WO2020056978A1 - 按键预测方法、装置及计算机可读存储介质 - Google Patents

按键预测方法、装置及计算机可读存储介质 Download PDF

Info

Publication number
WO2020056978A1
WO2020056978A1 PCT/CN2018/123596 CN2018123596W WO2020056978A1 WO 2020056978 A1 WO2020056978 A1 WO 2020056978A1 CN 2018123596 W CN2018123596 W CN 2018123596W WO 2020056978 A1 WO2020056978 A1 WO 2020056978A1
Authority
WO
WIPO (PCT)
Prior art keywords
incoming
prediction model
customer
prediction
preset time
Prior art date
Application number
PCT/CN2018/123596
Other languages
English (en)
French (fr)
Inventor
周俊琨
彭小明
李培彬
严江浩
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2020056978A1 publication Critical patent/WO2020056978A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04MTELEPHONIC COMMUNICATION
    • H04M3/00Automatic or semi-automatic exchanges
    • H04M3/42Systems providing special services or facilities to subscribers
    • H04M3/50Centralised arrangements for answering calls; Centralised arrangements for recording messages for absent or busy subscribers ; Centralised arrangements for recording messages
    • H04M3/51Centralised call answering arrangements requiring operator intervention, e.g. call or contact centers for telemarketing
    • H04M3/5166Centralised call answering arrangements requiring operator intervention, e.g. call or contact centers for telemarketing in combination with interactive voice response systems or voice portals, e.g. as front-ends
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • G06Q30/0203Market surveys; Market polls
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0281Customer communication at a business location, e.g. providing product or service information, consulting

