CN115858757A

CN115858757A - Customer service mail response method and device, equipment, medium and product thereof

Info

Publication number: CN115858757A
Application number: CN202211678113.8A
Authority: CN
Inventors: 李一龙; 吴文龙
Original assignee: Shenzhen Zhiyan Technology Co Ltd; Shenzhen Qianyan Technology Co Ltd
Current assignee: Shenzhen Zhiyan Technology Co Ltd; Shenzhen Qianyan Technology Co Ltd
Priority date: 2022-12-26
Filing date: 2022-12-26
Publication date: 2023-03-28

Abstract

The application relates to a customer service mail response method and a device, equipment, medium and product thereof, wherein the method comprises the following steps: acquiring mail content of a newly added mail in a target mailbox; determining a target intention and session information corresponding to the mail content, wherein the session information comprises at least one item of element information for defining a customer service question corresponding to the target intention, the customer service question is defined by complete element information, and the complete element information comprises a plurality of element information pointed by a plurality of preset element labels; determining a corresponding dialog template according to the target intention, wherein the dialog template comprises a format text and an element label corresponding to the element information lacking in the session information, and the lacking element information is determined by the session information relative to the complete element information of the customer service problem; and replying the new mail with the conversation template. The method and the device can guide the user to completely reflect all element information of the customer service problem through the mail, improve the communication efficiency of the user in reflecting the customer service problem, and improve the customer service satisfaction.

Description

Customer service mail response method and device, equipment, medium and product thereof

Technical Field

The application relates to the field of intelligent customer service, in particular to a customer service mail response method and a device, equipment, medium and product thereof.

Background

A merchant of a cross-border e-commerce usually operates a plurality of online shops through a plurality of e-commerce platforms, each e-commerce platform can usually provide an intelligent customer service robot to support customer service requirements of the corresponding online shop, but it is not practical for merchants correspondingly owning the plurality of online shops on the plurality of e-commerce platforms to realize centralized management and control of customer service data of the plurality of e-commerce platforms.

An alternative method is that customer service inlets are arranged in all online stores for users to feed back problems, in order to realize centralized management and control, an electronic mailbox is usually adopted to collect customer problems, a mailbox address of the electronic mailbox is configured in the corresponding online store so as to receive corresponding electronic mails, the customer service problems submitted by the online stores of the users are sent to corresponding mailbox servers in the form of the electronic mails, and customer service personnel can solve the customer service problems by checking and receiving the mails.

It is easy to see that, in such a scenario, although the customer service staff of the merchant can process the emails generated by the online shops in a centralized manner, the customer service staff can only process massive emails in a manual manner, and the processing manner is inefficient, cannot realize various analysis functions, and cannot embody the advantage of centralized management and control.

Theoretically, efficiency can be improved by automatically analyzing and replying each email through the customer service robot, however, the customer service robot cannot effectively provide effective reply texts due to the fact that the customer service robot often has the situations of incomplete information, unclear information and the like when a user reflects a problem, and therefore service quality is affected and expectations are not met.

In view of the above, a breakthrough needs to be found between the dilemma of the reply efficiency and reply effectiveness of the customer service problem based on the e-mail, so as to comprehensively improve the customer service response efficiency in the business scene of the cross-border e-commerce.

Disclosure of Invention

The present application aims to solve the above problems and provide a customer service mail response method, and a corresponding device, apparatus, non-volatile readable storage medium, and computer program product.

According to one aspect of the application, a customer service mail response method is provided, which comprises the following steps:

acquiring mail content of a newly added mail in a target mailbox;

determining a target intention and session information corresponding to the mail content, wherein the session information comprises at least one item of element information for defining a customer service question corresponding to the target intention, the customer service question is defined by corresponding complete element information, and the complete element information comprises a plurality of element information pointed by a plurality of preset element labels;

determining a corresponding dialog template according to the target intention, wherein the dialog template comprises a preset format text and an element label corresponding to the element information lacking in the session information, and the lacking element information is determined by the session information relative to the complete element information of the customer service problem;

and replying the new mail with the conversation template.

In an optional embodiment, determining the target intention and the session information corresponding to the mail content includes:

inputting the mail content into a preset named entity marking model, determining one or more element information in the named entity marking model, and forming session information;

and inputting the mail content and/or the conversation information into a preset intention analysis model, and predicting a corresponding target intention.

In an optional embodiment, determining the corresponding dialog template according to the target intention includes:

determining a corresponding dialogue relation tree according to the target intention, wherein the dialogue relation tree comprises a plurality of dialogue units pre-associated according to a tree structure, and each dialogue unit correspondingly provides the dialogue template;

matching element information contained in the session information with complete element information of the customer service problem, and determining a missing element label in the session information;

and determining a dialog template completely matched with the element label according to the missing element label in the session information.

