the session features from the session samples.
  • the session feature and state feature samples of the session sample are input into the Wide and Deep model for training.
  • Step 305 receiving input from a user.
  • step 310 the robot customer service gives a reply according to the user's question.
  • Step 315 Acquire a behavior record feature, a service state feature, and an identity information feature of the user.
  • step 320 it is determined whether the user's next input is received. If not received, the process ends; if received, step 325 is performed.
  • step 325 the session feature is extracted from the robot service and the user in all sessions of the service.
  • step 330 the session feature and the user's state feature are input into the trained completed Wide and Deep confidence assessment model, and the output of the model is the current confidence score.
  • step 335 it is determined whether the current confidence score evaluation value satisfies the predetermined manual condition. If yes, step 340 is performed; if not, step 345 is performed.
  • step 340 the manual customer service is transferred, and the process ends.
  • step 345 the user response is given by the robot service, and the process proceeds to step 320.
  • the embodiment of the present specification further provides a device for changing the robot customer service to the manual customer service.
  • the device can be implemented by software, or can be realized by hardware or a combination of hardware and software.
  • the CPU Central Process Unit
  • the device in which the device of the robot service to the manual customer service usually includes other hardware such as a chip for transmitting and receiving wireless signals, and/or Other hardware such as boards that implement network communication functions.
  • FIG. 5 is a schematic diagram of a device for changing a human service of a robot to a manual customer service according to an embodiment of the present disclosure, including a session feature acquisition unit, a state feature acquisition unit, a confidence component, and a transfer determination unit, wherein: a session feature acquisition unit is used.
  • the session feature is obtained from at least one round of conversation between the robot customer service and the user;
  • the state feature acquisition unit is configured to acquire the state feature of the user;
  • the confidence component evaluation unit is configured to input the conversation feature and the state feature into the confidence assessment model to obtain current confidence.
  • the evaluation value is a machine learning model, which is trained by a robot client and a user's session sample marked with an artificial point, and a user's state feature sample; the transfer judgment unit is used for the current confidence assessment When the value meets the predetermined manual condition, the user is transferred to the manual customer service.
  • the confidence score evaluation model is a depth and breadth Wide and Deep model
  • the Wide and Deep model includes a linear submodel and a deep neural network submodel, with session features as input to the linear submodel, with state features as Input to the deep neural network submodel.
  • the state feature may include at least one of: a user's behavior record feature, a service state feature, and an identity information feature; the deep neural network sub-model adopts the Dense tight neural layer to process the business state feature, and adopts the Dense neural layer.
  • the identity information characteristics are processed, and the long-short-term memory network LSTM neural layer is used to process the behavior record characteristics.
  • the session feature is a natural language processing NLP feature, including one or more of the following: a user's relevance to the robot answer, the number of questions and answers, and the answer type.
  • the status feature includes at least one of: a user's behavior record feature, a service state feature, and an identity information feature; and the behavior record feature includes at least one of: a user's access record, operation within a predetermined time period Recording;
  • the service status feature includes at least one of the following: a service activation status of the user account, an account authentication status, an account login status, and an account abnormal status;
  • the identity information feature includes at least one of the following: a user's gender, age, and Resident area.
  • the confidence assessment model includes: a machine learning model based on a support vector machine, a machine learning model based on a tree, a linear model, and a neural network model.
  • the predetermined manual condition includes: the current confidence score evaluation value is greater than or less than a predetermined confidence score threshold.
  • Embodiments of the present specification provide a computer device including a memory and a processor.
  • the computer stores a computer program executable by the processor; and when the processor runs the stored computer program, the processor executes the steps of the method of the robot service to the manual customer service in the embodiment of the present specification.
  • the various steps of the robot customer service to manual customer service please refer to the previous content, and will not repeat.
  • Embodiments of the present specification provide a computer readable storage medium having stored thereon computer programs that, when executed by a processor, perform various steps of a method of robotic customer service to manual customer service in the embodiments of the present specification .
  • a processor For a detailed description of the various steps of the robot customer service to manual customer service, please refer to the previous content, and will not repeat.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • the memory may include non-persistent memory, random access memory (RAM), and/or non-volatile memory in a computer readable medium, such as read only memory (ROM) or flash memory.
  • RAM random access memory
  • ROM read only memory
  • Memory is an example of a computer readable medium.
  • Computer readable media includes both permanent and non-persistent, removable and non-removable media.
  • Information storage can be implemented by any method or technology.
  • the information can be computer readable instructions, data structures, modules of programs, 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 disk read only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, Magnetic tape cartridges, magnetic tape storage or other magnetic storage devices or any other non-transportable media can be used to store information that can be accessed by a computing device.
  • computer readable media does not include temporary storage of computer readable media, such as modulated data signals and carrier waves.
  • embodiments of the present specification can be provided as a method, system, or computer program product.
  • embodiments of the present specification can take the form of an entirely hardware embodiment, an entirely software embodiment or a combination of software and hardware.
  • embodiments of the present specification can take the form of a computer program product embodied on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer usable program code embodied therein. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Educational Administration (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Manipulator (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种机器人客服转人工客服的方法，包括：从机器人客服与用户的至少一轮会话中获取会话特征（110）；获取用户的状态特征（120）；将所述会话特征和状态特征输入信心分评估模型，得到当前信心分评估值（130）；所述信心分评估模型为机器学习模型，采用标记有出人工点的机器人客户与用户的会话样本、以及用户的状态特征样本进行训练；在当前信心分评估值满足预定出人工条件时，将用户转接人工客服（140）。

