CN111611351A

CN111611351A - Control method and device for online customer service session and electronic equipment

Info

Publication number: CN111611351A
Application number: CN201910138876.5A
Authority: CN
Inventors: 陈永强
Original assignee: Beijing Didi Infinity Technology and Development Co Ltd
Current assignee: Beijing Didi Infinity Technology and Development Co Ltd
Priority date: 2019-02-25
Filing date: 2019-02-25
Publication date: 2020-09-01
Anticipated expiration: 2039-02-25
Also published as: CN111611351B

Abstract

The application provides a method and a device for controlling an online customer service session and electronic equipment, wherein the method comprises the following steps: monitoring the current online customer service session; if the service provider does not receive the message of the user within the preset time length after sending the current message, session information corresponding to the customer service session is obtained; the session information comprises session content and/or session statistical parameters in the customer service session; inputting the session information into a probability prediction model which is trained in advance to obtain the continuous probability of the customer service session; controlling an online state of the customer service session based on the continuation probability. According to the method and the device, the probability of the response of the user can be predicted according to the session information of the customer service session through the probability prediction model, and the online state of the session is controlled based on the probability; the method can avoid disconnection of the session when the user problem is not solved, can also avoid the problem that the customer service wastes time to wait for the response of the user after the user problem is solved, and improves the working efficiency of the customer service while ensuring the user experience.

Description

Control method and device for online customer service session and electronic equipment

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a method and a device for controlling an online customer service session and electronic equipment.

Background

The online customer service of the service providing system can provide various services for the user, solve the user's question in time and meet the user's requirements. In order to avoid waste of customer service resources caused by long-time occupation of online customer service by a certain user, in the related technology, when the user and the online customer service are in conversation, if the user does not respond any more, the conversation is not actively disconnected, usually, the system automatically disconnects the conversation in a mode of disconnection overtime, but the mode may cause that the requirements of the user are not solved, and in addition, the waiting time of disconnection overtime is long, the time of manual customer service is still wasted, so that the user quantity which can be served by the manual customer service is reduced, and the working efficiency of the manual customer service is reduced.

Disclosure of Invention

In view of this, an object of the embodiments of the present application is to provide a method and an apparatus for controlling an online customer service session, and an electronic device, so as to prevent a user from disconnecting the session when a problem is not solved, and also prevent the customer service from wasting time and waiting for a user response after the problem is solved, thereby improving the working efficiency of the customer service while ensuring the user experience.

According to one aspect of the present application, an electronic device is provided that may include one or more storage media and one or more processors in communication with the storage media. One or more storage media store machine-readable instructions executable by a processor. When the electronic device is operated, the processor communicates with the storage medium through the bus, and the processor executes the machine readable instructions to perform one or more of the following operations:

a method of controlling an online customer service session, the method comprising: monitoring the current online customer service session; the customer service session is a session between a user and a service provider; if the service provider does not receive the message of the user within the preset time length after sending the current message, session information corresponding to the customer service session is obtained; the session information comprises session content and/or session statistical parameters in the customer service session; inputting the session information into a probability prediction model which is trained in advance to obtain the continuous probability of the customer service session; controlling an online state of the customer service session based on the continuation probability.

In some embodiments, the step of acquiring session information corresponding to the customer service session if the service provider does not receive the message of the user within a preset time period after sending the current message includes: when the service provider sends the current message, starting a timer; the timing duration of the timer is a preset duration; monitoring whether a message of a user is received within the timing duration of the timer; and if not, acquiring the session information corresponding to the customer service session.

In some embodiments, the session content includes: taking the current message of the service provider as a reference, and continuously obtaining at least one part of messages before the current message and the current message; the session statistical parameters include: the interval duration between two adjacent messages in the customer service session, the duration of the current message from the current time, the number of messages of the user and the number of messages of the service provider.

In some embodiments, the probabilistic predictive model is trained by: extracting session data from a preset session database; determining a plurality of training samples based on the session data; each training sample comprises a session identifier, and session content, a session closing reason and session statistical parameters corresponding to the session identifier; training a preset initial model based on a training sample to obtain a probability prediction model; wherein the initial model comprises: convolutional networks, recursive networks, and loss functions.

In some embodiments, the step of determining a plurality of training samples based on the session data includes: dividing the session data into a plurality of pieces according to the session identification in the extracted session data; each session data corresponds to a session identifier; and extracting the session identification contained in each session data, and obtaining the training sample corresponding to each session data by the session content, the session closing reason and the session statistical parameters corresponding to the session identification.

In some embodiments, after obtaining the training sample corresponding to each session data, the method further includes: for each training sample, judging whether the training sample contains the message of the user and the message of the session service provider; if not, the training sample is rejected.

In some embodiments, the step of training the preset initial model based on the training samples includes: determining a current training sample from a plurality of training samples, inputting the conversation content of the current training sample into a convolution network of a preset initial model, and outputting a feature matrix corresponding to the conversation content; inputting the characteristic matrix and the conversation statistical parameter into a recursive network of the initial model, and outputting a characteristic value corresponding to the current training sample; inputting the characteristic value into a loss function of the initial model, and outputting a prediction result of a session closing reason of the current training sample; updating the prediction accuracy of the initial model according to the session closing reason of the current training sample and the prediction result of the session closing reason; judging whether to terminate the training based on preset conditions, wherein the preset conditions comprise: the prediction accuracy is greater than or equal to a preset accuracy threshold, and/or the number of current training samples input into the convolutional network is equal to a preset iteration threshold; if not, determining the next current training sample from the plurality of training samples, and continuing to train the convolutional network, the recursive network and the loss function; if yes, the initial model after the current training sample is trained is used as a probability prediction model.

In some embodiments, the convolutional network comprises a 3D convolutional neural network; the step of inputting the session content of the current training sample into the convolutional network of the preset initial model and outputting the feature matrix corresponding to the session content includes: and mapping the conversation content of the current training sample through a 3D convolutional neural network to obtain a characteristic matrix corresponding to the conversation content.

In some embodiments, the recursive network includes a predetermined number of gated cyclic unit networks; the gate control circulation unit networks with the preset number of layers are sequentially connected; the number of layers of the gated cyclic unit network corresponds to the number of messages in the session content; the step of inputting the feature matrix and the session statistic parameters into the recursive network of the initial model and outputting the feature value corresponding to the current training sample includes: inputting a feature matrix of a message corresponding to a first layer of gated cyclic unit network in the feature matrix and session statistical parameters of session content into the first layer of gated cyclic unit network to obtain an output result of the first layer of gated cyclic unit network; for each layer of gated cyclic unit network except the first layer of gated cyclic unit network, inputting the output result of the previous layer of gated cyclic unit network of the layer of gated cyclic unit network, the feature matrix of the message corresponding to the layer of gated cyclic unit network and the session statistical parameters of the session content into the layer of gated cyclic unit network until the last layer of gated cyclic unit network; and taking the output result of the last layer of gated loop unit network as a characteristic value corresponding to the current training sample.

In some embodiments, the loss function comprises a softmax function; the number of the characteristic values corresponding to the current training sample is multiple; the step of inputting the feature value into the loss function of the initial model and outputting the prediction result of the session closure reason of the current training sample includes: calculating an exponential function value of each characteristic value through a softmax function; determining the probability of each characteristic value according to the index function value of each characteristic value and the sum of the index function values of each characteristic value of the current training sample; and outputting the probability of the specified characteristic value in each characteristic value, and determining the probability of the specified characteristic value and the session closing reason corresponding to the specified characteristic value as the prediction result of the session closing reason of the current training sample.