Definitions

  • the present application relates to the field of artificial intelligence technology, and in particular, to a key prediction method, device, and computer-readable storage medium based on an automatic voice response technology.
  • Automatic voice response In the customer service system, automatic voice response (Interactive Voice Response (IVR)) and manual services are two important channels for product / service providers to interact with users.
  • Automatic voice response can interact with users through automatic processes, which is simple to complete. Clear inquiries, consultations, business transactions and other functions are fast, clear, reasonable, simple, and low in operating costs.
  • Manual services are humane, timely, and distinguished services. Automatic voice response and manual services complement each other. With With the rapid development of communication services, users have higher and higher requirements for customer service, and the labor resources of customer service systems are becoming more and more tight. This requires that customer service systems not only ensure user satisfaction, but also use automatic voice response to divert traffic. A part of valuable human resources.
  • IVR Automatic voice response system
  • Interactive Voice Response Interactive Voice Response
  • the product / service provider provides a unified, fixed IVR menu for all users based on factors such as the scope of business and the type of business it provides.
  • the automatic voice response system provides users with a variety of business items in a fixed order, including, for example, auto insurance reporting, life insurance business, credit card business, banking business, property and casualty insurance business, corporate annuity business, and securities business. Broadcast.
  • the present application provides a button prediction method, device, and computer-readable storage medium, whose main purpose is to adjust the sequence of voice announcements of the IVR menu by predicting the type of service that the user needs to handle when the user handles the business through the automatic voice response system. Therefore, the user can quickly understand the key corresponding to the service to be handled, so as to save the user time and improve the user experience.
  • a key prediction method provided in this application includes:
  • the user attributes of the customer the past behavior attributes performed by the customer through different channels within a preset period of time, and the incoming characteristic attributes of the incoming line are analyzed;
  • the prediction data is input into a pre-trained prediction model to predict the customer's current intent, and according to the predicted customer's current intent, according to a preset rule, the sequence of voice announcements of the automatic voice response menu is performed. auto-adjust.
  • the pre-trained prediction model is a Deep and Wide model, wherein the Deep and Wide model includes a linear softmax regression model and a DNN neural network model.
  • the prediction model includes a first prediction model, a second prediction model, a third prediction model, and a fourth prediction model, wherein:
  • the first prediction model is obtained by training using historical incoming data of an interval between adjacent incoming lines within a first preset time
  • the second prediction model is obtained by using historical incoming data with an interval time between adjacent incoming lines being greater than a first preset time and less than or equal to a second preset time;
  • the third prediction model is obtained by training using historical incoming data with an interval time between adjacent incoming lines greater than a second preset time and less than or equal to a third preset time;
  • the fourth prediction model is obtained by using historical incoming line data with an interval between adjacent incoming lines longer than a third preset time.
  • the preset rule includes: priority reporting of function buttons of a predetermined service, and for function buttons corresponding to other services, the selection probability of various services according to the predicted intention of the customer to enter the line from the largest to the smallest Function buttons corresponding to various services are broadcast in sequence.
  • the preset rule further includes: broadcasting only function buttons corresponding to services with a selection probability greater than or equal to a predetermined threshold, and hiding function buttons corresponding to services with a selection probability less than the predetermined threshold.
  • the present application also provides a key prediction device.
  • the device includes a memory and a processor.
  • the memory stores a key prediction program that can be run on the processor.
  • the key prediction program is When the processor executes, the following steps are implemented:
  • the user attributes of the customer the past behavior attributes performed by the customer through different channels within a preset period of time, and the incoming characteristic attributes of the incoming line are analyzed;
  • the prediction data is input into a pre-trained prediction model to predict the customer's current intent, and according to the predicted customer's current intent, according to a preset rule, the sequence of voice announcements of the automatic voice response menu auto-adjust.
  • the pre-trained prediction model is a Deep and Wide model, wherein the Deep and Wide model includes a linear softmax regression model and a DNN neural network model.
  • the prediction model includes a first prediction model, a second prediction model, a third prediction model, and a fourth prediction model, wherein:
  • the first prediction model is obtained by training using historical incoming data of an interval between adjacent incoming lines within a first preset time
  • the second prediction model is obtained by using historical incoming data with an interval time between adjacent incoming lines being greater than a first preset time and less than or equal to a second preset time;
  • the third prediction model is obtained by training using historical incoming data with an interval time between adjacent incoming lines greater than a second preset time and less than or equal to a third preset time;
  • the fourth prediction model is obtained by using historical incoming line data with an interval between adjacent incoming lines longer than a third preset time.
  • the preset rule includes: priority reporting of function buttons of a predetermined service, and for function buttons corresponding to other services, the selection probability of various services according to the predicted intention of the customer to enter the line from the largest to the smallest Function buttons corresponding to various services are broadcast in sequence.
  • the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a key prediction program, and the key prediction program can be executed by one or more processors to implement Steps of the key prediction method as described above.
  • the key prediction method, device, and computer-readable storage medium proposed in this application when a user handles a business through an automatic voice response system, acquires the user's user attributes, past behavior attributes, incoming line intention attributes, and the current progress according to the basic information of the user.
  • the attributes of the incoming characteristics of the line are combined to generate prediction data, and the pre-trained prediction model is used to predict the type of business the user currently needs to handle based on the prediction data, and the voice announcement of the IVR menu is adjusted accordingly Order, so that the user can quickly understand the key corresponding to the business to be processed, so as to save the user time and improve the user experience.
  • FIG. 1 is a schematic flowchart of a key prediction method according to an embodiment of the present application
  • FIG. 2 is a detailed flowchart of one step in the key prediction method provided by the embodiment shown in FIG. 1; FIG.
  • FIG. 3 is a schematic diagram of an internal structure of a key prediction device according to an embodiment of the present application.
  • FIG. 4 is a schematic block diagram of a key prediction program in a key prediction device provided by an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of a key prediction method according to an embodiment of the present application.
  • the method may be performed by a device, which may be implemented by software and / or hardware.
  • the key prediction method includes:
  • the device when the device receives an incoming line, the device connects the incoming line to an automatic voice response system, and simultaneously identifies the phone number of the incoming line.
  • the customer database stores basic information of all customers of a preset product / service provider.
  • the basic information includes, but is not limited to, the customer's name, gender, ID number, telephone number, address, type of product or service purchased, and so on.
  • the device may be connected to the customer database through a network or other means.
  • the product / service provider may be, for example, a company that provides insurance services, such as Ping An.
  • the telephone number is matched with each customer record in the customer database to find the basic information of the customer to which the telephone number belongs.
  • the historical behavior database records the time and behavior attributes of each incoming line for each incoming line.
  • the time attribute includes the year, month, day, and time when the line was entered; the behavior attribute includes the service handled when the line is entered.
  • the historical behavior database may include the following records:
  • a preferred embodiment of this solution performs matching in the historical behavior database according to the telephone number, and obtains all incoming behavior records of the telephone number in a preset time period.
  • the preset time period may be, for example, within half a year pushed forward from the current date.
  • the customer database and the historical behavior database may be integrated into a same database.
  • S3 further comprises obtaining the incoming intention attributes of the customer according to all incoming behavior records of the preset time period.
  • the attributes of the incoming line include the attributes of the last incoming line, the attributes of the last incoming line, the attributes of the latest week, the attributes of the last month, the attributes of the last three months, and the last six months Incoming attributes and more.
  • the attributes of the incoming line include the time of entering the line and the behavior attributes.
  • the user attribute may indicate a type of a product or service that a customer has purchased.
  • the customer attributes of the customer are marked with 1 for car insurance attributes, 1 for credit card attributes, and 0 for other attributes. If life insurance is bought, the life insurance attribute in the user attributes of the customer is marked as 1, other attributes are marked as 0, and so on.