In an optional embodiment, determining a dialog template completely matching the session information according to the missing element tag in the session information includes:

determining a platform identifier according to the receiver mailbox of the newly added mail;

determining a dialog template set completely matched with the missing element labels in the session information according to the missing element labels in the session information;

and determining a dialog template corresponding to the platform identification in the dialog template set.

In an optional embodiment, before obtaining the mail content of the newly added mail in the target mailbox, the method includes:

pulling at least one newly added mail from a target mailbox according to preset configuration;

constructing a corresponding mail message for each newly added mail;

and marking each mail message as a to-be-executed state, and adding the mail message into the task list for ordered listed consumption.

In an alternative embodiment, marking each email message as a to-be-executed state, and after adding the email message to the task list for in-order listed consumption, the method includes:

orderly reading the mail messages in the task list in a to-be-executed state for consumption;

marking the consumed email message which is replied by adopting the dialogue template as a finished state;

and transmitting the email message which is consumed but not replied by adopting the conversation template to a manual customer service interface for processing, and marking the email message as a manual processing state.

In an optional embodiment, the obtaining of the mail content of the newly added mail in the target mailbox includes:

responding to the consumption event of the mail message in the state to be executed in the task list, and extracting a sender mailbox address, a mail subject, a mail body and a mail signature file in the newly-added mail;

packaging any multiple items in the mail subject, the mail text and the mail signature file into mail content;

and determining the sender mailbox address as a target mailbox address for replying the newly added mail.

In an optional embodiment, after replying to the new email with the conversation template, the method includes:

monitoring a reply mail replied by the sender of the newly added mail based on the conversation template;

extracting the element information submitted by each element label corresponding to the dialogue template, perfecting the conversation information and obtaining complete conversation information;

predicting a corresponding target intention according to the complete session information;

and determining a reply text mapped with the target intention according to the target intention, and responding the reply mail with the reply text.

According to another aspect of the present application, there is provided a customer service mail answering apparatus, including:

the mail acquisition module is set to acquire the mail content of the newly added mail in the target mailbox;

the information analysis module is used for determining a target intention and session information corresponding to the mail content, wherein the session information comprises at least one item of element information used for defining a customer service question corresponding to the target intention, the customer service question is defined by corresponding complete element information, and the complete element information comprises a plurality of element information pointed by a plurality of preset element labels;

the template determining module is used for determining a corresponding dialog template according to the target intention, the dialog template comprises a preset format text and an element label corresponding to the element information lacking in the session information, and the lacking element information is determined by the session information relative to the complete element information of the customer service problem;

and the mail reply module is used for replying the newly added mail by the conversation template.

According to another aspect of the present application, there is provided a customer service mail response device, comprising a central processing unit and a memory, wherein the central processing unit is configured to invoke and run a computer program stored in the memory to perform the steps of the customer service mail response method described in the present application.

According to another aspect of the present application, a non-volatile readable storage medium is provided, which stores a computer program implemented according to the customer service mail response method in the form of computer readable instructions, and when the computer program is called by a computer, the steps included in the method are executed.

According to another aspect of the present application, there is provided a computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method described in any one of the embodiments of the present application.

Compared with the prior art, the method and the device have the advantages that the target intention and the session information of the customer service problem corresponding to the mail content of the newly added mail submitted by the user are determined firstly, then the conversation template is determined according to the target intention, the conversation template comprises the element labels which are deficient in the conversation information corresponding to the customer service problem, the deficient element labels are determined according to the complete element information of the customer service problem, the newly added mail is responded by the conversation template, accordingly, the function of guiding a sender of the newly added mail to supplement the complete information can be achieved through the element labels, various newly added mails can be automatically responded by adopting the mode, the session information of the same sender can be continuously supplemented, and finally, each customer service problem can obtain the complete element information, so that the solution efficiency of the customer service problem is improved, and the conversation efficiency in the customer service process is further improved by means of machine intelligence.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is an architectural diagram of an exemplary network environment of the present application;

FIG. 2 is a schematic diagram of a network architecture of an intent analysis model as exemplary employed herein;

FIG. 3 is a schematic diagram of a network architecture of a named entity tagging model used in an exemplary implementation of the present application;

FIG. 4 is a schematic diagram of a network architecture of an exemplary problem analysis model employed herein;

FIG. 5 is a flowchart illustrating an embodiment of a customer service mail response method according to the present application;

FIG. 6 is a schematic flowchart of determining a dialog template according to a target intention in an embodiment of the present application;

FIG. 7 is a schematic flowchart illustrating a process of determining a dialog template based on platform id according to an embodiment of the present application;

FIG. 8 is a schematic flow chart illustrating ordered processing of newly added mails based on a task list in an embodiment of the present application;

FIG. 9 is a flowchart illustrating status control of email messages in a task list according to an embodiment of the present application;

fig. 10 is a schematic flowchart illustrating parsing of an added email in the embodiment of the present application;

FIG. 11 is a flowchart illustrating the process of providing reply text to a customer service question in accordance with a reply email based on a reply to a conversation template in an embodiment of the present application;

FIG. 12 is a functional block diagram of a customer service mail answering device according to the present application;

fig. 13 is a schematic structural diagram of a customer service mail answering apparatus used in the present application.