In some embodiments, the step of controlling the online state of the customer service session based on the continuation probability includes: and if the continuing probability is smaller than a preset probability threshold value, disconnecting the customer service session or instructing the service provider to disconnect the customer service session.

In some embodiments, the step of controlling the online state of the customer service session based on the continuation probability includes: if the continuing probability is equal to or higher than a preset probability threshold, monitoring whether a message of the user is received within a preset timeout time threshold; if not, the customer service session is disconnected or the service provider is instructed to disconnect the customer service session.

In some embodiments, the step of controlling the online state of the customer service session based on the continuation probability includes: if the continuing probability is equal to or higher than a preset probability threshold, monitoring whether a message of the user is received within a preset timeout time threshold; if not, continuing to predict the new continuing probability of the customer service session through the probability prediction model; and when the new continuation probability is lower than a preset probability threshold value, disconnecting the customer service session or instructing the service provider to disconnect the customer service session.

According to another aspect of the present application, there is also provided a control apparatus for an online customer service session, the apparatus including: the session monitoring module is used for monitoring the current online customer service session; the customer service session is a session between a user and a service provider; the information acquisition module is used for acquiring session information corresponding to the customer service session if the service provider does not receive the message of the user within a preset time length after sending the current message; the session information comprises session content and/or session statistical parameters in the customer service session; the probability prediction module is used for inputting the session information into a probability prediction model which is trained in advance to obtain the continuous probability of the customer service session; and the session control module is used for controlling the online state of the customer service session based on the continuation probability.

In some embodiments, the information obtaining module is configured to: when the service provider sends the current message, starting a timer; the timing duration of the timer is a preset duration; monitoring whether a message of a user is received within the timing duration of the timer; and if not, acquiring the session information corresponding to the customer service session.

In some embodiments, the apparatus further comprises a model training module configured to: extracting session data from a preset session database; determining a plurality of training samples based on the session data; each training sample comprises a session identifier, and session content, a session closing reason and session statistical parameters corresponding to the session identifier; training a preset initial model based on a training sample to obtain a probability prediction model; wherein the initial model comprises: convolutional networks, recursive networks, and loss functions.

In some embodiments, the model training module is configured to: dividing the session data into a plurality of pieces according to the session identification in the extracted session data; each session data corresponds to a session identifier; and extracting the session identification contained in each session data, and obtaining the training sample corresponding to each session data by the session content, the session closing reason and the session statistical parameters corresponding to the session identification.

In some embodiments, the above apparatus further comprises: the message judging module is used for judging whether each training sample contains the message of the user and the message of the session service provider; and the sample removing module is used for removing the training sample if the training sample does not contain the message of the user and the message of the session service provider.

In some embodiments, the model training module is configured to: determining a current training sample from a plurality of training samples, inputting the conversation content of the current training sample into a convolution network of a preset initial model, and outputting a feature matrix corresponding to the conversation content; inputting the characteristic matrix and the conversation statistical parameter into a recursive network of the initial model, and outputting a characteristic value corresponding to the current training sample; inputting the characteristic value into a loss function of the initial model, and outputting a prediction result of a session closing reason of the current training sample; updating the prediction accuracy of the initial model according to the session closing reason of the current training sample and the prediction result of the session closing reason; judging whether to terminate the training based on preset conditions, wherein the preset conditions comprise: the prediction accuracy is greater than or equal to a preset accuracy threshold, and/or the number of current training samples input into the convolutional network is equal to a preset iteration threshold; if not, determining the next current training sample from the plurality of training samples, and continuing to train the convolutional network, the recursive network and the loss function; if yes, the initial model after the current training sample is trained is used as a probability prediction model.

In some embodiments, the convolutional network comprises a 3D convolutional neural network; the model training module is configured to: and mapping the conversation content of the current training sample through a 3D convolutional neural network to obtain a characteristic matrix corresponding to the conversation content.

In some embodiments, the recursive network includes a predetermined number of gated cyclic unit networks; the gate control circulation unit networks with the preset number of layers are sequentially connected; the number of layers of the gated cyclic unit network corresponds to the number of messages in the session content; the model training module is configured to: inputting a feature matrix of a message corresponding to a first layer of gated cyclic unit network in the feature matrix and session statistical parameters of session content into the first layer of gated cyclic unit network to obtain an output result of the first layer of gated cyclic unit network; for each layer of gated cyclic unit network except the first layer of gated cyclic unit network, inputting the output result of the previous layer of gated cyclic unit network of the layer of gated cyclic unit network, the feature matrix of the message corresponding to the layer of gated cyclic unit network and the session statistical parameters of the session content into the layer of gated cyclic unit network until the last layer of gated cyclic unit network; and taking the output result of the last layer of gated loop unit network as a characteristic value corresponding to the current training sample.

In some embodiments, the loss function comprises a softmax function; the number of the characteristic values corresponding to the current training sample is multiple; the model training module is configured to: calculating an exponential function value of each characteristic value through a softmax function; determining the probability of each characteristic value according to the index function value of each characteristic value and the sum of the index function values of each characteristic value of the current training sample; and outputting the probability of the specified characteristic value in each characteristic value, and determining the probability of the specified characteristic value and the session closing reason corresponding to the specified characteristic value as the prediction result of the session closing reason of the current training sample.

In some embodiments, the session control module is configured to: and if the continuing probability is smaller than a preset probability threshold value, disconnecting the customer service session or instructing the service provider to disconnect the customer service session.

In some embodiments, the session control module is configured to: if the continuing probability is equal to or higher than a preset probability threshold, monitoring whether a message of the user is received within a preset timeout time threshold; if not, the customer service session is disconnected or the service provider is instructed to disconnect the customer service session.

In some embodiments, the session control module is configured to: if the continuing probability is equal to or higher than a preset probability threshold, monitoring whether a message of the user is received within a preset timeout time threshold; if not, continuing to predict the new continuing probability of the customer service session through the probability prediction model; and when the new continuation probability is lower than a preset probability threshold value, disconnecting the customer service session or instructing the service provider to disconnect the customer service session.

According to another aspect of the present application, there is also provided an electronic device, including: the system comprises a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, when the electronic device runs, the processor and the storage medium are communicated through the bus, and the processor executes the machine-readable instructions to execute the steps of the control method of the online customer service session.

According to another aspect of the present application, there is also provided a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, performs the steps of the method of controlling an online customer service session as described above.

Based on any aspect, if the current online customer service session is monitored, the service provider does not receive the message of the user within the preset time after sending the current message, and session information corresponding to the customer service session is acquired; and then inputting the session information into a probability prediction model to obtain the continuous probability of the customer service session, and controlling the online state of the customer service session based on the continuous probability. In the method, the probability prediction model can predict the probability of the response of the user according to the session information of the customer service session, namely the continuous probability of the customer service session, and controls the online state of the session based on the continuous probability; compared with the overtime disconnection mode in the related technology, the mode can avoid disconnection of the session when the user problem is not solved, can also avoid the problem that the customer service wastes time to wait for the response of the user after the user problem is solved, and improves the working efficiency of the customer service while ensuring the user experience.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.