Description

机器人客服转人工客服的方法和装置 技术领域
本说明书涉及数据处理技术领域，尤其涉及一种机器人客服转人工客服的方法和装置。
背景技术
随着互联网的发展，基于人工智能技术的虚拟机器人在企业用户服务领域的应用越来越广泛。机器人客服不需要休息，可以更加快速和标准化的响应用户的问题，以语音对话或文字聊天的形式与用户进行沟通，将人工客服从大量重复性问答中解放出来。
对一些非常规的用户问题，机器人客服往往难以给出令用户满意的答复。目前客服中心最为常用的架构是机器人客服与人工客服并存，缺省由机器人客服先接待用户，当机器人客服不解决用户的问题时，转接人工客服。出人工点(即以人工客服代替机器人客服为用户服务的时点)是否适当对客服中心的运营效率和用户的满意度都有重要的影响。
发明内容
有鉴于此，本说明书提供一种机器人客服转人工客服的方法，包括：
从机器人客服与用户的至少一轮会话中获取会话特征；
获取用户的状态特征；
将所述会话特征和状态特征输入信心分评估模型，得到当前信心分评估值；所述信心分评估模型为机器学习模型，采用标记有出人工点的机器人客户与用户的会话样本、以及用户的状态特征样本进行训练；
在当前信心分评估值满足预定出人工条件时，将用户转接人工客服。
本说明书还提供了一种机器人客服转人工客服的装置，包括：
会话特征获取单元，用于从机器人客服与用户的至少一轮会话中获取会话特征；
状态特征获取单元，用于获取用户的状态特征；
信心分评估单元，用于将所述会话特征和状态特征输入信心分评估模型，得到当前信心分评估值；所述信心分评估模型为机器学习模型，采用标记有出人工点的机器人客户与用户的会话样本、以及用户的状态特征样本进行训练；
转接判断单元，用于在当前信心分评估值满足预定出人工条件时，将用户转接人工客服。
本说明书提供的一种计算机设备，包括：存储器和处理器；所述存储器上存储有可由处理器运行的计算机程序；所述处理器运行所述计算机程序时，执行上述机器人客服转人工客服的方法所述的步骤。
本说明书提供的一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器运行时，执行上述机器人客服转人工客服的方法所述的步骤。
由以上技术方案可见，本说明书的实施例中，以从机器人与用户的会话中获取的会话特征和用户的状态特征作为信心分评估机器学习模型的输入，得到当前信心分评估值，并根据当前信心分评估值来判断是否需要转接人工客服，由于用户的状态特征常常能够体现其具体需求和迫切程度，应用本说明书的实施例后可以提高出人工点的准确性，在提升客服中心服务效率的同时使用户对服务更加满意。
附图说明
图1是本说明书实施例中一种机器人客服转人工客服的方法的流程图；
图2是本说明书应用示例中一种Wide and Deep模型的结构示意图；
图3是本说明书应用示例中一次客户服务的处理流程图；
图4是运行本说明书实施例的设备的一种硬件结构图；
图5是本说明书实施例中一种机器人客服转人工客服的装置的逻辑结构图
具体实施方式
在用户接受机器人客服服务的过程中，用户与机器人客服的交互过程能够体现机器人客服的服务效果。例如，如果用户在交互中重复着同样的问题，或者表达了不满情绪，通常意味着机器人客服对用户问题的解决能力欠佳，需要人工客服处理。因此，用户与机器人客服的会话通常会被用来作为判断是否需要转接人工客服的依据。
另一方面，用户的自身因素也会影响对人工客户的需求程度。例如，账户被盗的用户通常都有较强的人工客服需求，而查询账单的用户一般有更好的耐心与机器人客户沟通。此外，不同年龄、职业、教育背景的用户，对机器人客服的接受度也不尽相同，青年用户能较快理解机器人客服的回答，而老年用户常常需要人工客服的详细指导。
本说明书的实施例提出一种新的机器人客服转人工客服的方法，采用用户的状态特征来描述用户的自身因素，利用以机器人客服与用户会话中提取的会话特征和用户的状态特征训练完成的信心分评估模型，得到当前信心分评估值，在当前信心分评估值满足预定出人工条件时转接人工客服，由于信心分评估模型不仅基于机器人客服与用户的会话过程，并且基于用户状态特征体现的用户对人工客服的需求程度和迫切程度来给出当前信心分评估值，本说明书的实施例能够对出人工点给出更加准确的判断，不仅能够提高客服中心的服务效率，而且还能够提高用户对服务的满意程度。
本说明书的实施例可以运行在任何具有计算和存储能力的设备上，如手机、平板电脑、PC(Personal Computer，个人电脑)、笔记本、服务器等设备；还可以由运行在两个或两个以上设备的逻辑节点来实现本说明书实施例中的各项功能。
本说明书的实施例中，采用从机器人客服与用户的会话中提取的会话特征、和用户的状态特征来建立机器学习模型，本说明书中称之为信心分评估模型。其中，从机器人客服与用户的会话中提取的会话特征可以是以机器人客服与用户的会话为基础，采用NLP(Natural Language Processing，自然语言处理)方法能够获取的任何特征，即NLP特征。具体而言，可以是用户提问与机器人回答的关联度、问答轮数、答案类型(机器人客服的答案是陈述还是提问、答案是具体问题的答案还是兜底答案等等)、答案重复次数、用户是否提出换人工、用户是否在解释自己的问题等特征中的任意个数。会话特征的确定和从会话中获取会话特征的具体方式均可参照现有技术实现，不再赘述。
除与机器人客服的会话以外，其他能够体现用户当前对人工客服的需求程度以及解决问题的迫切程度的用户信息都可以用来作为用户的状态特征，本说明书的实施例对具体的状态特征以及状态特征的数量均不做限定。以下举例说明。
第一个例子：用户的行为记录特征。用户的行为记录特征包括用户在预定时间段内对客服中心咨询服务范围内所有产品的访问记录、和/或功能使用记录等。例如，可以是过去72小时内在某个App(应用程序)内打开哪些页面、进行哪些功能操作等。用户的行为记录特征反映了短期内用户对被咨询产品的使用情况；如果某个用户在该时间段 内频繁使用客服功能，并且查询了同一个知识点，那么该用户很可能遇到了机器人客服解决不了的问题，需要人工服务；如果某个用户尝试了某个功能操作很多遍，该用户对人工客服有的需求往往更为迫切。
第二个例子：用户的业务状态特征。用户的业务状态特征体现了用户在被咨询产品上所开设账户的信息，可以包括用户账户的业务开通状态、账户认证状态、账户登录状态、和/或账户异常状态等。用户的业务状态特征常常能够体现用户解决问题的迫切程度，如果用户账户处在“被冻结”(一种账户异常状态)、“异地登录”(一种账户登录状态)等状态时，用户很可能遇到了被盗被骗等问题，对人工客服会有很迫切的需求。