Detailed Description

The models cited or possibly cited in the application comprise a traditional machine learning model or a deep learning model, and unless specified in clear text, the models can be deployed in a remote server and remotely called at a client, and can also be deployed in a client with qualified equipment capability to be directly called.

Referring to fig. 1, a network architecture adopted in an exemplary application scenario in the present application includes a terminal device 80, a mailbox server 81, an application server 82, and a provider server 83, where the application server 82 may be configured to deploy a computer program product implemented by programming according to a customer service mail response method in the present application, so as to provide a corresponding customer service system service; the mail server 81 may be configured to store an email sent by a consumer user, and allow the application server 82 to pull or reply to the corresponding email; the terminal device 80 can be used for a consumer user and/or a manual agent user to log in the customer service system so as to serve various correspondingly provided pages to realize various functions; the e-commerce server 83 is used for providing the customer user with access service of an independent station corresponding to the online shop.

In one embodiment, the mailbox server 81 may be multiple, and is respectively configured to provide a support service for a corresponding electronic mailbox, and the application server 82 may implement access to the corresponding electronic mailbox by pre-configuring login verification information of a corresponding mailbox address.

In one embodiment, the e-commerce server 83 may have a plurality of independent stations, each of which is used for providing independent station services corresponding to different online stores, and for each page of an independent station, one or more of the electronic mailboxes may be used as recipients for reflecting customer service problems, and when a corresponding customer user triggers a mail sending event for submitting a customer service problem, the corresponding independent station sends a corresponding mail to the corresponding recipient, and enters the corresponding mailbox server 81, so that the mail can be read by the application server 82.

Referring to fig. 2, an intention analysis model exemplarily employed by the present application is built based on a deep learning model, and is configured to extract deep semantic information according to given text information, perform classification mapping according to the deep semantic information, map the deep semantic information to each intention type in a preset classification space, obtain a classification probability corresponding to each intention type, and determine an intention type with the largest classification probability as a target intention corresponding to the text information.

In one embodiment, the intent analysis model employs a basic neural network model adapted to process serialized information as a backbone network, followed by a classifier based on the backbone network. The backbone network may include, but is not limited to, any of the following: a Recurrent Neural Network (RNN), a Long Short-Term Memory network (LSTM), a transform encoder, bert, and the like; the classifier may be a multi-classifier constructed using a Softmax function.

The intention analysis model adopts training samples in a training data set and corresponding supervision labels in advance to carry out iterative training to a convergence state and then is put into an on-line reasoning for use. The training sample is text information, and usually contains partial or all text content for describing the customer service problem, and the supervision label may be a target intention corresponding to the customer service problem. When the intention analysis model needs to be subjected to iterative training for one time, one training sample is adopted for coding and then input into the intention analysis model, corresponding deep semantic information is extracted, then corresponding result intention is predicted, then the supervision label is adopted for calculating a loss value corresponding to the result intention, when the loss value reaches a target threshold value for verifying whether the model is converged, the intention analysis model is represented to be converged, and therefore the iterative training process is terminated, otherwise, the intention analysis model is represented not to reach a convergence state, therefore, gradient updating is carried out on the intention analysis model according to the loss value, weight parameters of all links of the intention analysis model are corrected through back propagation to enable the model to be further converged, then, the next training sample is continuously called from the training data set to carry out the next iteration, and so on until the intention analysis model reaches the convergence state.

Referring to fig. 3, the named entity tagging model exemplarily adopted by the application is built based on a deep learning model, and is configured to extract deep semantic information of a given text message according to the deep learning model, and then extract each keyword and a corresponding type tag thereof according to the deep semantic information, so as to obtain mapping relationship data between the type tag and the keyword. In this application, the type tag may be an element tag corresponding to various types of element information required for describing information for defining a customer service problem, and the keyword is a type tag corresponding thereto, that is, an element tag.

In one embodiment, the named entity tagging model is implemented by taking a basic neural network model suitable for processing serialized information as a backbone network and then connecting a conditional random field network on the basis of the backbone network. The backbone network may include, but is not limited to, any of the following: a Recurrent Neural Network (RNN), a Long Short-term memory Network (LSTM), a transform encoder, a Bert, and the like; the Conditional Random Field network may be a CRF (Conditional Random Field) model.