Fig. 1 is a block diagram illustrating a service providing system provided by an embodiment of the present application;

FIG. 2 illustrates a schematic diagram of exemplary hardware and software components of an electronic device provided by embodiments of the present application;

FIG. 3 is a flow chart illustrating a method for controlling an online customer service session according to an embodiment of the present application;

FIG. 4 is a flow chart illustrating a method for training a probabilistic predictive model provided in an embodiment of the present application;

FIG. 5 is a flow chart illustrating another method for training a probabilistic predictive model provided in an embodiment of the present application;

FIG. 6 is a flow chart illustrating another method for controlling an online customer service session according to an embodiment of the present application;

fig. 7 is a schematic structural diagram illustrating a control apparatus for an online customer service session according to an embodiment of the present application.

Detailed Description

In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it should be understood that the drawings in the present application are for illustrative and descriptive purposes only and are not used to limit the scope of protection of the present application. Additionally, it should be understood that the schematic drawings are not necessarily drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of the present application. It should be understood that the operations of the flow diagrams may be performed out of order, and steps without logical context may be performed in reverse order or simultaneously. One skilled in the art, under the guidance of this application, may add one or more other operations to, or remove one or more operations from, the flowchart.

In addition, the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present application without making any creative effort, shall fall within the protection scope of the present application.

In order to enable a person skilled in the art to use the present disclosure, the following embodiments are given in connection with a specific application scenario "car rental service". It will be apparent to those skilled in the art that the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the application. Although the present application is described primarily in the context of a rental car service, it should be understood that this is merely one exemplary embodiment. The application can also be applied to providing other service systems, such as a system for sending and/or receiving express delivery, a service system for business transaction between buyers and sellers, and the like.

It should be noted that in the embodiments of the present application, the term "comprising" is used to indicate the presence of the features stated hereinafter, but does not exclude the addition of further features.

Fig. 1 is a block diagram of a service providing system 100 of some embodiments of the present application. For example, the service providing system 100 may be an online transportation service platform for transportation services such as taxi cab, designated drive service, express, carpool, bus service, driver rental, or regular service, or any combination thereof. The service providing system 100 may include one or more of a server 110, a network 120, a user terminal 130, a service provider terminal 140, and a database 150, and the server 110 may include a processor therein to perform an instruction operation.

In some embodiments, the server 110 may be a single server or a group of servers. The set of servers can be centralized or distributed (e.g., the servers 110 can be a distributed system). In some embodiments, the server 110 may be local or remote to the terminal. For example, server 110 may access information and/or data stored in user terminal 130, service provider terminal 140, or database 150, or any combination thereof, via network 120. As another example, the server 110 may be directly connected to at least one of the user terminal 130, the service provider terminal 140, and the database 150 to access stored information and/or data. In some embodiments, the server 110 may be implemented on a cloud platform. In some embodiments, the server 110 may be implemented on an electronic device 200 having one or more of the components shown in FIG. 2 in the present application. In some embodiments, the server 110 may include a processor. The processor may process information and/or data related to the service request to perform one or more of the functions described herein.

Network 120 may be used for the exchange of information and/or data. In some embodiments, one or more components in the service providing system 100 (e.g., the server 110, the user terminal 130, the service provider terminal 140, and the database 150) may send information and/or data to other components. For example, server 110 may obtain a service request from user terminal 130 via network 120. In some embodiments, the network 120 may be any type of wired or wireless network, or combination thereof. By way of example only, network 120 may include a wired network, a wireless network, a fiber optic network, a telecommunications network, and the like. In some embodiments, network 120 may include one or more network access points. For example, network 120 may include wired or wireless network access points, such as base stations and/or network switching nodes, through which one or more components of service providing system 100 may connect to network 120 to exchange data and/or information.

In some embodiments, "user" and "user terminal" may be used interchangeably, and "service provider" and "service provider terminal" may be used interchangeably. In some embodiments, user terminal 130 may comprise a mobile device, a tablet computer, a laptop computer, or a built-in device in a motor vehicle, among others. In some embodiments, the user terminal 130 may be a device having a positioning technology for locating a user and/or a location of the user terminal.

In some embodiments, the service provider terminal 140 may be a similar or the same device as the user terminal 130. In some embodiments, the service provider terminal 140 may be a device with location technology for locating the location of the service provider and/or the service provider terminal. In some embodiments, the user terminal 130 and/or the service provider terminal 140 may communicate with other positioning devices to determine the location of the user, the user terminal 130, the service provider, or the service provider terminal 140, or any combination thereof. In some embodiments, the user terminal 130 and/or the service provider terminal 140 may send the positioning information to the server 110.

Database 150 may store data and/or instructions. In some embodiments, the database 150 may store data obtained from the user terminal 130 and/or the service provider terminal 140. Database 150 may store data and/or instructions for the exemplary methods described herein. The database 150 may include mass storage, removable storage, volatile Read-and-write Memory, or Read-Only Memory (ROM), among others. Database 150 may be implemented on a cloud platform.

In some embodiments, a database 150 may be connected to the network 120 to communicate with one or more components in the service providing system 100 (e.g., the server 110, the user terminal 130, the service provider terminal 140, etc.). One or more components in the service providing system 100 may access data or instructions stored in the database 150 via the network 120. In some embodiments, the database 150 may be directly connected to one or more components in the service providing system 100 (e.g., the server 110, the user terminal 130, the service provider terminal 140, etc.); alternatively, the database 150 may be part of the server 110.

Fig. 2 illustrates a schematic diagram of exemplary hardware and software components of an electronic device 200 of a server 110, a user terminal 130, a service provider terminal 140, which may implement the concepts of the present application, according to some embodiments of the present application. For example, a processor may be used on the electronic device 200 and to perform the functions herein.

The electronic device 200 may be a general-purpose computer or a special-purpose computer, both of which may be used to implement the control method of the online customer service session of the present application. Although only a single computer is shown, for convenience, the functions described herein may be implemented in a distributed fashion across multiple similar platforms to balance processing loads.

For example, the electronic device 200 may include a network port 210 connected to a network, one or more processors 220 for executing program instructions, a communication bus 230, and a different form of storage medium 240, such as a disk, ROM, or RAM, or any combination thereof. Illustratively, the computer platform may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application may be implemented in accordance with these program instructions. The electronic device 200 also includes an Input/Output (I/O) interface 250 between the computer and other Input/Output devices (e.g., keyboard, display screen).

For ease of illustration, only one processor is depicted in the electronic device 200. However, it should be noted that the electronic device 200 in the present application may also comprise a plurality of processors, and thus the steps performed by one processor described in the present application may also be performed by a plurality of processors in combination or individually. For example, if the processor of the electronic device 200 executes steps a and B, it should be understood that steps a and B may also be executed by two different processors together or separately in one processor. For example, a first processor performs step a and a second processor performs step B, or the first processor and the second processor perform steps a and B together.

Based on the description of the electronic device, an embodiment of the present application first describes a method for controlling an online customer service session, as shown in fig. 3, the method includes the following steps:

step S302, monitoring the current online customer service session; wherein, the customer service session is a session between a user and a service provider;

the service provider can specifically provide online customer service, manual customer service and the like of the system for the service; the service providing system may be a car rental system, a shopping system, an education system, an administrative system, etc.; the user can enjoy the corresponding service by logging in the system, so the user can also be called a service user. In the process of using the service provided by the service providing system, a user often has some questions or disputes, and at this time, the service providing system needs to be connected to solve the questions or disputes. The user can specifically click the on-line customer service or manual customer service related button of the webpage or application page of the service providing system, and the service provider can be connected.