第三个例子：用户的身份信息特征。用户的身份信息特征是用户作为自然人的信息，可以包括用户的性别、年龄、常驻地域、和/或教育程度等。不同身份信息特征的用户对机器人客服的接受程度通常不同，如年轻的用户、教育程度高的用户更习惯机器人客服的问答模式，年长的用户、教育程度较低的用户对人工客服更加偏爱。
如前所述，信心分评估模型的输入包括从机器人客服与用户的会话中获取会话特征、以及用户的状态特征，其输出包括信心分评估值。信心分评估模型采用标记有出人工点的机器人客户与用户的会话样本、以及用户的状态特征样本进行训练，其中，出人工点是在与用户的会话过程中，以人工客服代替机器人客服的适当时点。在训练信心分评估模型时，可以在会话样本中标记了出人工点和会话特征后输入信心分评估模型进行训练；也可以只在会话样本中标记出人工点，由计算机程序自动对会话样本进行NLP处理后得到会话特征，再输入信心分评估模型进行训练。
信心分评估模型所采用的机器学习算法可以根据实际应用场景的特点来选择，不做限定。信心分评估模型可以是基于支持向量机的机器学习模型，如SVC(Support Vector Machine，支持向量机)等；可以是基于树型的机器学习模型，如GBDT(Gradient Boosting Decision Tree，梯度提升决策树)等；可以是线性模型，如LR(Logistic Regression，逻辑回归)等；也可以是神经网络模型，如DNN(Deep Neural Networks，深度神经网络)、RNN(Recurrent Neural Networks，循环神经网络)、CNN(Convolutional Neural Networks，卷积神经网络)等。
在一种实现方式中，采用Wide and Deep(深度和广度)模型来建立信心分评估模型。Wide and Deep模型包括线性子模型和深度神经网络子模型，采用将深度神经网络子模型与浅层线性子模型相结合的训练模式。通过结合线性子模型的记忆能力 (memorization)和深度神经网络子模型的泛化能力(generalization)，并且采用联合训练(joint training)，将整体模型的训练误差同时反馈到线性子模型和深度神经网络子模型中进行参数更新，同时优化2个子模型的参数，从而达到整体Wide and Deep模型的预测能力最优。
在用户与机器人客服的问答过程中，会话特征直接体现了用户对人工客服的需求，属于强相关特征；而用户的状态特征与问答过程没有直接的关联，只是间接影响着用户对人工客服的需求，属于弱相关特征。为了体现这两类特征对结果的影响，可以采用Wide and Deep learning模型中不同的子模型来处理这两类特征。具体而言，以会话特征作为线性子模型的输入，经线性子模型处理后得到线性子模型的输出向量；以用户的状态特征作为深度神经网络子模型的输入，经深度神经网络子模型处理后得到深度神经网络子模型的输出向量；然后将两个子模型的输出向量拼起来，经过一个神经元计算输出。
在Wide and Deep learning模型训练阶段，可以将是否标记为出人工点作为输出(假设信心分评估值分别为1或0)；在使用训练后的模型进行预测时，输出可以视为该时点是出人工点的可能性(信心分评估值为0到1之间的浮点数)。
本说明书的实施例中，机器人客服转人工客服的方法的流程如图1所示。
步骤110，从机器人客服与用户的至少一轮会话中获取会话特征。
在信心分评估模型训练完成后，可以在实时的机器人客服与用户的会话过程中，利用信心分评估模型来判断在会话中用户发言后的时点，是否需要切换为人工客服。
本说明书的实施例中，以一次用户发言、或者一次机器人客服发言和一次用户发言为一个轮次。通常机器人客服与用户的会话的第一轮是用户发言，第二轮及后续轮次是一次机器人客服发言和一次用户发言。每个轮次以用户发言来结束，该时点也即是可以由人工客服代替机器人客服的时点，或者说是可能的出人工点。
可以将机器人客服与用户会话的一个到多个轮次来作为获取会话特征的基础。例如将当前时点前预定轮次数目的会话作为获取会话特征的基础，如果当前时点已经进行的会话轮次小于预定轮次数目，则整个会话作为获取会话特征的基础；再如，始终将已经进行的整个会话作为获取会话特征的基础。
步骤120，获取用户的状态特征。
根据实际应用场景中客服中心和所咨询服务产品服务端的具体实现，可以从保存用 户状态特征的预定网络位置、预定数据库表等处读取到用户的状态特征，不做限定。
需要说明的是，步骤110和步骤120之间没有时序关系。机器人客服与用户的会话通常随着会话过程的继续不断更新，而用户的状态特征则一般在一次客户服务过程中不会发生变化，因此步骤110在一次客户服务过程中可能执行多次，而步骤120则通常执行一次。
步骤130，将会话特征和状态特征输入信心分评估模型，得到当前信心分评估值。
将从当前时点前若干轮次的机器人客服与用户的会话中提取的会话特征、用户的状态特征输入到信心分评估模型中，得到在会话中用户最后一次发言后的当前时点(即当前轮次会话的用户发言后时点)的信心分评估值。
步骤140，在当前信心分评估值满足预定出人工条件时，将用户转接人工客服。
如果当前信心分评估值满足预定出人工条件，则认为当前时点需要人工客服介入，将用户转接至人工客服。
预定出人工条件可以是当前信心分评估值大于或小于预定信心分阈值，视实际应用场景中当前信心分评估值更高是代表更强的对人工客服的需求程度，还是更弱的对人工客服的需求程度。预定信心分阈值可以综合考虑训练后信心评估分模型与会话样本的拟合程度、实际应用场景中的用户与人工客服的数量比例等因素来确定。也可以由设置一定的标准，由程序根据所设置的标准自动确定预定信心分阈值。例如，可以将会话样本输入到训练后的信心评估模型中，得到会话样本中对应于出人工点的样本信心分评估值；设定一系列不同的预定信心分阈值的具体数值，计算当选择不同数值的预定信心分阈值时样本信心分评估值的覆盖率和准确率，设定针对覆盖率和准确率的评判标准，按照评判标准的评价最好的覆盖率和准确率对应的数值作为预定信心分阈值。
可见，本说明书的实施例中，采用用户的状态特征来描述用户的自身因素，以从机器人与用户的会话中获取的会话特征和用户的状态特征作为信心分评估机器学习模型的输入，得到当前信心分评估值，在当前信心分评估值满足预定出人工条件时转接人工客服，由于信心分评估模型基于用户状态特征体现的用户对人工客服的需求程度和迫切程度来给出当前信心分评估值，应用本说明书的实施例后可以增加出人工点的准确性，不仅能够提高客服中心的服务效率，而且还能够提高用户对服务的满意程度。