Similarly, the named entity labeling model adopts training samples in a training data set and corresponding supervision labels in advance to carry out iterative training to a convergence state, and then is put into an online reasoning mode for use. The training sample is text information, and usually contains partial or all text contents corresponding to the customer service problem, and the supervision tag may be element information and an element tag thereof corresponding to the customer service problem. When the named entity tagging model needs to be subjected to one-time iterative training, one training sample is adopted for coding and then is input into the named entity tagging model, corresponding deep semantic information is extracted, corresponding element information and corresponding element labels thereof are predicted to serve as result data, then the supervision labels are adopted for calculating loss values corresponding to the result data, when the loss values reach a target threshold value used for verifying whether the model converges, the named entity tagging model is represented to be converged, and therefore the iterative training process is terminated, otherwise, the named entity tagging model does not reach a convergence state, so that gradient updating is carried out on the named entity tagging model according to the loss values, weight parameters of all links of the named entity tagging model are corrected through back propagation, the model is promoted to be further converged, then, a next training sample is continuously called from the training data set for carrying out the next iteration, and the analogy is carried out until the named entity tagging model reaches the convergence state.

Referring to fig. 4, another exemplary network architecture embodiment is disclosed, wherein the intention analysis model and the named entity tagging model can be constructed as the same problem analysis model by sharing the same backbone network, which is still used to extract deep semantic information of the text information inputted therein, and then determining the target intention through the respective corresponding branch networks, i.e. through the classifier in the first branch network, and determining the named entity through the conditional random field network in the second branch network. It is understood that, for such a network architecture, the intention analysis model and the named entity tagging model are jointly trained, and in each iterative training, the respective loss values of the intention analysis model and the named entity tagging model may be superimposed to be an overall loss value, and then whether to continue the iteration and update the weight is determined according to the overall loss value.

In one embodiment, the intent analysis model may be deployed in any runtime space that can be invoked by the application server to service the relevant needs of the application.

In another embodiment, the named entity tagging model can be deployed in any runtime space that can be called by the application server, so as to provide services for the relevant requirements of the application.

In another embodiment, the intention analysis model and the named entity tagging model may be encapsulated by a standardized service, the standardized service provides an interface for calling to receive input of text information, and then calls the intention analysis model and the named entity tagging model respectively to obtain corresponding result data and returns the result data correspondingly.

In another embodiment, the text information processed by the named entity tagging model may be first text information, after predicting each element information and its corresponding element tag according to the first text information by the named entity tagging model, all the predicted element information is constructed as session information indicating the same session, the session information is still text information, and then the session information is input into the intention analysis model as second text information alone or in combination with the first text information for predicting the corresponding target intention. Therefore, the input standardization degree of the intention analysis model can be improved, and the intention analysis model can predict the corresponding target intention more accurately.

On the basis of the above principle, please refer to fig. 5, in an embodiment of the method for responding to a customer service mail according to the present application, comprising the following steps:

s1100, acquiring mail content of a newly added mail in a target mailbox;

a merchant of a cross-border e-commerce typically operates one or more online stores, each operating in a respective independent station, each of which may be affiliated with a different e-commerce platform.

In one embodiment, when a customer user of any online store needs to submit a customer service question, the relevant question description can be edited and submitted through a corresponding page of the online store, and the question description submitted by the user is used as a mail text and sent to a default customer service mailbox address to enter a corresponding mailbox server. In another embodiment, the user can edit the corresponding problem description through the email box of the user and send the corresponding problem description to the designated customer service email address by himself to enter the corresponding email server.

Therefore, in the customer service system of the merchant, data communication connection to each customer service mailbox address is established in advance, and the e-mails are pulled to each corresponding e-mail mailbox in an incremental manner in a timed or non-timed manner, so that the newly added mails in each customer service mailbox are continuously acquired.

After a new mail of a customer service mailbox is obtained, the new mail can be analyzed, so that the mail content in the new mail can be read for implementing prediction. The mail content may be specified in advance according to actual needs, and specific components thereof may be specified, for example, the mail content may include only a mail body, may include a mail subject and a mail body, may include a mail body and a signature file, or further integrate the mail subject, the mail body, the mail signature file, and the like.

Step S1200, determining a target intention and session information corresponding to the mail content, wherein the session information comprises at least one item of element information for defining a customer service question corresponding to the target intention, the customer service question is defined by corresponding complete element information, and the complete element information comprises a plurality of element information pointed by a plurality of preset element labels;

in the application, a standardized service is realized in advance, and is used for predicting the target intention of the corresponding customer service problem according to the mail content and extracting the session information provided corresponding to the customer service problem.

The objective intention means that after customer service problems are classified in advance, a plurality of types of problems are formed, a corresponding intention type is calibrated corresponding to each type of problem, when one mail content is predicted to be one intention type, the intention type is an objective intention corresponding to the mail content, and therefore a customer service problem gate class to which the mail content belongs is determined. For example, for different door categories such as ' product failure ', ' logistics failure ' \8230; ' 8230; ' and the like, the corresponding target intentions can be correspondingly and firstly marked as ' failure ', ' logistics ' \8230; ' 8230, and the like, and can be represented by 0,1 ' \8230; ' 8230, and the like at the computer level.