  • the channels may include, for example, a phone channel, a web page channel, a mobile application software (APP) channel, a third-party payment channel, and the like.
  • APP mobile application software
  • the analysis server obtains the past behavior attributes of the customer in different channels according to a combination of one or more basic information such as the telephone number, name and ID number of the customer.
  • the past behavior attributes may include, but are not limited to, for example, customer A purchased a wealth management product through an app on January 1, 2018; on January 2, 2018, a WeChat payment method was used to The credit card in the name spent 10,000 yuan and so on.
  • the incoming feature attributes of this incoming line include 7-dimensional week feature attributes, 24-dimensional hour feature attributes, and 30-dimensional date feature attributes.
  • the feature combination may use an FM (Factorization Machine) factorization algorithm.
  • FM Vectorization Machine
  • the linear model only considers the influence of a single feature on the prediction result, and does not consider the influence of the combined feature on the prediction result, and the FM algorithm is designed to solve the problem of feature combination under sparse data.
  • n (n-1) / 2g combined feature parameters there are n (n-1) / 2g combined feature parameters. It is important that any two parameters are independent, but in the case of very sparse features, the combined feature (x i , x j ) appears at the same time as not In the case of less than 0, learning the parameter w ij directly by using the gradient descent method will make a large number of w ij learning results 0, so it may cause insufficient training samples and easily cause the parameter w ij to be inaccurate and affect the model. final effect.
  • the W matrix is decomposed into:
  • k represents the dimension of the v vector
  • the direct calculation complexity is O (kn 2 )
  • O (kn) the calculation complexity can be reduced from O (kn 2 ) to O (kn).
  • the prediction data is input into a pre-trained prediction model to predict a customer's intention to enter the current line.
  • the prediction model is obtained by training according to the user's historical incoming data.
  • the pre-trained prediction model is a Deep and Wide model.
  • the Deep and Wide model is composed of 2 parts: a linear softmax regression model and a DNN neural network model.
  • the Softmax regression model can save the original discrete features, the DNN neural network model can nonlinearly transform the original features into new features, and then use the combination of the original features + the new feature into the line to obtain a new feature into the model prediction.
  • Deep and wide models can achieve better results than traditional separate softmax regression models and DNN neural network models.
  • the prediction model includes four.
  • the first prediction model is obtained by using the history of adjacent line behavior interval time within a first preset time, such as less than 1 day;
  • the second prediction model is based on the use of adjacent line behavior interval time is greater than the first Preset time, such as 1 day and less than or equal to the second preset time, such as obtained from historical incoming data training within 3 days;
  • the third prediction model is to use the interval between adjacent incoming lines to be greater than the second preset time, such as 3 Days and is less than or equal to the third preset time, such as the historical entry data training within 7 days;
  • the fourth prediction model is to use the adjacent entry behavior interval time greater than the third preset time, such as 7 days of historical entry Data training.
  • step S8 further includes:
  • the preset rule includes: priority reporting of function buttons of a predetermined service, and for function buttons corresponding to other services, the selection probability of various services in descending order according to the predicted intention of the customer to enter the line is in order Function buttons corresponding to various services are announced.
  • the preset rule further includes: broadcasting only function buttons corresponding to services with a selection probability greater than or equal to a predetermined threshold, and hiding function buttons corresponding to services with a selection probability less than the predetermined threshold.
  • scheduled services such as auto insurance reporting buttons and function buttons corresponding to an account user
  • scheduled services are required and given priority to be reported.
  • function buttons corresponding to other services the predicted probability is greater than or equal to 0.85, ranked from highest probability Broadcast, services with a predicted probability less than 0.85 can be hidden, and by letting the user select the button that broadcasts all services, the corresponding function buttons are broadcasted in order according to the predicted probability ranking.
  • the IVR menu of the automatic voice response system may broadcast the following voice:
  • the invention also provides a key prediction device.
  • a key prediction device Referring to FIG. 3, a schematic diagram of an internal structure of a key prediction device according to an embodiment of the present application is shown.
  • the key prediction device 1 may be a PC (Personal Computer), a terminal device such as a smart phone, a tablet computer, a portable computer, or a server.
  • the key prediction device 1 includes at least a memory 11, a processor 12, a communication bus 13, and a network interface 14.
  • the memory 11 includes at least one type of readable storage medium.
  • the readable storage medium includes a flash memory, a hard disk, a multimedia card, a card-type memory (for example, SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, and the like.
  • the memory 11 may be an internal storage unit of the key prediction device 1 in some embodiments, such as a hard disk of the key prediction device 1.
  • the memory 11 may also be an external storage device of the key prediction device 1 in other embodiments, such as a plug-in hard disk, a smart memory card (SMC), and a secure digital (Secure Digital, SD) card, flash card, etc.
  • the memory 11 may include both an internal storage unit of the key prediction device 1 and an external storage device.
  • the memory 11 can be used not only to store application software installed in the key prediction device 1 and various types of data, such as the code of the key prediction program 01, but also to temporarily store data that has been or will be output.
  • the processor 12 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments, and is configured to run program codes or processes stored in the memory 11 Data, for example, the key prediction program 01 is executed.
  • CPU central processing unit
  • controller a controller
  • microcontroller a microcontroller
  • microprocessor or other data processing chip in some embodiments, and is configured to run program codes or processes stored in the memory 11 Data, for example, the key prediction program 01 is executed.
  • the communication bus 13 is used to implement connection and communication between these components.
  • the network interface 14 may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), and is generally used to establish a communication connection between the device 1 and other electronic devices.
  • a standard wired interface such as a WI-FI interface
  • the device 1 may further include a user interface.
  • the user interface may include a display, an input unit such as a keyboard, and the optional user interface may further include a standard wired interface and a wireless interface.
  • the display may be an LED display, a liquid crystal display, a touch-type liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light emitting diode) touch device, or the like.
  • the display may also be appropriately called a display screen or a display unit for displaying information processed in the key prediction device 1 and for displaying a visual user interface.
  • FIG. 2 only shows the key prediction device 1 having the components 11-14 and the key prediction program 01.
  • FIG. 1 does not constitute a limitation on the key prediction device 1, and may include There are fewer or more parts than shown, or some parts are combined, or different parts are arranged.
  • the key prediction program 01 is stored in the memory 11; when the processor 12 executes the key prediction program 01 stored in the memory 11, the following steps are implemented:
  • Step 1 Receive incoming calls and identify the phone number of the incoming call.
  • the device when the device receives an incoming line, the device connects the incoming line to an automatic voice response system, and simultaneously identifies the phone number of the incoming line.
  • Step 2 According to the telephone number, obtain basic information of a customer to which the telephone number belongs from a customer database.
  • the customer database stores basic information of all customers of a preset product / service provider.
  • the basic information includes, but is not limited to, the customer's name, gender, ID number, telephone number, address, type of product or service purchased, and so on.
  • the device may be connected to the customer database through a network or other means.
  • the product / service provider may be, for example, a company that provides insurance services, such as Ping An.
  • the telephone number is matched with each customer record in the customer database to find the basic information of the customer to which the telephone number belongs.
  • Step 3 According to the telephone number, obtain a record of incoming behavior of the telephone number in a preset time period from a historical behavior database.
  • the historical behavior database records the time and behavior attributes of each incoming line for each incoming line.
  • the time attribute includes the year, month, day, and time when the line was entered; the behavior attribute includes which services were handled when the line was entered.
  • the historical behavior database may include the following records:
  • a preferred embodiment of this solution performs matching in the historical behavior database according to the telephone number, and obtains all incoming behavior records of the telephone number in a preset time period.
  • the preset time period may be, for example, within half a year pushed forward from the current date.
  • the customer database and the historical behavior database may be integrated into a same database.
  • step 3 further includes obtaining the incoming intention attributes of the customer according to all incoming behavior records of the preset time period.
  • the attributes of the incoming line include the attributes of the last incoming line, the attributes of the last incoming line, the attributes of the latest week, the attributes of the last month, the attributes of the last three months, and the last six months Incoming attributes and more.
  • the attributes of the incoming line include the time of entering the line and the behavior attributes.
  • Step 4 Acquire the user attributes of the customer according to the basic information of the customer.