When a user connects with a service provider, if the number of the service providers is small and the number of the users connecting with the service provider is large, the user may need to queue to connect with the service provider and enter a customer service session with the service provider. When both the user and the service provider enter a session, the customer service session can be started. In practical implementation, a session window may be set on the web page or the application page for the user to perform a session with the service provider. A customer service session typically contains all messages that the user has switched on with the service provider and then switched off until the session, and a customer service session typically has a unique session identification associated with it. In the conversation process, the system can monitor the current online customer service conversation in real time in the whole course, namely monitor the messages sent by the user and the service provider (certainly, the messages sent by the system can also be included), and wait for the other party to reply the messages after one party sends the messages; the system can also listen to the message returned by the user only after the service provider sends the message, wait for the time for the user to return the message, and the like.

Step S304, if the service provider does not receive the message of the user within the preset time after sending the current message, acquiring the session information corresponding to the customer service session; wherein, the session information comprises session content and/or session statistical parameters in the customer service session;

when a service provider sends a current message, a user sometimes cannot reply the message in time, and at the moment, two possibilities exist, namely the question of the user is already answered, the session is ended, but the user does not actively disconnect the session; the other is that the user may not have viewed the message sent by the service provider, or the message sent by the service provider does not resolve the user's question, but the user is busy with something else to reply to the message.

In the related technology, if a service provider does not receive the message of a user for a long time after sending the current message, the system automatically disconnects the session; however, the method does not consider the possible reason that the user does not reply the message, on one hand, the question of the user is not answered, the user experience is affected, and on the other hand, the service provider is enabled to be idle for a long time, so that the working efficiency of the service provider is reduced; based on this, in this embodiment, it is supposed that the user is asked to answer through the messages that have been sent by both parties in the current customer service session, and the probability that the session should be ended is inferred, and then based on the probability, it is determined whether to disconnect the session.

Specifically, in step S304, the service provider may perform timing after sending the current message, and obtain session information corresponding to the customer service session when the user' S message is still not received after the timing reaches the preset duration. The preset time can be set according to actual requirements, such as 30 seconds, 60 seconds and the like; the preset time period is usually shorter than the time period used in the time-out disconnection manner in the related art, so as to save time for the service provider. The obtained session information corresponding to the customer service session may only include one of the session content or the session statistical parameter in the customer service session, or may also include the session content and the session statistical parameter in the customer service session. The session content in the customer service session may be all messages sent by the user and the service provider in the customer service session, or may be a part of messages, where the part of messages may be messages sent by the user in the customer service session, messages sent by the service provider, or messages at specified positions in a sequence after being sorted according to the message sending time, and the like. The session statistics parameter may be statistics information related to the customer service session, such as a waiting time for sending each message, a time between the time when the service provider sends the current message and the current time, and the like, and may further include statistics information related to the user in the customer service session, such as the number of times the user uses online customer service, and the like.

Step S306, inputting the session information into a probability prediction model which is trained in advance to obtain the continuous probability of the customer service session;

the probabilistic predictive model can be implemented by various machine learning tools, such as neural networks, decision tree models, and the like. In the training process of the probability prediction model, training samples can be obtained from a database of customer service sessions, the training samples usually comprise positive samples and negative samples, and the positive samples can be extracted from sessions disconnected in overtime; negative examples may be extracted from sessions actively disconnected by the user; through learning and training of the positive sample and the negative sample, the probability prediction model can predict the continuing probability of the current customer service session based on the session information, and the continuing probability can also be understood as the probability that the user replies the message after the service provider sends the message.

And step S308, controlling the online state of the customer service session based on the continuation probability.

The online state of the customer service session may be online or disconnected; if the continuation probability is higher, it indicates that the user's question is not resolved based on the session message, and the possibility of the user sending the message is higher, at this time, the session can be maintained in an online state; if the continuation probability is low, it means that the user's question is resolved based on the session information, and the possibility that the user will not send any more messages is high, and at this time, the customer service session may be disconnected. In addition, the system can directly disconnect the customer service session, or send prompt information to the service provider, and the service provider determines whether to disconnect the session according to actual conditions. In actual implementation, a probability threshold or a probability interval may be set, and the predicted continuation probability is compared with the probability threshold or the probability interval, so as to control the online state of the customer service session.

In the method for controlling the online customer service session provided by the embodiment of the invention, if the situation that the message of the user is not received within the preset time after the service provider sends the current message in the current online customer service session is monitored, the session information corresponding to the customer service session is acquired; and then inputting the session information into a probability prediction model to obtain the continuous probability of the customer service session, and controlling the online state of the customer service session based on the continuous probability. In the method, the probability prediction model can predict the probability of the response of the user according to the session information of the customer service session, namely the continuous probability of the customer service session, and controls the online state of the session based on the continuous probability; compared with the overtime disconnection mode in the related technology, the mode can avoid disconnection of the session when the user problem is not solved, can also avoid the problem that the customer service wastes time to wait for the response of the user after the user problem is solved, and improves the working efficiency of the customer service while ensuring the user experience.

The embodiment provides that the continuous probability of the customer service session is predicted through a probability prediction model, and then the online state of the customer service session is controlled based on the continuous probability; in this embodiment, the training mode of the probabilistic predictive model is described in detail.

FIG. 4 is a schematic diagram of a training method for a probabilistic predictive model; the method comprises the following steps:

step S402, extracting session data from a preset session database;

the session database may be a hive-type database, but may also be implemented by other Hadoop architectures or databases (or data warehouses) of other architectures. The session database stores session data generated when a user in the service providing system makes a session with a service provider. In the process of extracting the session data, a screening field may be set, and the session data meeting the condition may be screened out from the session database according to the screening field. For example, the screening field may contain a user identifier (also referred to as a user ID), a session identifier (also referred to as a session ID), a session close reason (including normal close and timeout disconnection), and message types in the session (including voice, text, picture, etc.); the time of transmission of each message in the session, the source of transmission of the message (including the service provider, user, system), the content of the message, etc.

Step S404, determining a plurality of training samples based on the session data; each training sample comprises a session identifier, and session content, a session closing reason and session statistical parameters corresponding to the session identifier;

the extracted session data includes session data corresponding to a plurality of sessions, and each session corresponds to a unique session identifier. A training sample is usually determined from session data corresponding to a session, and specifically, the session identifier, the session content, the session close reason, the session statistical parameters, and the like may be extracted from the session data of a session. The session content is a message sent by a user and a service provider in the session; the session content may contain all messages in the session or only part of the messages. For example, after the messages in the conversation are sorted in time, only the last messages are used as the conversation contents, and the number of messages included in the conversation contents may be set in advance. The session statistical parameters may specifically be a time difference between adjacent messages after each message in the session is sorted according to time, a time difference between the last message and the session hang-up time, the number of messages sent by the user, the number of messages sent by the service provider, and the like.

In addition, the determined training samples can be divided into positive samples and negative samples according to the reason of session closing; the reason for closing the session of the positive sample is time-out disconnection, and the reason for closing the session of the negative sample is normal closing. In practical implementation, the number of the positive samples and the number of the negative samples may be predetermined, and the ratio of the positive samples to the negative samples is set within a reasonable range, for example, the ratio of the number of the positive samples to the number of the negative samples is 1: 1. 4: 1, etc.

Step S406, training a preset initial model based on a training sample to obtain a probability prediction model; wherein the initial model comprises: convolutional networks, recursive networks, and loss functions.

In the training process, training samples can be input into the initial model one by one, so that the initial model outputs a prediction result, the prediction result can comprise the probability of normal closing and the probability of overtime disconnection corresponding to the training samples, the prediction result is compared with the reason of session closing in the training samples, and then parameters in the initial model are adjusted to achieve the purpose of training the model.