The session information is a set of each element information for describing a certain customer service problem, and this set of element information constitutes description information of the customer service problem. Generally, complete description of a customer service problem requires providing a plurality of different element information according to a predetermined specification, and when a session information completely contains all the element information in the predetermined specification, a complete description of the corresponding customer service problem is formed. Each element information may be referred to using a corresponding element tag for calling, and thus, may be defined using a corresponding set of element tags when a predetermined specification of a customer service problem is given. It is understood that the element label sets of different customer service problems may be the same or different, and may be predetermined according to actual needs.

For example, describing a customer service problem corresponding to a fault, according to a preset specification, element information corresponding to element tags such as an order number, a contact number, a commodity identification and the like needs to be provided, when all the element information is included in session information, a complete description of the customer service problem is formed, otherwise, the session information is incomplete in describing the customer service problem.

Mail content extracted from a newly added mail is transferred to the standardized service by calling an interface provided by the standardized service, and the standardized service further determines corresponding target intentions and session information according to the mail content, wherein the target intentions are usually determined and unique, and the session information usually comprises element information corresponding to one or more element labels, but it should be noted that the element information contained in the session information may still not completely correspond to all element information required by a customer service problem pointed by the target intentions.

In an embodiment of the standardized service, an intention analysis model and a named entity tagging model can be respectively adopted to correspondingly determine corresponding target intention and session information according to the mail content; in another embodiment, the named entity tagging model may be used to predict the mail content to obtain corresponding session information, and then the intention analysis model may be used to predict a corresponding target intention according to the session information and the text set of the mail content.

In some embodiments, encapsulation of the standardized service can be dispensed with, and corresponding calls can be directly made to the intention analysis model and the named entity tagging model, so as to realize prediction of corresponding target intention and session information according to the mail content.

Step S1300, determining a corresponding dialog template according to the target intention, wherein the dialog template comprises a preset format text and an element label corresponding to the element information lacking in the session information, and the lacking element information is determined by the session information relative to the complete element information of the customer service problem;

as previously mentioned, each target intent is actually directed to an intent type, corresponding to a class of customer service questions. The application is provided with a corresponding dialogue template for each type of customer service problem in advance. The dialogue template can be used for replying the new mail so as to guide the consumer user to continuously perfect the conversation information and provide the element information which is lacked in the conversation information. Therefore, the dialog template comprises at least two parts of information, wherein the first part can be a preset format text for giving prompt information for guiding the user to supplement the materials; the second part may be one or more element tags of the element information that is missing when the session information corresponds to all the element information required for the customer service problem.

The following is an example of a dialog template:

we want to solve the problem quickly for you, but in the process of checking information provided by you, we lack further information, and have to trouble you to provide further materials as follows:

# name:

order number:

after receiving the element information corresponding to the above element tag supplemented by you, we will reply to you!

Thank you for your vertical inquiry!

As can be seen from the above example, the element tag defined by "#" is used to instruct the user to submit the corresponding element information.

In one embodiment, a single standardized dialog template may be provided for each type of customer care issue, whereby efficiency of the lead may be improved by giving the element tags of all the element information required for one type of customer care issue at once in order to lead the consumer user to provide the full amount of element information.

In another embodiment, for the same type of customer service problem, a plurality of dialog templates may be preset, each dialog template including different conditions in which element information is missing to provide different combinations of element tags, and when a corresponding dialog template needs to be matched, the element tags of the element information of the session information and the element tags of the dialog template are integrated and then matched with the set of element tags in the preset specification of the customer service problem, so as to determine the dialog template matching the session information, so that the element tags provided in the dialog template are exactly the element tags of all the element information in the session information that is missing from the description information corresponding to the customer service problem.

And step S1400, replying the newly added mail with the conversation template.

And after determining the corresponding conversation template according to the mail content, calling a mail sending interface, taking the sender of the newly added mail as a receiver, and sending the conversation template as a mail text to a consumer user.

After the consumer user receives the conversation template, the element information can be further improved according to the element label in the conversation template, so that the element information required by the customer service problem reflected by the element information is more complete, and the solution of the customer service problem can be conveniently and correspondingly determined in the follow-up process.

It is understood that, the newly added email after the customer user responds to the dialog template may still be processed iteratively through the steps S1100 to S1400, so as to further determine whether the session information is complete, and if the session information is still incomplete, the customer may be further guided to complete the session information until the description of the customer service problem by the session information meets the requirement of the preset specification corresponding to the customer service problem.

According to the embodiment, the target intention and the conversation information of the customer service problem corresponding to the added mail are determined according to the mail content of the added mail submitted by the user, then the conversation template is determined according to the target intention, the conversation template comprises the element labels which are deficient in the conversation information corresponding to the customer service problem, the deficient element labels are determined according to the complete element information of the customer service problem, the added mail is responded by the conversation template, the element labels can play a role in guiding a sender of the added mail to supplement the complete information, various added mails are automatically responded by adopting the mode, the conversation information of the same sender can be supplemented continuously, each customer service problem can obtain the complete element information finally, the solution efficiency of the customer service problem is improved, and the conversation efficiency in the customer service process is further improved by means of machine intelligence.