  • the user attribute may indicate a type of a product or service that a customer has purchased.
  • the customer attributes of the customer are marked with 1 for car insurance attributes, 1 for credit card attributes, and 0 for other attributes. If life insurance is bought, the life insurance attribute in the user attributes of the customer is marked as 1, other attributes are marked as 0, and so on.
  • Step 5 According to the basic information of the customer, acquire the past behavior attributes performed by the customer through different channels within a preset time period.
  • the channels may include, for example, a phone channel, a web page channel, a mobile application software (APP) channel, a third-party payment channel, and the like.
  • APP mobile application software
  • the analysis server obtains the past behavior attributes of the customer in different channels according to a combination of one or more basic information such as the telephone number, name and ID number of the customer.
  • the past behavior attributes may include, but are not limited to, for example, customer A purchased a wealth management product through an app on January 1, 2018; on January 2, 2018, a WeChat payment method was used to The credit card in the name spent 10,000 yuan and so on.
  • Step 6 Analyze the incoming line characteristic attributes of this incoming line.
  • the incoming feature attributes of this incoming line include 7-dimensional week feature attributes, 24-dimensional hour feature attributes, and 30-dimensional date feature attributes.
  • Step 7 The feature combination of the user attributes, past behavior attributes, incoming intention attributes, and incoming feature attributes of the incoming line obtained above is used to obtain prediction data.
  • the feature combination may use an FM (Factorization Machine) factorization algorithm.
  • FM Vectorization Machine
  • the linear model only considers the influence of a single feature on the prediction result, and does not consider the influence of the combined feature on the prediction result, and the FM algorithm is designed to solve the problem of feature combination under sparse data.
  • n (n-1) / 2g combined feature parameters there are n (n-1) / 2g combined feature parameters. It is important that any two parameters are independent, but in the case of very sparse features, the combined feature (x i , x j ) appears at the same time as not In the case of less than 0, learning the parameter w ij directly by using the gradient descent method will make a large number of w ij learning results 0, so it may cause insufficient training samples and easily cause the parameter w ij to be inaccurate and affect the model. final effect.
  • the W matrix is decomposed into:
  • k represents the dimension of the v vector
  • the direct calculation complexity is O (kn 2 )
  • O (kn) the calculation complexity can be reduced from O (kn 2 ) to O (kn).
  • Step 8 The prediction data is input into a pre-trained prediction model to predict a customer's intention to enter the line this time.
  • the prediction model is obtained by training according to the user's historical incoming data.
  • the pre-trained prediction model is a Deep and Wide model.
  • the Deep and Wide model is composed of 2 parts: a linear softmax regression model and a DNN neural network model.
  • the Softmax regression model can save the original discrete features, the DNN neural network model can nonlinearly transform the original features into new features, and then use the combination of the original features + the new feature into the line to obtain a new feature into the model prediction.
  • Deep and wide models can achieve better results than traditional separate softmax regression models and DNN neural network models.
  • the prediction model includes four.
  • the first prediction model is obtained by using the history of adjacent line behavior interval time within a first preset time, such as less than 1 day;
  • the second prediction model is based on the use of adjacent line behavior interval time is greater than the first Preset time, such as 1 day and less than or equal to the second preset time, such as obtained from historical incoming data training within 3 days;
  • the third prediction model is to use the interval between adjacent incoming lines to be greater than the second preset time, such as 3 Days and is less than or equal to the third preset time, such as the historical entry data training within 7 days;
  • the fourth prediction model is to use the adjacent entry behavior interval time greater than the third preset time, such as 7 days of historical entry Data training.
  • step eight further includes:
  • Sub-step 1 Calculate the time interval between the current incoming line of the telephone number and the incoming line.
  • Sub-step 2. Determine whether the time interval is within one day.
  • substep 3 input the prediction data into a first training model trained in advance to predict the customer's intention of entering the line this time.
  • substep 4 determine whether the time interval is within one day and less than or equal to three days.
  • sub-step 5 input the prediction data into a pre-trained second prediction model to predict the current customer Intent to enter the line.
  • sub-step 4 determines whether the time interval is more than 1 day and less than or equal to 3 days.
  • sub-step 7 input the prediction data into a pre-trained third prediction model to predict the customer's current entry intention.
  • sub-step 8 input the prediction data into a pre-trained fourth prediction model to predict the customer's current entry intention.
  • Step 9 According to the predicted intention of the customer to enter the line, the sequence of voice announcements of the automatic voice response menu is automatically adjusted according to a preset rule.
  • the preset rule includes: priority reporting of function buttons of a predetermined service, and for function buttons corresponding to other services, the selection probability of various services in descending order according to the predicted intention of the customer to enter the line is in order Function buttons corresponding to various services are announced.
  • the preset rule further includes: broadcasting only function buttons corresponding to services with a selection probability greater than or equal to a predetermined threshold, and hiding function buttons corresponding to services with a selection probability less than the predetermined threshold.
  • scheduled services such as auto insurance reporting buttons and function buttons corresponding to an account user
  • scheduled services are required and given priority to be reported.
  • function buttons corresponding to other services the predicted probability is greater than or equal to 0.85, ranked from highest probability Broadcast, services with a predicted probability less than 0.85 can be hidden, and by letting the user select the button that broadcasts all services, the corresponding function buttons are broadcasted in order according to the predicted probability ranking.
  • the IVR menu of the automatic voice response system may broadcast the following voice:
  • the key prediction program may also be divided into one or more modules, and the one or more modules are stored in the memory 11 and processed by one or more processors (the processing in this embodiment is 12) is executed to complete the present application.
  • the module referred to in the present application refers to a series of computer program instruction segments capable of performing specific functions and is used to describe the execution process of the key prediction program in the key prediction device.
  • FIG. 4 it is a schematic diagram of a program module of a key prediction program in an embodiment of the key prediction device of the present application.
  • the key prediction program can be divided into a customer identification module 10 and a customer feature acquisition module. 20.
  • the prediction data calculation module 30 and the key prediction module 40 for example:
  • the customer identification module 10 is configured to receive an incoming line, identify a telephone number of the incoming line, and obtain basic information of a customer to which the telephone number belongs from a customer database according to the telephone number.
  • the customer feature obtaining module 20 is configured to: according to the telephone number, obtain a record of incoming behavior of the telephone number in a preset time period from a historical behavior database, and according to all incoming behaviors of the preset time period Record, get the customer's incoming intention attributes, and learn the customer's user attributes, the customer's past behavior attributes performed through different channels within a preset period of time based on the customer's basic information, and analyze this time Incoming line characteristic attributes of incoming line.
  • the predictive data calculation module 30 is configured to perform feature combination of the user attributes, past behavior attributes, incoming intention attributes, and incoming feature attributes of the incoming line obtained above to obtain predicted data.
  • the button prediction module 40 is configured to input the prediction data into a pre-trained prediction model to predict the customer's intention of entering the line, and according to the predicted customer's intention to enter the line, the preset The voice announcement order of the automatic voice response menu is automatically adjusted.
  • an embodiment of the present application further provides a computer-readable storage medium.
  • the computer-readable storage medium stores a key prediction program, and the key prediction program can be executed by one or more processors to implement the following operations:
  • an incoming behavior record of the telephone number in a preset period of time is obtained from a historical behavior database, and the incoming intention of the customer is obtained according to all the incoming behavior records of the preset period of time. Attributes;
  • the user attributes of the customer the past behavior attributes performed by the customer through different channels within a preset period of time, and the incoming characteristic attributes of the incoming line are analyzed;
  • the sequence of voice announcements of the automatic voice response menu is automatically adjusted according to a preset rule.
  • the methods in the above embodiments can be implemented by means of software plus a necessary universal hardware platform, and of course, also by hardware, but in many cases the former is better.
  • Implementation Based on such an understanding, the technical solution of this application that is essentially or contributes to the existing technology can be embodied in the form of a software product.
  • the computer software product is stored in a storage medium (such as ROM / RAM) as described above. , Magnetic disk, optical disc), including a number of instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in the embodiments of the present application.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Signal Processing (AREA)
  • Telephonic Communication Services (AREA)