上述对本说明书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要求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。
在本说明书的一个应用示例中，第三方支付平台的客服中心为使用其客户端App的用户提供在线服务。客户中心的技术人员采用从机器人客服与用户的会话中提取的会话特征、以及用户的状态特征，建立Wide and Deep模型，作为信心分评估模型。本应用示例中，用户的状态特征包括用户的行为记录特征、业务状态特征和身份信息特征。
Wide and Deep模型的结构如图2所示，以会话特征作为线性子模型的输入，以用户的状态特征作为深度神经网络子模型的输入。通常Wide and Deep模型中深度神经网络子模型的各层神经层均为Dense(紧密)神经层，即该神经层采用若干个Dense神经元进行数据处理。本应用示例中，深度神经网络子模型采用Dense神经层处理业务状态特征，采用Dense神经层处理身份信息特征，采用LSTM(Long Short-Term Memory，长短期记忆网络)神经层处理行为记录特征；上述三个神经层的输出经Dense神经层将三种用户状态特征的处理结果综合为一个深度神经网络子模型的输出向量，再由一个Dense神经层将两个子模型的输出向量综合为整个模型的输出，即信心分评估值。经本申请的发明人在试验中发现，图2所示的神经层结构能够达到更好的效果。
技术人员在若干机器人客服与用户的会话的历史记录中标注出人工点后，将其作为会话样本，并将这些用户的状态特征作为状态特征样本，由程序自动从会话样本中提取出会话特征后，将会话样本的会话特征和状态特征样本输入到Wide and Deep模型中进行训练。
当客服中心接到用户的服务请求时，其处理过程如图3所示。
步骤305，接收用户的输入。
步骤310，由机器人客服根据用户的问题给出回复。
步骤315，获取用户的行为记录特征、业务状态特征和身份信息特征。
步骤320，判断是否收到用户的下一次输入，如果未收到，流程结束；如果收到，执行步骤325。
步骤325，从机器人客服与用户在本次服务的所有会话中提取会话特征。
步骤330，将会话特征和用户的状态特征输入到训练完成的Wide and Deep信心分评估模型中，模型的输出为当前信心分评估值。
步骤335，判断当前信心分评估值是否满足预定出人工条件，如果是，执行步骤340；如果否，执行步骤345。
步骤340，转接人工客服，流程结束。
步骤345，由机器人客服给出用户回复，转步骤320。
与上述流程实现对应，本说明书的实施例还提供了一种机器人客服转人工客服的装置。该装置均可以通过软件实现，也可以通过硬件或者软硬件结合的方式实现。以软件实现为例，作为逻辑意义上的装置，是通过所在设备的CPU(Central Process Unit，中央处理器)将对应的计算机程序指令读取到内存中运行形成的。从硬件层面而言，除了图4所示的CPU、内存以及存储器之外，机器人客服转人工客服的装置所在的设备通常还包括用于进行无线信号收发的芯片等其他硬件，和/或用于实现网络通信功能的板卡等其他硬件。
图5所示为本说明书实施例提供的一种机器人客服转人工客服的装置，包括会话特征获取单元、状态特征获取单元、信心分评估单元和转接判断单元，其中：会话特征获取单元用于从机器人客服与用户的至少一轮会话中获取会话特征；状态特征获取单元用于获取用户的状态特征；信心分评估单元用于将所述会话特征和状态特征输入信心分评估模型，得到当前信心分评估值；所述信心分评估模型为机器学习模型，采用标记有出人工点的机器人客户与用户的会话样本、以及用户的状态特征样本进行训练；转接判断单元用于在当前信心分评估值满足预定出人工条件时，将用户转接人工客服。
一个例子中，所述信心分评估模型为深度和广度Wide and Deep模型，所述Wide and Deep模型包括线性子模型和深度神经网络子模型，以会话特征作为线性子模型的输入，以状态特征作为深度神经网络子模型的输入。
上述例子中，所述状态特征可以包括以下至少一项：用户的行为记录特征、业务状态特征和身份信息特征；所述深度神经网络子模型采用Dense紧密神经层处理业务状态特征，采用Dense神经层处理身份信息特征，采用长短期记忆网络LSTM神经层处理行为记录特征。
可选的，所述会话特征为自然语言处理NLP特征，包括以下的一项到多项：用户提 问与机器人回答的关联度、问答轮数、答案类型。
可选的，所述状态特征包括以下至少一项：用户的行为记录特征、业务状态特征和身份信息特征；所述行为记录特征包括以下至少一项：用户在预定时间段内的访问记录、操作记录；所述业务状态特征包括以下至少一项：用户账户的业务开通状态、账户认证状态、账户登录状态、账户异常状态；所述身份信息特征包括以下至少一项：用户的性别、年龄、常驻地域。
可选的，所述信心分评估模型包括：基于支持向量机的机器学习模型、基于树型的机器学习模型、线性模型、神经网络模型。
可选的，所述预定出人工条件包括：当前信心分评估值大于或小于预定信心分阈值。
本说明书的实施例提供了一种计算机设备，该计算机设备包括存储器和处理器。其中，存储器上存储有能够由处理器运行的计算机程序；处理器在运行存储的计算机程序时，执行本说明书实施例中机器人客服转人工客服的方法的各个步骤。对机器人客服转人工客服的方法的各个步骤的详细描述请参见之前的内容，不再重复。
本说明书的实施例提供了一种计算机可读存储介质，该存储介质上存储有计算机程序，这些计算机程序在被处理器运行时，执行本说明书实施例中机器人客服转人工客服的方法的各个步骤。对机器人客服转人工客服的方法的各个步骤的详细描述请参见之前的内容，不再重复。
以上所述仅为本说明书的较佳实施例而已，并不用以限制本申请，凡在本申请的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本申请保护的范围之内。
在一个典型的配置中，计算设备包括一个或多个处理器(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，本说明书的实施例可提供为方法、系统或计算机程序产品。因此，本说明书的实施例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，本说明书的实施例可采用在一个或多个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。