On the basis of any embodiment of the application, determining the target intention and the session information corresponding to the mail content comprises the following steps:

step S1210, inputting the mail content into a preset named entity marking model, determining one or more element information in the named entity marking model, and forming session information;

before the mail content is used to determine the target intention and session information, in one embodiment, the mail content is subjected to formatting pre-processing, including but not limited to: and removing punctuation marks, stop words, invalid characters and the like to obtain the purified text of the mail content.

Then, a bag-of-words model or other word segmentation algorithms such as N-Gram and the like are adopted to segment the purified text to obtain a plurality of lemmas, a preset mapping dictionary is adopted to inquire out the characteristic value corresponding to each lemma, and the characteristic values of all the lemmas are constructed into the coded information corresponding to the purified text.

In one embodiment, in the encoded information, position information of each lemma in the mail content may be further superimposed, and the position information may be absolute position information and/or relative position information.

And after the coded information corresponding to the mail content is determined, inputting the coded information into a named entity marking model, extracting corresponding deep semantic information by the named entity marking model according to the coded information, then extracting each named entity according to the deep semantic information, and determining element labels corresponding to the named entities as corresponding element information. And the mapping relation data between the element label and the element information determined by the named entity labeling model can form the session information corresponding to the mail content.

Step S1220, inputting the mail content and/or the session information into a preset intention analysis model, and predicting a corresponding target intention.

Similarly, in an embodiment, the coded information corresponding to the purified text of the mail content may be synchronously input into an intention analysis model adopted in the present application, deep semantic information in the coded information is extracted by the intention analysis model, and then the deep semantic information is mapped to a preset classification space, and an intention type indicated by a category corresponding to the maximum classification probability is determined as a target intention.

In another embodiment, the encoded information of the session information may be used instead as the input of the intention analysis model, thereby determining the corresponding target intention.

In still another embodiment, the session information and the mail content may be jointly encoded to obtain corresponding encoded information thereof as an input of the intention analysis model, so as to determine a corresponding target intention.

In another embodiment, the intention analysis model and the named entity tagging model can have a shared backbone network, and accordingly, by inputting the encoded information of the purified text of the mail content into the backbone network, the target intention and the session information corresponding to the mail content can be respectively determined in two branch networks of the whole network architecture.

According to the embodiment, the mail content can be flexibly determined by combining the named entity labeling model and the intention analysis model, the corresponding target intention and the corresponding session information can be intelligently predicted due to the application of the deep learning model, the efficiency is high, the accuracy is high, the manual intervention is avoided, and the batch processing of a large amount of mail content is facilitated.

On the basis of any embodiment of the present application, please refer to fig. 6, determining a dialog template corresponding to the target intent includes:

step S1310, determining a corresponding dialogue relation tree according to the target intention, wherein the dialogue relation tree comprises a plurality of dialogue units pre-associated according to a tree structure, and each dialogue unit correspondingly provides the dialogue template;

in the embodiment, a large number of preset dialogue templates are organized through the dialogue relation tree so as to realize effective organization of multi-customer service problems and multiple element information missing conditions, and the determined dialogue templates have finer and more precise functions and steps in a guiding process for perfecting the dialogue information for a consumer user. For the convenience of retrieval, a dialog relation tree can be established correspondingly to each target intention, so that when a target intention is determined, the corresponding dialog relation tree can be determined.

In an embodiment of the dialog relationship tree, the dialog relationship tree may be constructed based on relational data, and includes a plurality of dialog units, each of which stores a corresponding dialog template and hierarchical dependency relationship information thereof in association with each other, the dialog template includes an element tag set corresponding to an element information missing condition in addition to a preset format text, and different dialog units, that is, different dialog templates, have different element information missing conditions corresponding to each other, so as to include different element tag sets.

In another embodiment, the dialogue relation tree can also be constructed based on a knowledge graph, wherein the dialogue templates corresponding to the dialogue units are stored in a node manner, and the edge connection information is constructed between any two dialogue units according to the actual hierarchical dependency information.

It can be seen that, a hierarchical relationship between massive dialog templates is constructed in a dialog relationship tree manner, which is helpful for improving matching efficiency between the dialog information and the dialog templates, and thus, in some embodiments, all element tags of description information required to be provided for each corresponding subdivided customer service problem are directly defined by the dialog templates as preset specifications, and after a consumer user replies to a dialog template provided by a first newly added email, it is not necessary to iterate step S1100 to step S1400 again for subsequent newly added emails replied by the user, but only by using a first determined target intention, in a corresponding dialog relationship tree thereof, the element information supplemented by the user is combined with the first determined dialog intention, and then the result is compared with the preset specification corresponding to the dialog template in a lower-level dialog unit of the first determined dialog template to match the dialog template, so as to gradually guide the consumer user to provide corresponding element information, and implement a multi-branch and multi-detail guidance flow.