Considering that most of training samples are text data, the initial model comprises a convolution network, a recursion network and a loss function; the convolution network can convert text data into vectors or matrixes which can participate in calculation; the recursion network can take the time sequence of each message in the sample data into consideration, so as to extract whether the user is predicted to reply to the related characteristic data subsequently or not; the loss function can calculate the probability of normal closing and the probability of overtime disconnection of the training sample based on the characteristic data.

The model training mode can obtain a probability prediction model, and the continuous probability of the customer service session can be predicted and obtained based on the probability prediction model. In order to further improve the accuracy and stability of the probability prediction model, another model training method is provided in this embodiment, and the method further describes a specific training mode of the model and a data cleaning process of training data; as shown in fig. 5, the method includes the steps of:

step S502, extracting session data from a preset session database;

step S504, according to the conversation label in the above-mentioned conversation data extracted, divide the conversation data into a plurality of; each session data corresponds to a session identifier;

because the extracted session data relates to a plurality of sessions, and each session corresponds to a unique session identifier, based on this, the session data can be divided to obtain the session data corresponding to each session.

Step S506, extracting a session identifier included in each piece of session data, a session content corresponding to the session identifier, a session close reason, and a session statistical parameter, to obtain a training sample corresponding to each piece of session data.

The session data corresponding to each divided session usually includes a large amount of information, especially information unrelated to probability prediction, such as a user account; in order to make the training data more concise and efficient, the session content, the session closing reason and the session statistical parameters related to the probability prediction need to be extracted from each session data, and the extracted information is combined into a training sample.

Step S508, for each training sample, judging whether the training sample contains the message of the user and the message of the session service provider; if not, the training sample is rejected. And obtaining a plurality of final training samples until all the training samples are traversed.

After the training sample corresponding to each session data is obtained, a plurality of training samples can be finally obtained; among these training samples, there still exist some training samples that do not significantly contribute to model training, and these training samples may also be referred to as invalid samples; for example, only a sample of messages sent by the user, only a sample of messages sent by the service provider; these samples need to be culled because there is no message interaction between the user and the service provider, which is not conducive to model training.

In actual implementation, the message sending source of each message in the session content can be identified one by one, and whether the training sample contains the message of the user and the message of the session service provider is determined; if the identified message sending sources of all the messages comprise the user and the session service provider, the training sample does not need to be rejected; the training samples are culled if the identified message delivery origin of each message includes only the user or the session service provider.

Step S510, determining a first current training sample from the plurality of training samples;

step S512, inputting the conversation content of the current training sample into a convolution network of a preset initial model, and outputting a feature matrix corresponding to the conversation content;

in practical implementation, a plurality of training samples can be sequenced according to a preset rule, then the training samples are determined as current training samples one by one according to the sequencing order, and the current training samples are input into a convolution network; of course, the current training sample may be randomly determined from a plurality of training samples. The convolution network can comprise a plurality of convolution layers, each convolution layer carries out convolution calculation on the conversation content through a corresponding convolution core, and then the conversation content in the text form is converted into a characteristic matrix; compared with the conversation content in the text form, the feature matrix is more convenient to participate in calculation, and the feature extraction processing of the subsequent network is facilitated.

In particular, the convolutional network can be implemented by a 3D convolutional neural network, which is beneficial for extracting the correlation between adjacent words or adjacent messages; when the convolutional network in the model is a 3D convolutional neural network, the session content of the current training sample can be mapped through the 3D convolutional neural network, and a feature matrix corresponding to the session content is obtained. In actual implementation, mapping each message in the session content through a 3D convolutional neural network to obtain a feature matrix corresponding to each message; the process of mapping the message may also be referred to as a process of text embedding the message.

Step S514, inputting the feature matrix and the session statistical parameters into a recursive network of the initial model, and outputting a feature value corresponding to the current training sample;

the recursive network can be an LSTM (Long Short-Term Memory) network, a GRU (gated recurrent Unit) network, a structural recurrent neural network and the like; the recursive network generally has a tree-shaped hierarchical structure and network nodes, and can carry out recursive processing on input information; the recursive network, when processing a text language, can parse sentences in the text language to extract language meaning of the text language. The feature value corresponding to the current training sample output by the recursive network usually represents the meaning of the session content in the current training sample.

For further understanding, the recursive network including a gated loop unit network with a preset number of layers is described as an example; the gate control circulation unit networks with the preset number of layers are sequentially connected; the number of layers of the gated cyclic unit network corresponds to the number of messages in the session content; for example, if the session content of the training sample includes five messages, the recursive network includes five gated cyclic unit networks, and the five gated cyclic unit networks are connected in sequence; after the five messages are sequenced according to time sequence, a first layer of gated cycle unit network processes the first message, a second layer of gated cycle unit network processes the second message until a fifth layer of gated cycle unit network processes the fifth message, and a final characteristic value is output; in these gated cycle unit networks, the input data of the current gated cycle unit network usually includes the output data of the previous gated cycle unit network, and also includes the message corresponding to the current gated cycle unit network, and other related data.

Based on the gated loop unit network, the step S514 can be further implemented by the following steps 02 to 06:

step 02, inputting a feature matrix of a message corresponding to a first layer of gated cyclic unit network in the feature matrix and session statistical parameters of session contents into the first layer of gated cyclic unit network to obtain an output result of the first layer of gated cyclic unit network;

specifically, the feature matrices of the messages in the feature matrix may be sorted according to the time sequence of the messages in the session content; the number of messages corresponds to the number of layers of the gated cyclic unit network; therefore, a feature matrix corresponding to a message can be obtained from the feature matrix according to the arrangement sequence; inputting the acquired feature rectangle and the session statistical parameters of the session content into a first layer of gate control cycle unit network for training; generally, the first layer of gated cyclic unit network performs convolution calculation on input data, and an output result of the gated cyclic unit network is a feature matrix or a feature vector obtained by performing convolution calculation on the input data. The session statistical parameters input into the first-layer gated cyclic unit network may be all session statistical parameters corresponding to the session content, or may be partial session statistical parameters associated with the message corresponding to the input feature matrix.

Step 04, for each layer of gated cyclic unit network except the first layer of gated cyclic unit network, inputting an output result of a previous layer of gated cyclic unit network of the layer of gated cyclic unit network, a feature matrix of a message corresponding to the layer of gated cyclic unit network and session statistical parameters of session content into the layer of gated cyclic unit network until the last layer of gated cyclic unit network;

for each layer of gated cyclic unit network except the first layer of gated cyclic unit network, the gated cyclic unit network of the previous layer is connected in front of the gated cyclic unit network; the input data of the gating circulation unit network comprises the output result of the previous layer of gating circulation unit network, the feature matrix of the message corresponding to the layer of gating circulation unit network and the session statistical parameters of the session content; the gated cyclic unit network performs convolution calculation on input data, and an output result of the gated cyclic unit network is usually a feature matrix or a feature vector obtained by performing convolution calculation on the input data. Because the input data of each layer of gated cyclic unit network comprises the output result of the previous layer of gated cyclic unit network, the input data of the last layer of gated cyclic unit network comprises the output results corresponding to all previous layers, and the output result of the last layer of gated cyclic unit network can represent the characteristics corresponding to all messages in the session content.

And step 06, taking the output result of the last layer of gated cyclic unit network as a characteristic value corresponding to the current training sample.