Claims (16)

  1. 一种机器人客服转人工客服的方法，包括：
    从机器人客服与用户的至少一轮会话中获取会话特征；
    获取用户的状态特征；
    将所述会话特征和状态特征输入信心分评估模型，得到当前信心分评估值；所述信心分评估模型为机器学习模型，采用标记有出人工点的机器人客户与用户的会话样本、以及用户的状态特征样本进行训练；
    在当前信心分评估值满足预定出人工条件时，将用户转接人工客服。
  2. 根据权利要求1所述的方法，所述信心分评估模型为深度和广度Wide and Deep模型，所述Wide and Deep模型包括线性子模型和深度神经网络子模型，以会话特征作为线性子模型的输入，以状态特征作为深度神经网络子模型的输入。
  3. 根据权利要求2所述的方法，所述状态特征包括以下至少一项：用户的行为记录特征、业务状态特征和身份信息特征；
    所述深度神经网络子模型采用Dense紧密神经层处理业务状态特征，采用Dense神经层处理身份信息特征，采用长短期记忆网络LSTM神经层处理行为记录特征。
  4. 根据权利要求1所述的方法，所述会话特征为自然语言处理NLP特征，包括以下的一项到多项：用户提问与机器人回答的关联度、问答轮数、答案类型。
  5. 根据权利要求1所述的方法，所述状态特征包括以下至少一项：用户的行为记录特征、业务状态特征和身份信息特征；
    所述行为记录特征包括以下至少一项：用户在预定时间段内的访问记录、操作记录；
    所述业务状态特征包括以下至少一项：用户账户的业务开通状态、账户认证状态、账户登录状态、账户异常状态；
    所述身份信息特征包括以下至少一项：用户的性别、年龄、常驻地域。
  6. 根据权利要求1所述的方法，所述信心分评估模型包括：基于支持向量机的机器学习模型、基于树型的机器学习模型、线性模型、神经网络模型。
  7. 根据权利要求1所述的方法，所述预定出人工条件包括：当前信心分评估值大于或小于预定信心分阈值。
  8. 一种机器人客服转人工客服的装置，包括：
    会话特征获取单元，用于从机器人客服与用户的至少一轮会话中获取会话特征；
    状态特征获取单元，用于获取用户的状态特征；
    信心分评估单元，用于将所述会话特征和状态特征输入信心分评估模型，得到当前 信心分评估值；所述信心分评估模型为机器学习模型，采用标记有出人工点的机器人客户与用户的会话样本、以及用户的状态特征样本进行训练；
    转接判断单元，用于在当前信心分评估值满足预定出人工条件时，将用户转接人工客服。
  9. 根据权利要求8所述的装置，所述信心分评估模型为深度和广度Wide and Deep模型，所述Wide and Deep模型包括线性子模型和深度神经网络子模型，以会话特征作为线性子模型的输入，以状态特征作为深度神经网络子模型的输入。
  10. 根据权利要求9所述的装置，所述状态特征包括以下至少一项：用户的行为记录特征、业务状态特征和身份信息特征；
    所述深度神经网络子模型采用Dense紧密神经层处理业务状态特征，采用Dense神经层处理身份信息特征，采用长短期记忆网络LSTM神经层处理行为记录特征。
  11. 根据权利要求8所述的装置，所述会话特征为自然语言处理NLP特征，包括以下的一项到多项：用户提问与机器人回答的关联度、问答轮数、答案类型。
  12. 根据权利要求8所述的装置，所述状态特征包括以下至少一项：用户的行为记录特征、业务状态特征和身份信息特征；
    所述行为记录特征包括以下至少一项：用户在预定时间段内的访问记录、操作记录；
    所述业务状态特征包括以下至少一项：用户账户的业务开通状态、账户认证状态、账户登录状态、账户异常状态；
    所述身份信息特征包括以下至少一项：用户的性别、年龄、常驻地域。
  13. 根据权利要求8所述的装置，所述信心分评估模型包括：基于支持向量机的机器学习模型、基于树型的机器学习模型、线性模型、神经网络模型。
  14. 根据权利要求8所述的装置，所述预定出人工条件包括：当前信心分评估值大于或小于预定信心分阈值。
  15. 一种计算机设备，包括：存储器和处理器；所述存储器上存储有可由处理器运行的计算机程序；所述处理器运行所述计算机程序时，执行如权利要求1到7任意一项所述的步骤。
  16. 一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器运行时，执行如权利要求1到7任意一项所述的步骤。
PCT/CN2018/125297 2018-01-26 2018-12-29 机器人客服转人工客服的方法和装置 WO2019144773A1 (zh)