Step S1320, matching the element information contained in the session information with the complete element information of the customer service problem, and determining the missing element label in the session information;

when the conversation template is matched with the conversation information in one conversation relation tree, because the conversation information contains mapping relation data between the element labels and the element information, the element label set of each element information existing in the conversation information is compared with the element label set corresponding to the complete element information of the customer service problem corresponding to the target intention, and a difference element label subset is obtained, wherein the element label subset contains all the element labels which are lacked in the conversation information and correspond to the customer service problem.

Step S1330, determining a dialog template completely matching the session information according to the missing element tag in the session information.

Because each conversation template correspondingly comprises an element label set under the condition of lacking element information, the element label subset lacking in the conversation information and each conversation template under the conversation relation tree corresponding to the target intention are subjected to set operation to obtain the conversation templates with the same set, namely the conversation templates suitable for replying the newly-added mail.

According to the embodiment, the conversation template for guiding the consumer user to perfect the customer service problem description can be determined according to the target intention and the conversation information, so that the conversation template has a flow guiding function, and before the consumer user incompletely describes the customer service problem, an intermediate process before the customer service problem is solved is intelligently guided, so that the solving efficiency of the customer service problem can be improved, and manpower and material resources are saved.

On the basis of any embodiment of the present application, please refer to fig. 7, determining a dialog template completely matching with the missing element tag according to the missing element tag in the session information includes:

step S1331, determining a platform identifier according to the receiver mailbox of the newly added mail;

in order to distinguish different independent stations of the merchant conveniently, a special customer service mailbox, namely a receiver mailbox, can be configured for the online shop of each independent station, so that each independent station can share one customer service mailbox, and the different independent stations can be distinguished through the customer service mailboxes.

In an embodiment, considering that different independent stations are usually supported by different e-commerce platforms, an enterprise mailbox provided by a mailbox server of the corresponding e-commerce platform is also provided, and such enterprise mailbox is usually distinguished by using a website of the e-commerce platform or the independent station as a suffix and using the suffix, so that after a new added mail is received, a suffix part of a recipient mailbox, specifically, a recipient mailbox address of the recipient mailbox, can be extracted and used as a platform identifier corresponding to the new added mail to represent a source independent station of the new added mail.

Step S1332, determining a dialog template set which is completely matched with the dialog template set according to the missing element labels in the session information;

in the dialogue relation tree, aiming at the preset specification of the same customer service problem, a plurality of dialogue templates corresponding to texts with different styles, languages and different formats can be set, and the mapping relation is established between the dialogue templates and the platform identifier in advance, so that the element label sets of the different dialogue templates corresponding to the same customer service problem are the same, and therefore the dialogue templates can be used for correspondingly serving different independent stations and providing the guide information of the same element label.

Therefore, when the dialog template is matched in the dialog relation tree corresponding to the target intention by using the session information, a plurality of dialog templates can be determined according to the matching relation between the missing element label subset in the session information and the element label set in the dialog template to form a dialog template set. It will be readily appreciated that each dialog template in the set of dialog templates is identical in ideographic and element tag formation as corresponding to the same customer service question.

And S1333, determining a dialogue template corresponding to the platform identification in the dialogue template set.

Furthermore, by using the mapping relation data between the conversation templates and the platform identifiers determined according to the email addresses of the recipients of the newly added emails, only one conversation template in the conversation templates can be determined, and the conversation template can be used for replying the newly added emails.

According to the embodiment, the platform identification is set, the mailbox of the recipient of the newly-added mail and the conversation template are associated, the requirements of different independent stations can be met, the personalized conversation template is provided for the reply mail of the customer service problem of each independent station, and the communication differences such as language, habits, styles and the like are overcome.

On the basis of any embodiment of the present application, please refer to fig. 8, before acquiring the mail content of the newly added mail in the target mailbox, the method includes:

step S2100, pulling at least one newly added mail from a target mailbox according to preset configuration;

in order to pull the mail from the service mailbox corresponding to each independent station, a preset configuration of the service mailbox may be provided, where the preset configuration is determined according to a mailbox-related communication protocol, such as POP, IMAP, SMTP, or other protocols customized by a mailbox server, where POP and IMAP are recipient protocols, and SMTP is a delivery protocol. The protocol is suitable for receiving the mail and provides various corresponding parameters according to the protocol to form corresponding preset configuration. The various parameters may be, for example, an account number, an authorization code, a port number, and the like of the customer service mailbox, which are specifically determined according to an interface specification of a corresponding protocol.

When the preset configuration of the service mailbox is provided, the mail can be pulled to the corresponding service mailbox at a certain reasonable frequency, for example, every time of 10 to 20 minutes. Of course, the corresponding customer service mailbox or other mailboxes can be used for sending the mails at any time. The frequency of pulling the mails is reasonably set, so that the balance between mail updating and system computing power can be realized. When a mail is pulled to a target mailbox, i.e. a target customer service mailbox, according to a certain frequency, the mail is usually acquired in an increment mode, and the mail pulled each time is treated as a newly added mail.