The output result of the last layer of gated cyclic unit network usually includes a multi-dimensional feature matrix, so that the feature value corresponding to the current training sample also includes a multi-dimensional feature matrix, such as 300 dimensions or even more.

Step S516, inputting the characteristic value into a loss function of the initial model, and outputting a prediction result of the session closing reason of the current training sample;

the loss function can be a cross entropy function, a softmax function, a sigmoid function and the like; the loss function can calculate the probability of the characteristic value of each dimension and output a specified characteristic value; specifically, in this embodiment, the feature value of each dimension usually represents different meanings, but usually includes a feature value representing a reason for closing a session, for example, feature value a represents normal closing, and feature value B represents a timeout hang-up; the loss function may specify a probability of outputting the feature value a and a probability of outputting the feature value B, where the probability of the feature value a and the probability of the feature value B are prediction results of the session close reason.

For further explanation, the implementation of step S516 is described below by taking an example in which the loss function includes a softmax function, and specifically includes the following steps 12 to 16:

step 12, calculating an exponential function value of each characteristic value through a softmax function;

specifically, the formula of the softmax function is as follows:

wherein x is_iRepresenting the ith characteristic value; x is the number of_jRepresents the jth characteristic value; n represents the total number of eigenvalues.

The exponential function value of the feature value may expand the difference between the feature values with respect to the feature value itself, for example, the vector of the feature value is [3,1, -3], and after the exponential function value of each feature value is calculated, the corresponding vector of the exponential function value of the feature value is [20,2.7,0.05 ]. The probability of each characteristic value is calculated by adopting the exponential function values of the characteristic values, so that the probability difference between the characteristic values can be increased, the probability of a correct recognition result is higher, and the accuracy of the recognition result is facilitated.

Step 14, determining the probability of each feature value according to the index function value of each feature value and the sum of the index function values of each feature value of the current training sample;

specifically, the probability of the feature value is obtained by dividing the index function value of each feature value by the sum of the index function values of each feature value.

And step 16, outputting the probability of the specified characteristic value in each characteristic value, and determining the probability of the specified characteristic value and the session closing reason corresponding to the specified characteristic value as the prediction result of the session closing reason of the current training sample.

Since the feature value output by the recursive network usually includes multiple dimensions, and only the probability of the feature value related to the reason of session closure is concerned in this embodiment, the feature value related to the reason of session closure may be determined in advance as the specified feature value, and the loss function may output only the probability of the specified feature value.

For example, the specified characteristic values are a characteristic value a and a characteristic value B, wherein the characteristic value a represents normal closing, and the characteristic value B represents time-out hanging-up; the probability corresponding to the characteristic value A is 0.7, the probability corresponding to the characteristic value B is 0.2, which shows that the probability of the normal closing of the session content is 0.7, the probability of the overtime hangup is 0.2, and the probability of the normal closing is higher; if the session closing reason of the session content is normally closed, the prediction result of the session content is consistent with the real session closing reason.

Step S518, updating the prediction accuracy of the initial model according to the session closing reason of the current training sample and the prediction result of the session closing reason;

in the model training process, whether the model predicts each training sample correctly or incorrectly can be recorded, and the prediction accuracy of the initial model is further calculated; the prediction accuracy may be updated based on the current training sample each time a new current training sample is predicted.

Step S520, judging whether to terminate training based on preset conditions; if not, go to step S522; if yes, go to step S524; wherein, the preset conditions include: the prediction accuracy is greater than or equal to a preset accuracy threshold, and/or the number of current training samples input into the convolutional network is equal to a preset iteration threshold;

judging whether the preset conditions are met or not after the prediction of one current training sample is completed; the preset condition may only include that the prediction accuracy is greater than or equal to the preset accuracy threshold, or the number of the current training samples input to the convolutional network is equal to one of the preset iteration threshold, or may also include that the prediction accuracy is greater than or equal to the preset accuracy threshold, and the number of the current training samples input to the convolutional network is equal to the preset iteration threshold, that is, when the prediction accuracy of the model is greater than or equal to the preset accuracy threshold, and the number of the current training samples input to the convolutional network is equal to the preset iteration threshold, the training may be terminated.

The accuracy threshold and the iteration number threshold can be set according to actual requirements, for example, the accuracy threshold can be set to 80%, 90% and the like; the iteration number threshold may be positioned 1000 times, 5000 times, etc.

Step S522, determining the next current training sample from the plurality of training samples, and continuing to execute step S512, i.e. continuing to train the convolutional network, the recursive network, and the loss function. And (6) ending.

Step S524, the initial model after the training of the current training sample is used as a probability prediction model.

The probability prediction model obtained by training in the mode has reliable and stable performance and higher accuracy, can predict the continuous probability of the customer service session according to the session information of the customer service session, and controls the online state of the session based on the continuous probability, thereby avoiding the disconnection of the session when the user problem is not solved, avoiding the waste of time for the customer service to wait for the response of the user after the user problem is solved, and improving the working efficiency of the customer service while ensuring the user experience.

Further, another control method for online customer service sessions is provided in the embodiments of the present invention, where the method focuses on describing a specific implementation manner for controlling an online state of a customer service session based on a continuation probability; as shown in fig. 6, the method includes the steps of:

step S602, monitoring the current online customer service session; wherein, the customer service session is a session between a user and a service provider;

step S604, when the service provider sends the current message, a timer is started; the timing duration of the timer is a preset duration;

step S606, monitoring whether the message of the user is received in the timing duration of the timer; if yes, ending; if not, go to step S608;

step S608, session information corresponding to the customer service session is acquired.

The session information comprises session content and session statistical parameters; wherein the session content includes: taking the current message of the service provider as a reference, and continuously obtaining at least one part of messages before the current message and the current message; for example, it may be preset that the session content includes a specified number of messages, and the specified number of messages is obtained continuously from the current message of the service provider; such as five, or other numbers. If the number of messages in the conversation is more, the specified number of messages are only a part of messages of the conversation; if the number of the messages in the conversation is totally specified, the specified number of the messages are all messages of the conversation; and if the number of the messages in the session is less than the specified number, acquiring the number of the messages in all the sessions.

The session statistic parameters include: the interval duration between two adjacent messages in the customer service session, the duration of the current message from the current time, the number of messages of the user and the number of messages of the service provider. Among these session statistical parameters, the interval duration between two adjacent messages in the customer service session and the duration of the current message from the current time usually represent the intention of the user more obviously, and have a greater influence on the probability prediction, so that in practical implementation, the session statistical parameters may only include the interval duration between two adjacent messages in the customer service session and the duration of the current message from the current time. Because the system usually automatically records the sending time of the message when the user or the service provider sends the message, the messages in the conversation can be sequenced according to the sending time, the sending time of the adjacent conversation is subjected to difference operation, the interval duration between the two adjacent messages in the customer service conversation can be obtained, and the current time and the sending time of the current message are subjected to difference operation, so the duration between the current message and the current time can be obtained.

Step S610, inputting the session information into a probability prediction model which is trained in advance to obtain the continuous probability of the customer service session;

step S612, judging whether the continuing probability is smaller than a preset probability threshold value; if not, executing step S614; if yes, go to step S616;

the probability threshold may be set according to actual requirements, for example, the probability threshold may be 50%, 70%, etc. If the continuing probability is smaller than the preset probability threshold, the possibility that the user replies the message is not high, and the system can directly disconnect the customer service session at the moment; in addition, in order to more reliably control the customer service session, the system may not directly disconnect the customer service session, and send a prompt message to the service provider of the customer service session to inform the service provider that the possibility of a user replying a message is low, and after receiving the prompt message, the service provider may further determine whether the customer service session can be disconnected according to the session content.