Priority Applications (5)

Application Number Priority Date Filing Date Title
JP2020536672A JP6991341B2 (ja) 2018-01-26 2018-12-29 ロボットの顧客サービスから人間の顧客サービスへの移行のための方法および装置
SG11202006127PA SG11202006127PA (en) 2018-01-26 2018-12-29 Method and apparatus for transferring from robot customer service to human customer service
EP18902124.9A EP3719732A1 (en) 2018-01-26 2018-12-29 Method and apparatus for transferring from robot customer service to human customer service
KR1020207018960A KR102445992B1 (ko) 2018-01-26 2018-12-29 로봇 고객 서비스로부터 사람 고객 서비스로 전환하기 위한 방법 및 장치
US16/888,801 US10977664B2 (en) 2018-01-26 2020-05-31 Method and apparatus for transferring from robot customer service to human customer service

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810076926.7A CN108363745B (zh) 2018-01-26 2018-01-26 机器人客服转人工客服的方法和装置
CN201810076926.7 2018-01-26

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/888,801 Continuation US10977664B2 (en) 2018-01-26 2020-05-31 Method and apparatus for transferring from robot customer service to human customer service

Publications (1)

Publication Number Publication Date
WO2019144773A1 true WO2019144773A1 (zh) 2019-08-01

Family

ID=63007175

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/125297 WO2019144773A1 (zh) 2018-01-26 2018-12-29 机器人客服转人工客服的方法和装置

Country Status (8)

Country Link
US (1) US10977664B2 (zh)
EP (1) EP3719732A1 (zh)
JP (1) JP6991341B2 (zh)
KR (1) KR102445992B1 (zh)
CN (1) CN108363745B (zh)
SG (1) SG11202006127PA (zh)
TW (1) TWI698830B (zh)
WO (1) WO2019144773A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112182189A (zh) * 2020-10-10 2021-01-05 网易（杭州）网络有限公司 一种对话处理方法、装置、电子设备及存储介质
US11875362B1 (en) 2020-07-14 2024-01-16 Cisco Technology, Inc. Humanoid system for automated customer support
US11907670B1 (en) 2020-07-14 2024-02-20 Cisco Technology, Inc. Modeling communication data streams for multi-party conversations involving a humanoid