Abstract

本申请涉及一种人工智能技术,揭露了一种按键预测方法,包括:接收进线,识别电话号码,获取客户的基本信息;获取该电话号码在预设时间段的进线行为记录以得到客户的进线意图属性;根据客户的基本信息获知客户的用户属性、在预设时间段内通过不同渠道执行的过往行为属性,本次进线的进线特征属性;将上述得到的各种属性数据进行特征组合,得到预测数据;及将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图,并按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。本申请还提出一种按键预测装置以及一种计算机可读存储介质。本申请实现了对IVR菜单的语音播报顺序的自能调整,节省用户时间,提高用户体验。

Description

按键预测方法、装置及计算机可读存储介质
本申请要求于2018年9月19日提交中国专利局,申请号为201811096236.4、发明名称为“按键预测方法、装置及计算机可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域,尤其涉及一种基于自动语音应答技术的按键预测方法、装置及计算机可读存储介质。
背景技术
在客服系统中,自动语音应答((Interactive Voice Response,IVR)与人工服务是产品/服务提供商与用户交互的两个重要的通道。自动语音应答可以通过自动流程与用户进行信息交互,完成简单明确的查询、咨询、业务办理等功能,具有快捷、清晰、合理、简便、运营成本低的特点。人工服务具有人性化、及时化、体现尊贵服务的特点。自动语音应答与人工服务相辅相成。随着通信业务的飞速发展,用户对客服业务的要求越来越高,而客服系统的人工资源也越来越紧张,这就要求客服系统既要保证用户的满意度,又要用自动语音应答分流一部分宝贵的人工资源。
目前,有多种方法可以为用户提供自动语音应答系统(IVR,Interactive Voice Response)菜单,产品/服务提供商为用户提供统一的菜单即是其中的一种方法。在这种方法中,产品/服务提供商完全根据自身提供的业务范围、业务类型等因素,为所有的用户提供统一的、固定的IVR菜单。
然而,在实际应用中,对于统一的、固定的IVR菜单,用户在办理一个业务时,可能需要在菜单中进行多个层次的选择,才能找到自己需要的服务项目。例如,用户通过95511拨打平安公司的客服电话时,首先接入的为平安公司提供的自动语音应答系统。传统上,该自动语音应答系统为用户提供了按照固定顺序的,包括例如车险报案、人寿业务、信用卡业务、银行业务、财产险意外险业务、企业年金业务、证券业务等多种业务项目的语音播报。对于不熟悉IVR菜单结构的用户来说,可能会听完很多不相关的语音播报之后,才能找到自己需要办理的业务选择按键,浪费了用户的时间,给用户带来不好的使用体验。此外,为了节省时间,用户很可能直接选择人工服务办理业务,这样就会增加人工资源的负担。
发明内容
本申请提供一种按键预测方法、装置及计算机可读存储介质,其主要目的在于在用户通过自动语音应答系统办理业务时,通过预测用户所需办理的业务类型,调整IVR菜单的语音播报顺序,从而使用户快速了解所需办理的业务对应的按键,以节省用户时间,提高用户体验。
为实现上述目的,本申请提供的一种按键预测方法,包括:
接收电话进线,识别该次进线的电话号码,并根据所述电话号码,从数据库中获取该电话号码所属客户的基本信息;
根据所述电话号码,从数据库中获取该电话号码在预设时间段的进线行为记录,并根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性;
根据所述客户的基本信息获知所述客户的用户属性、所述客户在预设时间段内通过不同渠道执行的过往行为属性,并分析本次进线的进线特征属性;
将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据;及
将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图,并根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
可选地,所述预先训练的预测模型为Deep and wide模型,其中,所述Deep and wide模型包括线性softmax回归模型和DNN神经网络模型。
可选地,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
可选地,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
可选地,所述预设规则还包括:仅播报选择概率大于或者等于预定阈值的业务对应的功能按键,对选择概率小于所述预定阈值的业务对应的功能按键进行隐藏。
此外,为实现上述目的,本申请还提供一种按键预测装置,该装置包括存储器和处理器,所述存储器中存储有可在所述处理器上运行的按键预测程序,所述按键预测程序被所述处理器执行时实现如下步骤:
接收电话进线,识别该次进线的电话号码,并根据所述电话号码,从数据库中获取该电话号码所属客户的基本信息;
根据所述电话号码,从数据库中获取该电话号码在预设时间段的进线行 为记录,并根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性;
根据所述客户的基本信息获知所述客户的用户属性、所述客户在预设时间段内通过不同渠道执行的过往行为属性,并分析本次进线的进线特征属性;
将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据;及
将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图,并根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
可选地,所述预先训练的预测模型为Deep and wide模型,其中,所述Deep and wide模型包括线性softmax回归模型和DNN神经网络模型。
可选地,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
可选地,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
此外,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有按键预测程序,所述按键预测程序可被一个或者多个处理器执行,以实现如上所述的按键预测方法的步骤。
本申请提出的按键预测方法、装置及计算机可读存储介质,在用户通过自动语音应答系统办理业务时,根据用户的基本信息获取用户的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性,对这些属性进行特征组合,生成预测数据,并利用预先训练的预测模型,根据所述预测数据预测用户当前所需办理的业务类型,并据此调整IVR菜单的语音播报顺序,从而使用户快速了解所需办理的业务对应的按键,以节省用户时间,提高用户体验。
附图说明
图1为本申请一实施例提供的按键预测方法的流程示意图;
图2为图1所述实施例提供的按键预测方法中其中一个步骤的详细流程示意图;
图3为本申请一实施例提供的按键预测装置的内部结构示意图;
图4为本申请一实施例提供的按键预测装置中按键预测程序的模块示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供一种按键预测方法。参照图1所示,为本申请一实施例提供的按键预测方法的流程示意图。该方法可以由一个装置执行,该装置可以由软件和/或硬件实现。
在本实施例中,按键预测方法包括:
S1、接收进线,识别该次进线的电话号码。
本方案中,所述装置在接收到一个进线时,将该进线接入到一个自动语音应答系统中,并同时识别该次进线的电话号码。
S2、根据所述电话号码,从一个客户数据库中获取该电话号码所属客户的基本信息。
本方案中,所述客户数据库中存储有预设产品/服务提供商的所有客户的基本信息。所述基本信息包括,但不限于,客户的姓名、性别、身份证号码、电话号码、住址、所购买过的产品或者服务的种类等等。
所述装置可以通过网络或者其他方式连接所述客户数据库。
所述产品/服务提供商可以是,例如,提供保险服务的公司,如平安等。
本申请较佳实施例通过所述电话号码与所述客户数据库中的每条客户记录进行匹配,以找到该电话号码所属客户的基本信息。
S3、根据所述电话号码,从一个历史行为数据库中获取该电话号码在预设时间段的进线行为记录。
本申请较佳实施例中,所述历史行为数据库中记录着每一个进线号码每次进线的时间以及行为属性。所述时间属性包括进线的年、月、日、及几时几分;所述行为属性包括进线时办理的业务。
例如,所述历史行为数据库中可能包括如下记录:
1、2017年6月30日的8:00AM,进线号码13411111111,办理了车险业务;
2、2017年7月1日的8:00PM,进线号码13412345678,办理了信用卡业务;以及
3、2017年7月2日的9:30AM,进线号码13411111111,办理了银行业务;
......。
本方案较佳实施例根据所述电话号码在所述历史行为数据库进行匹配, 获取该电话号码在预设时间段的所有进线行为记录。
其中,所述预设时间段可以是,例如从当前日期开始往前推的半年内。
其他方案中,所述的客户数据库以及所述的历史行为数据库可以整合为同一个数据库。
进一步地,S3还包括根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性。
本方案中,所述进线意图属性包括上次进线属性、上上次进线属性、最近一星期进线属性、最近一个月进线属性、最近三个月进线属性、最近六个月进线属性等等。其中,所述进线属性包括进线的时间以及行为属性。
S4、根据所述客户的基本信息,获知所述客户的用户属性。
本申请较佳实施例中,所述用户属性可以指示客户所购买过的产品或者服务的种类。
例如,对于平安公司来说,若客户购买了车险以及办理了信用卡,则所述客户的用户属性中,车险属性标记为1,信用卡属性标记为1,其他属性标记为0;若客户在平安公司买了寿险,则所述客户的用户属性中的寿险属性标记为1,其他属性标记为0等。
S5、根据所述客户的基本信息,获取所述客户在预设时间段内通过不同渠道执行的过往行为属性。
本方案中,所述渠道可以包括,例如,电话渠道、网页渠道、手机应用软件(APP)渠道、第三方支付渠道等。
所述分析服务器根据所述客户的电话号码、姓名以及身份证号码等一项或者多项基本信息的组合,获取所述客户在不同渠道的过往行为属性。
其中,所述的过往行为属性可以包括,但不限于,例如,客户A在2018年1月1日,通过APP购买了一款理财产品;在2018年1月2日,通过微信支付的方式从名下的信用卡消费了1万元钱等等。
S6、分析本次进线的进线特征属性。
本方案中,本次进线的进线特征属性包括7维星期特征属性、24维小时特征属性、30维日期特征属性等。
S7、将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据。
本申请较佳实施例中,优选地,所述特征组合可以采用FM(Factorization Machine,因子分解机)算法。
线性模型只考虑了单一特征对预测结果的影响,没有考虑组合特征对预测结果的影响,而所述FM算法是旨在解决稀疏数据下特征组合问题。
本申请较佳实施例定义目标函数如下:
Figure PCTCN2018123596-appb-000001
上述函数中,组合特征参数一共有n(n-1)/2g个,重要的是任意两个参数独立,但在特征非常稀疏的情况下,组合特征(x i,x j)出现同时不为0的 情况较少的情况下,直接用梯度下降法对参数w ij进行学习会使得大量的w ij学习结果为0,因此可能造成训练样本不足,很容易导致参数w ij不准确,影响模型的最终效果。
在所述FM算法中,将W矩阵分解为:
W=V*V t
因此,上述的目标函数进一步写成:
Figure PCTCN2018123596-appb-000002
Figure PCTCN2018123596-appb-000003
其中:k代表v向量的维度,直接计算复杂度为O(kn 2),因为需要计算所有的两两组合特征的,但通过重新分析目标函数,计算复杂度可以从O(kn 2)降低到O(kn)。
S8,将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图。
本方案中,所述预测模型为根据用户的历史进线数据进行训练得到。
优选地,本申请较佳实施例中,所述预先训练的预测模型为Deep and wide模型。所述Deep and wide模型由2部分组成:线性softmax回归模型和DNN神经网络模型。
所述Softmax回归模型能够保存原始离散特征,DNN神经网络模型能够对原始特征进线非线性变换得到新的特征,再利用原始特征+新特征进线组合串联得到新特征进线模型预测。
Deep and wide模型可以比传统的单独softmax回归模型、DNN神经网络模型达到更好的效果。
优选地,在本申请较佳实施例中,所述预测模型包括四个。其中,第一预测模型为利用相邻进线行为间隔时间在第一预设时间,如1天以内的历史进线数据训练得到;第二预测模型为利用相邻进线行为间隔时间大于第一预设时间,如1天且小于等于第二预设时间,如3天以内的历史进线数据训练得到;第三预测模型为利用相邻进线行为间隔时间大于第二预设时间,如3天且小于等于第三预设时间,如7天以内的历史进线数据训练得到;及第四预测模型为利用相邻进线行为间隔时间大于第三预设时间,如7天的历史进线数据训练得到。