Step S614, monitoring whether the message of the user is received within a preset timeout time threshold value; if yes, ending; if not, go to step S616;

if the continuation threshold is higher, the customer service session is temporarily not disconnected or the service provider is instructed to disconnect the customer service session, a new timer can be started at the moment, the timing duration of the timer is the overtime threshold, and if the user information is still not received in the timing process of the timer, the customer service session is disconnected or the service provider is instructed to disconnect the customer service session.

Step S616, disconnect the customer service session or instruct the service provider to disconnect the customer service session.

In the above manner, if the continuation probability output by the probability prediction model is smaller than the preset probability threshold, the customer service session is disconnected or the service provider is instructed to disconnect the customer service session; if the continuing probability is larger than or equal to a preset probability threshold, disconnecting the customer service session in a mode of overtime disconnection or indicating a service provider to disconnect the customer service session; the method can avoid disconnection of the session when the user problem is not solved, can also avoid the problem that the customer service wastes time to wait for the response of the user after the user problem is solved, and improves the working efficiency of the customer service while ensuring the user experience.

In another way, if the continuation probability is equal to or higher than a preset probability threshold, whether the message of the user is received within a preset timeout time threshold can be monitored; if not, continuing to predict the new continuing probability of the customer service session through the probability prediction model; and when the new continuation probability is lower than a preset probability threshold value, disconnecting the customer service session or instructing the service provider to disconnect the customer service session.

Specifically, if the message of the user is still not received within the timeout time threshold, the session information corresponding to the customer service session may be continuously acquired; the session content in the session information is generally the same as the session content input to the probabilistic predictive model last time; the session statistical parameters in the session information are usually different from the session statistical parameters input to the probability prediction model last time; for example, the duration of the current message sent by the service provider from the current time may be longer due to the change of the current time, and thus the continuation probability output by the probabilistic predictive model may also be changed. It will be appreciated that the longer the duration of the current message from the current time, the lower the probability of continuation. Therefore, even if the probability prediction model has a low continuation probability of the first prediction output, if the user does not send a message all the time, the new continuation probability is lower and lower in the subsequent prediction process until the new continuation probability is lower than the preset probability threshold, and the customer service session is disconnected or the service provider is instructed to disconnect the customer service session.

It should be noted that, if the multiple prediction mode is adopted, it is necessary that the training samples of the probabilistic prediction model include a time length from a last message sent by the service provider to a session disconnection in a training process, and in a first prediction process, the session statistical parameter input to the probabilistic prediction model includes a time length from a current time to a current message sent by the service provider.

In the above manner, if the continuous probability output by the probability prediction model is greater than or equal to the preset probability threshold, the session information corresponding to the customer service session is obtained again, and is input into the probability prediction model for prediction, and a new probability threshold is output; the method can also avoid disconnection of the session when the user problem is not solved, and can also avoid the problem that the customer service wastes time to wait for the response of the user after the user problem is solved, thereby improving the working efficiency of the customer service while ensuring the user experience.

In correspondence to the above embodiment of the control method of the online customer service session, refer to a schematic structural diagram of a control device of the online customer service session shown in fig. 7; the functions performed by the device correspond to the steps performed by the method. The apparatus may be understood as the server or the processor of the server, or may be understood as a component which is independent of the server or the processor and implements the functions of the present application under the control of the server, as shown in fig. 7, the apparatus includes:

a session monitoring module 70, configured to monitor a current online customer service session; the customer service session is a session between a user and a service provider;

an information obtaining module 72, configured to obtain session information corresponding to the customer service session if the service provider does not receive the user's message within a preset duration after sending the current message; the session information comprises session content and/or session statistical parameters in the customer service session;

a probability prediction module 74, configured to input session information into a probability prediction model that is trained in advance, so as to obtain a continuous probability of the customer service session;

a session control module 76 for controlling the online state of the customer service session based on the continuation probability.

In the control device for the online customer service session provided by the embodiment of the invention, if the situation that the message of the user is not received within the preset time after the service provider sends the current message in the current online customer service session is monitored, the session information corresponding to the customer service session is acquired; and then inputting the session information into a probability prediction model to obtain the continuous probability of the customer service session, and controlling the online state of the customer service session based on the continuous probability. In the method, the probability prediction model can predict the probability of the response of the user according to the session information of the customer service session, namely the continuous probability of the customer service session, and controls the online state of the session based on the continuous probability; compared with the overtime disconnection mode in the related technology, the mode can avoid disconnection of the session when the user problem is not solved, can also avoid the problem that the customer service wastes time to wait for the response of the user after the user problem is solved, and improves the working efficiency of the customer service while ensuring the user experience.

The modules in the control device of the above-mentioned online customer service session may be connected or communicate with each other via a wired connection or a wireless connection. The wired connection may include a metal cable, an optical cable, a hybrid cable, etc., or any combination thereof. The wireless connection may comprise a connection over a LAN, WAN, bluetooth, ZigBee, NFC, or the like, or any combination thereof. Two or more modules may be combined into a single module, and any one module may be divided into two or more units.

In some embodiments, the loss function comprises a softmax function; the number of the characteristic values corresponding to the current training sample is multiple; the model training module is configured to: calculating an exponential function value of each characteristic value through a softmax function; determining the probability of each characteristic value according to the index function value of each characteristic value and the sum of the index function values of each characteristic value of the current training sample; and outputting the probability of a specified characteristic value in each characteristic value, and determining the probability of the specified characteristic value and the session closing reason corresponding to the specified characteristic value as the prediction result of the session closing reason of the current training sample.

The present embodiment also provides an electronic device, including: the system comprises a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, when the electronic device runs, the processor and the storage medium are communicated through the bus, and the processor executes the machine-readable instructions to execute the steps of the control method of the online customer service session.

The present embodiment also provides a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, performs the steps of the method for controlling an online customer service session as described above.

It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the system and the apparatus described above may refer to corresponding processes in the method embodiments, and are not described in detail in this application. In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is merely a logical division, and there may be other divisions in actual implementation, and for example, a plurality of modules or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or modules through some communication interfaces, and may be in an electrical, mechanical or other form.

The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for controlling an online customer service session, the method comprising:

monitoring the current online customer service session; the customer service session is a session between a user and a service provider;

if the service provider does not receive the message of the user within a preset time length after sending the current message, session information corresponding to the customer service session is obtained; wherein the session information comprises session content and/or session statistical parameters in the customer service session;

inputting the session information into a probability prediction model trained in advance to obtain the continuous probability of the customer service session;

controlling an online state of the customer service session based on the continuation probability.

2. The method according to claim 1, wherein the step of obtaining the session information corresponding to the customer service session if the service provider does not receive the user's message within a preset time period after sending the current message comprises:

starting a timer when the service provider sends a current message; the timing duration of the timer is preset duration;

monitoring whether the message of the user is received within the timing duration of the timer;

and if not, acquiring the session information corresponding to the customer service session.

3. The method according to claim 1 or 2, wherein the session content comprises: based on the current message of the service provider, the current message and at least a part of messages which are continuous before the current message;

the session statistical parameters include: the interval duration between two adjacent messages in the customer service session, the duration of the current message from the current time, the number of messages of the user, and the number of messages of the service provider.