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108363745B (zh) 2018-01-26 2020-06-30 阿里巴巴集团控股有限公司 机器人客服转人工客服的方法和装置
US20200050941A1 (en) * 2018-08-07 2020-02-13 Amadeus S.A.S. Machine learning systems and methods for attributed sequences
CN109202908B (zh) * 2018-10-19 2021-01-29 和美（深圳）信息技术股份有限公司 机器人的控制方法、装置、设备、系统和存储介质
CN110020426B (zh) * 2019-01-21 2023-09-26 创新先进技术有限公司 将用户咨询分配到客服业务组的方法及装置
CN111861610A (zh) * 2019-04-30 2020-10-30 北京嘀嘀无限科技发展有限公司 一种数据处理方法、装置、电子设备及存储介质
CN110602334A (zh) * 2019-09-03 2019-12-20 上海航动科技有限公司 一种基于人机协同的智能外呼方法及系统
CN110838014A (zh) * 2019-09-20 2020-02-25 北京智齿博创科技有限公司 在线客服系统中人工智能路由策略
CN111143537A (zh) * 2019-12-30 2020-05-12 税友软件集团股份有限公司 一种基于智能客服系统的服务方法、装置、设备及介质
CN111538822B (zh) * 2020-04-24 2023-05-09 支付宝(杭州)信息技术有限公司 一种智能客户服务机器人训练数据的生成方法和系统
CN111369080B (zh) * 2020-05-27 2020-08-28 支付宝(杭州)信息技术有限公司 一种智能客服解决率预测方法和系统以及多业务预测模型
CN113037935B (zh) * 2020-08-13 2022-09-27 深圳市世纪中正科技开发有限公司 基于大数据处理的用户语音呼叫系统
CN111932144B (zh) * 2020-08-25 2023-09-19 腾讯科技（上海）有限公司 一种客服坐席分配方法、装置、服务器及存储介质
CN112508585A (zh) * 2020-12-03 2021-03-16 大唐融合通信股份有限公司 一种广电客户服务业务处理的方法、设备及装置
CN112329928B (zh) * 2020-12-30 2021-04-30 四川新网银行股份有限公司 基于异构模型的用户满意度分析方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909930A (zh) * 2015-12-23 2017-06-30 神州数码信息系统有限公司 一种基于政务机器问答系统的人机自动切换的模型与方法
US20170195486A1 (en) * 2015-12-30 2017-07-06 Shanghai Xiaoi Robot Technology Co., Ltd. Intelligent customer service systems, customer service robots, and methods for providing customer service
CN107071193A (zh) * 2016-11-28 2017-08-18 阿里巴巴集团控股有限公司 互动应答系统接入用户的方法和装置
CN107451199A (zh) * 2017-07-05 2017-12-08 阿里巴巴集团控股有限公司 问题推荐方法及装置、设备
CN107590159A (zh) * 2016-07-08 2018-01-16 阿里巴巴集团控股有限公司 机器人客服转人工客服的方法和装置
CN108363745A (zh) * 2018-01-26 2018-08-03 阿里巴巴集团控股有限公司 机器人客服转人工客服的方法和装置

Family Cites Families (28)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7222075B2 (en) * 1999-08-31 2007-05-22 Accenture Llp Detecting emotions using voice signal analysis
US6275806B1 (en) 1999-08-31 2001-08-14 Andersen Consulting, Llp System method and article of manufacture for detecting emotion in voice signals by utilizing statistics for voice signal parameters
JP5033994B2 (ja) * 2006-01-19 2012-09-26 株式会社国際電気通信基礎技術研究所 コミュニケーションロボット
JP4412504B2 (ja) 2007-04-17 2010-02-10 本田技研工業株式会社 音声認識装置、音声認識方法、及び音声認識用プログラム
JP2011033680A (ja) 2009-07-30 2011-02-17 Sony Corp 音声処理装置及び方法、並びにプログラム
CN103456297B (zh) 2012-05-29 2015-10-07 中国移动通信集团公司 一种语音识别匹配的方法和设备
US8788439B2 (en) 2012-12-21 2014-07-22 InsideSales.com, Inc. Instance weighted learning machine learning model
US9177318B2 (en) 2013-04-22 2015-11-03 Palo Alto Research Center Incorporated Method and apparatus for customizing conversation agents based on user characteristics using a relevance score for automatic statements, and a response prediction function
US9728184B2 (en) 2013-06-18 2017-08-08 Microsoft Technology Licensing, Llc Restructuring deep neural network acoustic models
US9613619B2 (en) 2013-10-30 2017-04-04 Genesys Telecommunications Laboratories, Inc. Predicting recognition quality of a phrase in automatic speech recognition systems
US9413891B2 (en) 2014-01-08 2016-08-09 Callminer, Inc. Real-time conversational analytics facility
US20150201077A1 (en) 2014-01-12 2015-07-16 Genesys Telecommunications Laboratories, Inc. Computing suggested actions in caller agent phone calls by using real-time speech analytics and real-time desktop analytics
KR102143238B1 (ko) * 2014-03-13 2020-08-10 에스케이플래닛 주식회사 메신저 서비스에서 프로필 정보를 이용한 부가정보 제공 방법, 이를 위한 시스템 및 장치
US10235639B1 (en) * 2014-03-14 2019-03-19 Directly Software, Inc. Al based CRM system
CN105592237B (zh) * 2014-10-24 2019-02-05 中国移动通信集团公司 一种会话切换的方法、装置及智能客服机器人
CN104409075B (zh) 2014-11-28 2018-09-04 深圳创维－Rgb电子有限公司 语音识别方法和系统
CN105072173A (zh) * 2015-08-03 2015-11-18 谌志群 自动客服和人工客服自动切换的客服方法及系统
US9635181B1 (en) 2015-10-19 2017-04-25 Genesys Telecommunications Laboratories, Inc. Optimized routing of interactions to contact center agents based on machine learning
CN107025228B (zh) 2016-01-29 2021-01-26 阿里巴巴集团控股有限公司 一种问题推荐方法及设备
CN105808652B (zh) * 2016-02-26 2021-05-25 北京京东尚科信息技术有限公司 在线客服的实现方法和装置
CN105701088B (zh) * 2016-02-26 2018-12-28 北京京东尚科信息技术有限公司 从机器对话切换到人工对话的方法和装置
CN109196527A (zh) * 2016-04-13 2019-01-11 谷歌有限责任公司 广度和深度机器学习模型
EP4030295B1 (en) * 2016-04-18 2024-06-05 Google LLC Automated assistant invocation of appropriate agent
US10560575B2 (en) * 2016-06-13 2020-02-11 Google Llc Escalation to a human operator
US10403273B2 (en) 2016-09-09 2019-09-03 Oath Inc. Method and system for facilitating a guided dialog between a user and a conversational agent
CN106611597B (zh) 2016-12-02 2019-11-08 百度在线网络技术（北京）有限公司 基于人工智能的语音唤醒方法和装置
CN107506372A (zh) * 2017-07-11 2017-12-22 哈尔滨工业大学深圳研究生院 一种机器人客服在混合类型会话下的自动会话切换方法
US10558852B2 (en) * 2017-11-16 2020-02-11 Adobe Inc. Predictive analysis of target behaviors utilizing RNN-based user embeddings