因此,优选地,参阅图2所示,步骤S8还包括:
S81:计算所述电话号码本次进线与上述进线的时间间隔。
S82:判断所述时间间隔是否在1天之内。
若S82中,判断所述时间间隔是在1天之内,则执行S83:将所述预测数据输入预先训练的第一预测模型中,预测客户本次进线的意图。
若S82中,判断所述时间间隔不在1天之内,则执行S84:判断所述时 间间隔是否在大于1天且小于等于3天之内。
若在S84中,判断所述时间间隔是在大于1天且小于等于3天之内,则执行S85:将所述预测数据输入预先训练的第二预测模型中,预测客户本次进线的意图。
若在S84中,判断所述时间间隔不在大于1天且小于等于3天之内,则执行S86:判断所述时间间隔是否在大于3天且小于等于7天之内。
若S86中,判断所述时间间隔在大于3天且小于等于7天之内,则执行S87:将所述预测数据输入预先训练的第三预测模型中,预测客户本次进线的意图。
若S86中,判断所述时间间隔不在大于3天且小于等于7天之内,则执行S88:将所述预测数据输入预先训练的第四预测模型中,预测客户本次进线的意图。
S9,根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
优选地,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
进一步地,所述预设规则还包括:仅播报选择概率大于或者等于预定阈值的业务对应的功能按键,对选择概率小于所述预定阈值的业务对应的功能按键可以进行隐藏。
例如,根据监管要求,预定业务,如车险报案的按键以及一账通用户对应的功能按键需要且优先播报,对于其他业务对应的功能按键,其预测的概率大于等于0.85的,从最高概率排名依次播报,预测的概率小于0.85的业务可以进行隐藏,并通过让用户选择了播报全部业务的按键时,按照预测的概率排名依次播报其对应功能按键。
又如,根据所述预测的客户本次进线的意图,用户选择办理信用卡业务的可能性为90%,选择办理银行业务的可能性为88%,选择办理产险业务的可能性为86%,选择办理其他业务的可能性均小于85%,则所述自动语音应答系统的IVR菜单可能进播报如下语音:
车险报案请按1;一账通业务请按2;信用卡业务请按9;银行业务请按3;产险业务请按4;听取全部业务功能按键请按0等。
发明还提供一种按键预测装置。参照图3所示,为本申请一实施例提供的按键预测装置的内部结构示意图。
在本实施例中,所述按键预测装置1可以是PC(Personal Computer,个人电脑),或者是智能手机、平板电脑、便携计算机等终端设备,也可以是一种服务器等。该按键预测装置1至少包括存储器11、处理器12,通信总线13,以及网络接口14。
其中,存储器11至少包括一种类型的可读存储介质,所述可读存储介质 包括闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、磁性存储器、磁盘、光盘等。存储器11在一些实施例中可以是按键预测装置1的内部存储单元,例如该按键预测装置1的硬盘。存储器11在另一些实施例中也可以是按键预测装置1的外部存储设备,例如按键预测装置1上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器11还可以既包括按键预测装置1的内部存储单元也包括外部存储设备。存储器11不仅可以用于存储安装于按键预测装置1的应用软件及各类数据,例如按键预测程序01的代码等,还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器或其他数据处理芯片,用于运行存储器11中存储的程序代码或处理数据,例如执行按键预测程序01等。
通信总线13用于实现这些组件之间的连接通信。
网络接口14可选的可以包括标准的有线接口、无线接口(如WI-FI接口),通常用于在该装置1与其他电子设备之间建立通信连接。
可选地,该装置1还可以包括用户接口,用户接口可以包括显示器(Display)、输入单元比如键盘(Keyboard),可选的用户接口还可以包括标准的有线接口、无线接口。可选地,在一些实施例中,显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode,有机发光二极管)触摸器等。其中,显示器也可以适当的称为显示屏或显示单元,用于显示在按键预测装置1中处理的信息以及用于显示可视化的用户界面。
图2仅示出了具有组件11-14以及按键预测程序01的按键预测装置1,本领域技术人员可以理解的是,图1示出的结构并不构成对按键预测装置1的限定,可以包括比图示更少或者更多的部件,或者组合某些部件,或者不同的部件布置。
在图2所示的装置1实施例中,存储器11中存储有按键预测程序01;处理器12执行存储器11中存储的按键预测程序01时实现如下步骤:
步骤一、接收进线,识别该次进线的电话号码。
本方案中,所述装置在接收到一个进线时,将该进线接入到一个自动语音应答系统中,并同时识别该次进线的电话号码。
步骤二、根据所述电话号码,从一个客户数据库中获取该电话号码所属客户的基本信息。
本方案中,所述客户数据库中存储有预设产品/服务提供商的所有客户的基本信息。所述基本信息包括,但不限于,客户的姓名、性别、身份证号码、电话号码、住址、所购买过的产品或者服务的种类等等。
所述装置可以通过网络或者其他方式连接所述客户数据库。
所述产品/服务提供商可以是,例如,提供保险服务的公司,如平安等。
本申请较佳实施例通过所述电话号码与所述客户数据库中的每条客户记 录进行匹配,以找到该电话号码所属客户的基本信息。
步骤三、根据所述电话号码,从一个历史行为数据库中获取该电话号码在预设时间段的进线行为记录。
本申请较佳实施例中,所述历史行为数据库中记录着每一个进线号码每次进线的时间以及行为属性。所述时间属性包括进线的年、月、日、及几时几分;所述行为属性包括进线时办理了哪些业务。
例如,所述历史行为数据库中可能包括如下记录:
1、2017年6月30日的8:00AM,进线号码13411111111,办理了车险业务;
2、2017年7月1日的8:00PM,进线号码13412345678,办理了信用卡业务;以及
3、2017年7月2日的9:30AM,进线号码13411111111,办理了银行业务;
......。
本方案较佳实施例根据所述电话号码在所述历史行为数据库进行匹配,获取该电话号码在预设时间段的所有进线行为记录。
其中,所述预设时间段可以是,例如从当前日期开始往前推的半年内。
其他方案中,所述的客户数据库以及所述的历史行为数据库可以整合为同一个数据库。
进一步地,所述步骤三还包括根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性。
本方案中,所述进线意图属性包括上次进线属性、上上次进线属性、最近一星期进线属性、最近一个月进线属性、最近三个月进线属性、最近六个月进线属性等等。其中,所述进线属性包括进线的时间以及行为属性。
步骤四、根据所述客户的基本信息,获知所述客户的用户属性。
本申请较佳实施例中,所述用户属性可以指示客户所购买过的产品或者服务的种类。
例如,对于平安公司来说,若客户购买了车险以及办理了信用卡,则所述客户的用户属性中,车险属性标记为1,信用卡属性标记为1,其他属性标记为0;若客户在平安公司买了寿险,则所述客户的用户属性中的寿险属性标记为1,其他属性标记为0等。
步骤五、根据所述客户的基本信息,获取所述客户在预设时间段内通过不同渠道执行的过往行为属性。
本方案中,所述渠道可以包括,例如,电话渠道、网页渠道、手机应用软件(APP)渠道、第三方支付渠道等。
所述分析服务器根据所述客户的电话号码、姓名以及身份证号码等一项或者多项基本信息的组合,获取所述客户在不同渠道的过往行为属性。
其中,所述的过往行为属性可以包括,但不限于,例如,客户A在2018年1月1日,通过APP购买了一款理财产品;在2018年1月2日,通过微信 支付的方式从名下的信用卡消费了1万元钱等等。
步骤六、分析本次进线的进线特征属性。
本方案中,本次进线的进线特征属性包括7维星期特征属性、24维小时特征属性、30维日期特征属性等。
步骤七、将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据。
本申请较佳实施例中,优选地,所述特征组合可以采用FM(Factorization Machine,因子分解机)算法。
线性模型只考虑了单一特征对预测结果的影响,没有考虑组合特征对预测结果的影响,而所述FM算法是旨在解决稀疏数据下特征组合问题。
本申请较佳实施例定义目标函数如下:
Figure PCTCN2018123596-appb-000004
上述函数中,组合特征参数一共有n(n-1)/2g个,重要的是任意两个参数独立,但在特征非常稀疏的情况下,组合特征(x i,x j)出现同时不为0的情况较少的情况下,直接用梯度下降法对参数w ij进行学习会使得大量的w ij学习结果为0,因此可能造成训练样本不足,很容易导致参数w ij不准确,影响模型的最终效果。
在所述FM算法中,将W矩阵分解为:
W=V*V t
因此,上述的目标函数进一步写成:
Figure PCTCN2018123596-appb-000005
Figure PCTCN2018123596-appb-000006
其中:k代表v向量的维度,直接计算复杂度为O(kn 2),因为需要计算所有的两两组合特征的,但通过重新分析目标函数,计算复杂度可以从O(kn 2)降低到O(kn)。
步骤八、将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图。
本方案中,所述预测模型为根据用户的历史进线数据进行训练得到。
优选地,本申请较佳实施例中,所述预先训练的预测模型为Deep and wide模型。所述Deep and wide模型由2部分组成:线性softmax回归模型和DNN神经网络模型。
所述Softmax回归模型能够保存原始离散特征,DNN神经网络模型能够对原始特征进线非线性变换得到新的特征,再利用原始特征+新特征进线组合串联得到新特征进线模型预测。
Deep and wide模型可以比传统的单独softmax回归模型、DNN神经网络模型达到更好的效果。
优选地,在本申请较佳实施例中,所述预测模型包括四个。其中,第一预测模型为利用相邻进线行为间隔时间在第一预设时间,如1天以内的历史进线数据训练得到;第二预测模型为利用相邻进线行为间隔时间大于第一预设时间,如1天且小于等于第二预设时间,如3天以内的历史进线数据训练得到;第三预测模型为利用相邻进线行为间隔时间大于第二预设时间,如3天且小于等于第三预设时间,如7天以内的历史进线数据训练得到;及第四预测模型为利用相邻进线行为间隔时间大于第三预设时间,如7天的历史进线数据训练得到。
因此,优选地,步骤八还包括:
子步骤1、计算所述电话号码本次进线与上述进线的时间间隔。
子步骤2、判断所述时间间隔是否在1天之内。
若子步骤2中,判断所述时间间隔是在1天之内,则执行子步骤3:将所述预测数据输入预先训练的第一预测模型中,预测客户本次进线的意图。
若子步骤2中,判断所述时间间隔不在1天之内,则执行子步骤4:判断所述时间间隔是否在大于1天且小于等于3天之内。
若在子步骤4中,判断所述时间间隔是在大于1天且小于等于3天之内,则执行子步骤5:将所述预测数据输入预先训练的第二预测模型中,预测客户本次进线的意图。
若在子步骤4中,判断所述时间间隔不在大于1天且小于等于3天之内,则执行子步骤6:判断所述时间间隔是否在大于3天且小于等于7天之内。
若子步骤6中,判断所述时间间隔在大于3天且小于等于7天之内,则执行子步骤7:将所述预测数据输入预先训练的第三预测模型中,预测客户本次进线的意图。
若子步骤6中,判断所述时间间隔不在大于3天且小于等于7天之内,则执行子步骤8:将所述预测数据输入预先训练的第四预测模型中,预测客户本次进线的意图。
步骤九、根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
优选地,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
进一步地,所述预设规则还包括:仅播报选择概率大于或者等于预定阈值的业务对应的功能按键,对选择概率小于所述预定阈值的业务对应的功能按键可以进行隐藏。
例如,根据监管要求,预定业务,如车险报案的按键以及一账通用户对应的功能按键需要且优先播报,对于其他业务对应的功能按键,其预测的概率大于等于0.85的,从最高概率排名依次播报,预测的概率小于0.85的业务 可以进行隐藏,并通过让用户选择了播报全部业务的按键时,按照预测的概率排名依次播报其对应功能按键。
又如,根据所述预测的客户本次进线的意图,用户选择办理信用卡业务的可能性为90%,选择办理银行业务的可能性为88%,选择办理产险业务的可能性为86%,选择办理其他业务的可能性均小于85%,则所述自动语音应答系统的IVR菜单可能进播报如下语音:
车险报案请按1;一账通业务请按2;信用卡业务请按9;银行业务请按3;产险业务请按4;听取全部业务功能按键请按0等。
可选地,在其他实施例中,按键预测程序还可以被分割为一个或者多个模块,一个或者多个模块被存储于存储器11中,并由一个或多个处理器(本实施例为处理器12)所执行以完成本申请,本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段,用于描述按键预测程序在按键预测装置中的执行过程。
例如,参照图4所示,为本申请按键预测装置一实施例中的按键预测程序的程序模块示意图,该实施例中,所述按键预测程序可以被分割为客户识别模块10、客户特征获取模块20、预测数据计算模块30以及按键预测模块40,示例性地:
所述客户识别模块10用于:接收进线,识别该次进线的电话号码,并根据所述电话号码,从一个客户数据库中获取该电话号码所属客户的基本信息。
所述客户特征获取模块20用于:根据所述电话号码,从一个历史行为数据库中获取该电话号码在预设时间段的进线行为记录,并根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性,以及根据所述客户的基本信息获知所述客户的用户属性、所述客户在预设时间段内通过不同渠道执行的过往行为属性,并分析本次进线的进线特征属性。
所述预测数据计算模块30用于:将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据。
所述按键预测模块40用于:将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图,并根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
上述客户识别模块10、客户特征获取模块20、预测数据计算模块30以及按键预测模块40等程序模块被执行时所实现的功能或操作步骤与上述实施例大体相同,在此不再赘述。
此外,本申请实施例还提出一种计算机可读存储介质,所述计算机可读存储介质上存储有按键预测程序,所述按键预测程序可被一个或多个处理器执行,以实现如下操作:
接收进线,识别该次进线的电话号码,并根据所述电话号码,从一个客户数据库中获取该电话号码所属客户的基本信息;
根据所述电话号码,从一个历史行为数据库中获取该电话号码在预设时间段的进线行为记录,并根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性;
根据所述客户的基本信息获知所述客户的用户属性、所述客户在预设时间段内通过不同渠道执行的过往行为属性,并分析本次进线的进线特征属性;
将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据;
将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图;
根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
本申请计算机可读存储介质具体实施方式与上述按键预测装置和方法各实施例基本相同,在此不作累述。
需要说明的是,上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。并且本文中的术语“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、装置、物品或者方法不仅包括那些要素,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、装置、物品或者方法所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种按键预测方法,其特征在于,所述方法包括:
    接收电话进线,识别该次进线的电话号码,并根据所述电话号码,从数据库中获取该电话号码所属客户的基本信息;
    根据所述电话号码,从数据库中获取该电话号码在预设时间段的进线行为记录,并根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性;
    根据所述客户的基本信息获知所述客户的用户属性、所述客户在预设时间段内通过不同渠道执行的过往行为属性,并分析本次进线的进线特征属性;
    将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据;及
    将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图,并根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
  2. 如权利要求1所述的按键预测方法,其特征在于,所述预先训练的预测模型为Deep and wide模型,其中,所述Deep and wide模型包括线性softmax回归模型和DNN神经网络模型。
  3. 如权利要求1所述的按键预测方法,其特征在于,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
    所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
    所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
    所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
    所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
  4. 如权利要求2所述的按键预测方法,其特征在于,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
    所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
    所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
    所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
    所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
  5. 如权利要求1所述的按键预测方法,其特征在于,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
  6. 如权利要求2-4任一项所述的按键预测方法,其特征在于,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
  7. 如权利要求6所述的按键预测方法,其特征在于,所述预设规则还包括:仅播报选择概率大于或者等于预定阈值的业务对应的功能按键,对选择概率小于所述预定阈值的业务对应的功能按键进行隐藏。
  8. 一种按键预测装置,其特征在于,所述装置包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的按键预测程序,所述按键预测程序被所述处理器执行时实现如下步骤:
    接收电话进线,识别该次进线的电话号码,并根据所述电话号码,从数据库中获取该电话号码所属客户的基本信息;
    根据所述电话号码,从数据库中获取该电话号码在预设时间段的进线行为记录,并根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性;
    根据所述客户的基本信息获知所述客户的用户属性、所述客户在预设时间段内通过不同渠道执行的过往行为属性,并分析本次进线的进线特征属性;
    将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据;及
    将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图,并根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
  9. 如权利要求8所述的按键预测装置,其特征在于,所述预先训练的预测模型为Deep and wide模型,其中,所述Deep and wide模型包括线性softmax回归模型和DNN神经网络模型。
  10. 如权利要求8所述的按键预测装置,其特征在于,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
    所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
    所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
    所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
    所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
  11. 如权利要求9所述的按键预测装置,其特征在于,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
    所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
    所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
    所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
    所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
  12. 如权利要求8所述的按键预测装置,其特征在于,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
  13. 如权利要求9-11任一项所述的按键预测装置,其特征在于,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
  14. 如权利要求13所述的按键预测装置,其特征在于,所述预设规则还包括:仅播报选择概率大于或者等于预定阈值的业务对应的功能按键,对选择概率小于所述预定阈值的业务对应的功能按键进行隐藏。
  15. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有按键预测程序,所述按键预测程序可被一个或者多个处理器执行,以实现如下步骤:
    接收电话进线,识别该次进线的电话号码,并根据所述电话号码,从数据库中获取该电话号码所属客户的基本信息;
    根据所述电话号码,从数据库中获取该电话号码在预设时间段的进线行为记录,并根据所述预设时间段的所有进线行为记录,得到所述客户的进线意图属性;
    根据所述客户的基本信息获知所述客户的用户属性、所述客户在预设时间段内通过不同渠道执行的过往行为属性,并分析本次进线的进线特征属性;
    将上述得到的用户属性、过往行为属性、进线意图属性以及本次进线的进线特征属性进行特征组合,得到预测数据;及
    将所述预测数据输入预先训练的预测模型中,预测客户本次进线的意图,并根据所述预测的客户本次进线的意图,按照预设规则对自动语音应答菜单的语音播报顺序进行自动调整。
  16. 如权利要求15所述的计算机可读存储介质,其特征在于,所述预先训练的预测模型为Deep and wide模型,其中,所述Deep and wide模型包括线性softmax回归模型和DNN神经网络模型。
  17. 如权利要求15所述的计算机可读存储介质,其特征在于,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
    所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
    所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
    所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
    所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
  18. 如权利要求16所述的计算机可读存储介质,其特征在于,所述预测模型包括第一预测模型、第二预测模型、第三预测模型以及第四预测模型,其中:
    所述第一预测模型为利用相邻进线行为间隔时间在第一预设时间以内的历史进线数据训练得到;
    所述第二预测模型为利用相邻进线行为间隔时间大于第一预设时间且小于等于第二预设时间以内的历史进线数据训练得到;
    所述第三预测模型为利用相邻进线行为间隔时间大于第二预设时间且小于等于第三预设时间以内的历史进线数据训练得到;及
    所述第四预测模型为利用相邻进线行为间隔时间大于第三预设时间的历史进线数据训练得到。
  19. 如权利要求8-11任一项所述的计算机可读存储介质,其特征在于,所述预设规则包括:优先播报预定业务的功能按键,对于其他业务对应的功能按键,按照预测的客户本次进线的意图中各种业务的选择概率从大到小的顺序依次播报各种业务对应的功能按键。
  20. 如权利要求19所述的计算机可读存储介质,其特征在于,所述预设规则还包括:仅播报选择概率大于或者等于预定阈值的业务对应的功能按键,对选择概率小于所述预定阈值的业务对应的功能按键进行隐藏。
PCT/CN2018/123596 2018-09-19 2018-12-25 按键预测方法、装置及计算机可读存储介质 WO2020056978A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811096236.4A CN109274843B (zh) 2018-09-19 2018-09-19 按键预测方法、装置及计算机可读存储介质
CN201811096236.4 2018-09-19