4. The method of claim 1, wherein the probabilistic predictive model is trained by:

extracting session data from a preset session database;

determining a plurality of training samples based on the session data; each training sample comprises a session identifier, and session content, a session closing reason and session statistical parameters corresponding to the session identifier;

training a preset initial model based on the training sample to obtain a probability prediction model; wherein the initial model comprises: convolutional networks, recursive networks, and loss functions.

5. The method of claim 4, wherein the step of determining a plurality of training samples based on the session data comprises:

dividing the session data into a plurality of pieces according to the session identification in the extracted session data; each session data corresponds to a session identifier;

and extracting a session identifier contained in each piece of session data, wherein the session content, the session closing reason and the session statistical parameters corresponding to the session identifier obtain a training sample corresponding to each piece of session data.

6. The method of claim 5, wherein after obtaining the training sample corresponding to each session data, the method further comprises:

for each training sample, judging whether the training sample contains the message of the user and the message of the session service provider;

if not, the training sample is rejected.

7. The method of claim 4, wherein the step of training a preset initial model based on the training samples comprises:

determining a current training sample from a plurality of training samples, inputting the conversation content of the current training sample into the convolution network of the preset initial model, and outputting a feature matrix corresponding to the conversation content;

inputting the feature matrix and the session statistical parameters into a recursive network of the initial model, and outputting a feature value corresponding to the current training sample;

inputting the characteristic value into a loss function of the initial model, and outputting a prediction result of a session closing reason of the current training sample;

updating the prediction accuracy of the initial model according to the session closing reason of the current training sample and the prediction result of the session closing reason;

judging whether to terminate training based on preset conditions, wherein the preset conditions comprise: the prediction accuracy is greater than or equal to a preset accuracy threshold, and/or the number of current training samples input to the convolutional network is equal to a preset iteration threshold;

if not, determining the next current training sample from the plurality of training samples, and continuing to train the convolutional network, the recursive network and the loss function;

if so, taking the initial model after the current training sample is trained as a probability prediction model.

8. The method of claim 7, wherein the convolutional network comprises a 3D convolutional neural network;

inputting the session content of the current training sample into the convolutional network of the preset initial model, and outputting a feature matrix corresponding to the session content, wherein the step of outputting the feature matrix comprises the following steps:

and mapping the session content of the current training sample through the 3D convolutional neural network to obtain a feature matrix corresponding to the session content.

9. The method of claim 7, wherein the recursive network comprises a predetermined number of gated cyclic unit networks; the gate control circulation unit networks with the preset number of layers are sequentially connected; the number of layers of the gating cycle unit network corresponds to the number of messages in the session content;

inputting the feature matrix and the session statistical parameters into a recursive network of the initial model, and outputting a feature value corresponding to the current training sample, wherein the step comprises the following steps:

inputting a feature matrix of a message corresponding to a first layer of gated cyclic unit network in the feature matrix and session statistical parameters of the session content into the first layer of gated cyclic unit network to obtain an output result of the first layer of gated cyclic unit network;

for each layer of gated cyclic unit network except the first layer of gated cyclic unit network, inputting the output result of the previous layer of gated cyclic unit network of the layer of gated cyclic unit network, the feature matrix of the message corresponding to the layer of gated cyclic unit network and the session statistical parameters of the session content into the layer of gated cyclic unit network until the last layer of gated cyclic unit network;

and taking the output result of the last layer of gating cycle unit network as a characteristic value corresponding to the current training sample.

10. The method of claim 7, wherein the loss function comprises a softmax function; the number of the characteristic values corresponding to the current training sample is multiple;

inputting the characteristic value into a loss function of the initial model, and outputting a prediction result of a session closure reason of the current training sample, wherein the step comprises the following steps:

calculating an exponential function value of each of the feature values by the softmax function;

determining the probability of each feature value according to the index function value of each feature value and the sum of the index function values of each feature value of the current training sample;

and outputting the probability of a specified characteristic value in each characteristic value, and determining the probability of the specified characteristic value and the session closing reason corresponding to the specified characteristic value as the prediction result of the session closing reason of the current training sample.

11. The method of claim 1, wherein the step of controlling the online state of the customer service session based on the continuation probability comprises:

and if the continuation probability is smaller than a preset probability threshold, disconnecting the customer service session or indicating the service provider to disconnect the customer service session.

12. The method of claim 1, wherein the step of controlling the online state of the customer service session based on the continuation probability comprises:

if the continuing probability is equal to or higher than a preset probability threshold, monitoring whether the message of the user is received within a preset timeout time threshold;

if not, disconnecting the customer service session or instructing the service provider to disconnect the customer service session.

13. The method of claim 1, wherein the step of controlling the online state of the customer service session based on the continuation probability comprises:

if not, continuing to predict the new continuing probability of the customer service session through the probability prediction model;

and when the new continuation probability is lower than a preset probability threshold, disconnecting the customer service session or instructing the service provider to disconnect the customer service session.

14. An apparatus for controlling an online customer service session, the apparatus comprising:

the session monitoring module is used for monitoring the current online customer service session; the customer service session is a session between a user and a service provider;

the information acquisition module is used for acquiring the session information corresponding to the customer service session if the service provider does not receive the message of the user within a preset time length after sending the current message; wherein the session information comprises session content and/or session statistical parameters in the customer service session;

the probability prediction module is used for inputting the session information into a probability prediction model which is trained in advance to obtain the continuous probability of the customer service session;

and the session control module is used for controlling the online state of the customer service session based on the continuation probability.

15. The apparatus of claim 14, wherein the information obtaining module is configured to:

16. The apparatus according to claim 14 or 15, wherein the session content comprises: based on the current message of the service provider, the current message and at least a part of messages which are continuous before the current message;

17. The apparatus of claim 14, further comprising a model training module to:

extracting session data from a preset session database;

18. The apparatus of claim 17, wherein the model training module is configured to:

19. The apparatus of claim 18, further comprising:

the message judging module is used for judging whether each training sample contains the message of the user and the message of the session service provider;

and the sample removing module is used for removing the training sample if the training sample does not contain the message of the user and the message of the session service provider.

20. The apparatus of claim 17, wherein the model training module is configured to:

21. The apparatus of claim 20, wherein the convolutional network comprises a 3D convolutional neural network;

the model training module is configured to: and mapping the session content of the current training sample through the 3D convolutional neural network to obtain a feature matrix corresponding to the session content.

22. The apparatus of claim 20, wherein the recursive network comprises a predetermined number of gated cyclic unit networks; the gate control circulation unit networks with the preset number of layers are sequentially connected; the number of layers of the gating cycle unit network corresponds to the number of messages in the session content;

the model training module is configured to: inputting a feature matrix of a message corresponding to a first layer of gated cyclic unit network in the feature matrix and session statistical parameters of the session content into the first layer of gated cyclic unit network to obtain an output result of the first layer of gated cyclic unit network;

23. The apparatus of claim 20, wherein the loss function comprises a softmax function; the number of the characteristic values corresponding to the current training sample is multiple;

the model training module is configured to: calculating an exponential function value of each of the feature values by the softmax function;

24. The apparatus of claim 14, wherein the session control module is configured to:

25. The apparatus of claim 14, wherein the session control module is configured to:

26. The apparatus of claim 14, wherein the session control module is configured to:

27. An electronic device, comprising: processor, storage medium and bus, the storage medium storing machine readable instructions executable by the processor, the processor and the storage medium communicating via the bus when the electronic device is running, the processor executing the machine readable instructions to perform the steps of the method of controlling an online customer service session according to any one of claims 1 to 13.

28. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps of the method of controlling an online customer service session according to any one of claims 1 to 13.