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909930A (zh) * 2015-12-23 2017-06-30 神州数码信息系统有限公司 一种基于政务机器问答系统的人机自动切换的模型与方法
US20170195486A1 (en) * 2015-12-30 2017-07-06 Shanghai Xiaoi Robot Technology Co., Ltd. Intelligent customer service systems, customer service robots, and methods for providing customer service
CN107590159A (zh) * 2016-07-08 2018-01-16 阿里巴巴集团控股有限公司 机器人客服转人工客服的方法和装置
CN107071193A (zh) * 2016-11-28 2017-08-18 阿里巴巴集团控股有限公司 互动应答系统接入用户的方法和装置
CN107451199A (zh) * 2017-07-05 2017-12-08 阿里巴巴集团控股有限公司 问题推荐方法及装置、设备
CN108363745A (zh) * 2018-01-26 2018-08-03 阿里巴巴集团控股有限公司 机器人客服转人工客服的方法和装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3719732A4 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11875362B1 (en) 2020-07-14 2024-01-16 Cisco Technology, Inc. Humanoid system for automated customer support
US11907670B1 (en) 2020-07-14 2024-02-20 Cisco Technology, Inc. Modeling communication data streams for multi-party conversations involving a humanoid
CN112182189A (zh) * 2020-10-10 2021-01-05 网易（杭州）网络有限公司 一种对话处理方法、装置、电子设备及存储介质
CN112182189B (zh) * 2020-10-10 2023-06-30 网易（杭州）网络有限公司 一种对话处理方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
TW201933267A (zh) 2019-08-16
JP2021524071A (ja) 2021-09-09
JP6991341B2 (ja) 2022-01-12
TWI698830B (zh) 2020-07-11
KR20200095516A (ko) 2020-08-10
US10977664B2 (en) 2021-04-13
EP3719732A4 (en) 2020-10-07
EP3719732A1 (en) 2020-10-07
US20200294063A1 (en) 2020-09-17
CN108363745B (zh) 2020-06-30
CN108363745A (zh) 2018-08-03
SG11202006127PA (en) 2020-07-29
KR102445992B1 (ko) 2022-09-21

Similar Documents

Publication Publication Date Title
WO2019144773A1 (zh) 机器人客服转人工客服的方法和装置
US11669918B2 (en) Dialog session override policies for assistant systems
Lalwani et al. Implementation of a Chatbot System using AI and NLP
US10762892B2 (en) Rapid deployment of dialogue system
WO2018006727A1 (zh) 机器人客服转人工客服的方法和装置
US11012569B2 (en) Insight based routing for help desk service
US20180082184A1 (en) Context-aware chatbot system and method
US20180075335A1 (en) System and method for managing artificial conversational entities enhanced by social knowledge
US20180330232A1 (en) Identification and classification of training needs from unstructured computer text using a neural network
US10770072B2 (en) Cognitive triggering of human interaction strategies to facilitate collaboration, productivity, and learning
CA3147634A1 (en) Method and apparatus for analyzing sales conversation based on voice recognition
CN116541504B (zh) 对话生成方法、装置、介质及计算设备
Shah et al. Problem solving chatbot for data structures
CN115914148A (zh) 具有两侧建模的对话智能体
US20230113524A1 (en) Reactive voice device management
US10909422B1 (en) Customer service learning machine
US20210142180A1 (en) Feedback discriminator
US10878339B2 (en) Leveraging machine learning to predict user generated content
Jiang et al. Large Language Model for Causal Decision Making
US20240152871A1 (en) Intelligent generation of job profiles
Guvindan Raju et al. Cognitive Virtual Admissions Counselor
US12008661B2 (en) Social media management platform
US11875127B2 (en) Query response relevance determination
Perez Dialog state tracking, a machine reading approach using a memory-enhanced neural network
US20230267558A1 (en) Social media management platform

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: 18902124

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020536672

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20207018960

Country of ref document: KR

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2018902124

Country of ref document: EP

Effective date: 20200629

NENP Non-entry into the national phase

Ref country code: DE