Publications (1)

Publication Number Publication Date
WO2020056978A1 true WO2020056978A1 (zh) 2020-03-26

Family

ID=65197763

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/123596 WO2020056978A1 (zh) 2018-09-19 2018-12-25 按键预测方法、装置及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN109274843B (zh)
WO (1) WO2020056978A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110233944B (zh) * 2019-06-21 2020-12-18 携程旅游信息技术(上海)有限公司 交互式语音应答的方法、系统、电子设备和介质
CN110602333B (zh) * 2019-08-23 2021-08-03 绍兴文理学院 基于深度学习的呼叫中心应答系统及方法
CN111833072A (zh) * 2019-08-28 2020-10-27 北京嘀嘀无限科技发展有限公司 一种播报事项的推送方法、推送装置及可读存储介质
CN111614845A (zh) * 2020-04-21 2020-09-01 深圳追一科技有限公司 人机对话中人设匹配方法、装置、计算机设备和存储介质
CN111787164B (zh) * 2020-07-16 2021-06-15 阳光保险集团股份有限公司 来电提示方法、装置、电子设备和计算机可读介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070121837A1 (en) * 2000-03-21 2007-05-31 Sbc Properties, L.P., Of Reno Nevada Customer-centric interface and method of designing an interface
CN104735237A (zh) * 2013-12-20 2015-06-24 中国移动通信集团公司 一种路径预测方法及交互式语音应答ivr路径预测系统
CN106993104A (zh) * 2016-11-14 2017-07-28 阿里巴巴集团控股有限公司 来电处理方法、装置及终端
CN107155010A (zh) * 2016-03-02 2017-09-12 北京京东尚科信息技术有限公司 基于大数据处理用户语音呼叫的方法、装置和系统
CN107704946A (zh) * 2017-08-29 2018-02-16 平安科技(深圳)有限公司 电子装置、语音导航需求预测方法和存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8661112B2 (en) * 2002-12-20 2014-02-25 Nuance Communications, Inc. Customized interactive voice response menus
CN101094271A (zh) * 2007-07-20 2007-12-26 华为技术有限公司 自动语音应答系统菜单的生成方法及装置
CN106572272A (zh) * 2015-10-12 2017-04-19 阿里巴巴集团控股有限公司 一种ivr语音菜单确定方法及装置
CN106993089B (zh) * 2017-03-23 2020-01-31 中国联合网络通信集团有限公司 语音菜单显示的方法与装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070121837A1 (en) * 2000-03-21 2007-05-31 Sbc Properties, L.P., Of Reno Nevada Customer-centric interface and method of designing an interface
CN104735237A (zh) * 2013-12-20 2015-06-24 中国移动通信集团公司 一种路径预测方法及交互式语音应答ivr路径预测系统
CN107155010A (zh) * 2016-03-02 2017-09-12 北京京东尚科信息技术有限公司 基于大数据处理用户语音呼叫的方法、装置和系统
CN106993104A (zh) * 2016-11-14 2017-07-28 阿里巴巴集团控股有限公司 来电处理方法、装置及终端
CN107704946A (zh) * 2017-08-29 2018-02-16 平安科技(深圳)有限公司 电子装置、语音导航需求预测方法和存储介质

Also Published As

Publication number Publication date
CN109274843B (zh) 2021-06-11
CN109274843A (zh) 2019-01-25

Similar Documents

Publication Publication Date Title
WO2020056978A1 (zh) 按键预测方法、装置及计算机可读存储介质
CN107818344B (zh) 用户行为进行分类和预测的方法和系统
US9564040B1 (en) Systems and methods for automated alerts
US11803884B2 (en) System and methods for automatically generating regulatory compliance manual using modularized and taxonomy-based classification of regulatory obligations
CN111178705B (zh) 事项评价方法、装置、设备及存储介质
US9430534B2 (en) Systems and methods for improved security and precision in executing analytics using SDKS
WO2017133615A1 (zh) 一种业务参数获取方法及装置
US20080104037A1 (en) Automated scheme for identifying user intent in real-time
CN111506723B (zh) 问答响应方法、装置、设备及存储介质
CN110825956A (zh) 一种信息流推荐方法、装置、计算机设备及存储介质
WO2019041523A1 (zh) 电子装置、语音导航需求预测方法和存储介质
CN103493089A (zh) 跟踪社交网络中的馈送
CN107341187A (zh) 搜索处理方法、装置、设备和计算机存储介质
CN110852785B (zh) 用户分级方法、装置及计算机可读存储介质
CN111400465B (zh) 客服机器人的生成方法、装置、电子设备及介质
US11586618B2 (en) Query parsing from natural language questions supported by captured subject matter knowledge
US20210334593A1 (en) Recommending scripts for constructing machine learning models
CN112308590B (zh) 一种参数处理方法、装置以及计算设备、存储介质
CN117522538A (zh) 招投标信息处理方法、装置、计算机设备及存储介质
CN116661936A (zh) 页面数据的处理方法、装置、计算机设备及存储介质
CN116132589A (zh) 一种电力客服系统语音反馈菜单的优化方法及系统
CN116501979A (zh) 信息推荐方法、装置、计算机设备及计算机可读存储介质
CN115525192A (zh) 面向用户的报价计费方法、装置、计算机设备及存储介质
CN113190562A (zh) 一种报表生成方法、装置及电子设备
CN116756398A (zh) 标签体系构建系统及标签体系构建方法

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: 18934149

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18934149

Country of ref document: EP

Kind code of ref document: A1