Step S2200, construct the correspondent mail message to each new mail;

the customer service system may be responsible for processing newly added mails of a plurality of customer mailboxes at the same time, and the newly added mails of the plurality of customer service mailboxes are easy to block when the number of the newly added mails is too large.

And step S2300, marking each mail message as a to-be-executed state, adding the mail message into the task list, and performing out-of-line consumption.

And then, adding the mail message into a task list, wherein the task list is used for storing the processing state of each mail message and assisting in realizing a message mechanism of mail processing, and accordingly, when the mail message of a newly added mail is added into the task list, the state initialization of the mail message is marked as a to-be-executed state.

According to the embodiment, in order to prevent the response disorder caused by a large number of newly added mails, the ordered consumption of the mail messages is realized by establishing a message mechanism by virtue of the task list and recording the state of the mail messages of each newly added mail, so that the higher scheduling efficiency of the scheduling of each newly added mail can be ensured.

On the basis of any embodiment of the present application, please refer to fig. 9, marking each email message as a to-be-executed state, and after adding the email message to the task list for in-order listed consumption, the method includes:

step S2400, orderly reading the mail messages in the task list in the to-be-executed state for consumption;

when the newly added mails are processed, orderly scheduling is carried out according to the mail messages in the task list, specifically, only the mail messages marked as the to-be-executed state are consumed, and the mail content of the corresponding newly added mails is extracted from the mail messages to be used for determining the conversation template. To indicate the status of the mail message in time, when the mail message starts to be consumed, the status thereof may be modified to be in a processing state to indicate the difference.

In one embodiment, multiple concurrent threads can be started, and multiple mail messages marked as a to-be-executed state are consumed at a time, so that the consumption efficiency of the mail messages is improved.

The method for consuming the mail message refers to starting a corresponding thread to process the mail message listed from a corresponding queue according to a computer message queue operation mechanism, wherein the thread is realized in advance as business logic corresponding to the mail message, and the business logic related to the corresponding newly added mail is completely processed by consuming the mail message, for example, a conversation template corresponding to the newly added mail is finally determined.

Step S2500, marking the consumed email message which is replied by adopting the dialogue template as a finished state;

after each email message is consumed, that is, the conversation template of the corresponding newly added email is determined and replied, in order to avoid repeated processing or missed processing of the email message, the state of the email message may be further set to be a finished state.

And step S2600, transmitting the email message which is consumed but not replied by adopting the conversation template to a manual customer service interface for processing, and marking the email message as a manual processing state.

When the consumption process of a mail message is finished but a corresponding conversation template is not obtained, aiming at the situation, the problem which cannot be solved by the automation process usually exists, and manual agent intervention is needed, so that the corresponding mail message is further transmitted to a manual agent interface and sent to a manual agent user, and the manual agent user can perform further processing.

According to the embodiment, the state of the mail messages is maintained in time, so that massive mail messages passing through the task list can be guaranteed to be orderly and stably consumed, and the efficient operation of the customer service system is guaranteed.

On the basis of any embodiment of the present application, please refer to fig. 10, acquiring mail content of a newly added mail in a target mailbox includes:

step S1110, extracting a sender mailbox address, a mail subject, a mail body, and a mail signature file in the newly added mail in response to the consumption event of the mail message in the to-be-executed state in the task list;

on the basis of establishing a message mechanism by means of the task list, the mail messages in the task list in the state to be executed are sequentially listed according to a queue mechanism, and a consumption event is correspondingly triggered when one mail message is listed.

In response to the consumption event, the newly added mails carried in the consumption event can be analyzed, and the mailbox address of the sender, the mail subject, the mail body and the mail signature file in the newly added mails are respectively extracted.

Step S1120, packaging any items of the mail subject, the mail body and the mail signature file into mail content;

in the various embodiments described in the foregoing application, depending on the requirement of the formation of the mail content, the content of each part extracted from the newly added mail may be used flexibly, for example, the mail content may be formed by combining the mail subject with the mail body, or the mail content may be formed by combining the mail body with the mail signature file, or the mail content may be formed by combining the mail subject, the mail body, and the mail signature file, and so on. Currently, as described in one embodiment above, it is still possible to use the mail body alone as the mail content.

Step S1130, determining the sender mailbox address as a target mailbox address for replying the new mail.

And after determining a corresponding conversation template aiming at the newly added mail, configuring the sender mail address as a target mail address for replying the newly added mail, namely a receiver mail address, and simultaneously using the conversation template as a mail text and referring the original theme of the newly added mail as a reply theme so as to construct a reply mail, and sending the reply mail through a preset sending interface to reply the newly added mail.

According to the embodiment, in the process of consuming each mail, standardized analysis can be performed on the corresponding newly-added mail, the mail content is packaged, the mailbox address of the sender is determined, the early-stage standardization work is done for the subsequent reply consumer user, and the information processing efficiency is improved.

On the basis of any embodiment of the present application, please refer to fig. 11, where after replying to the new email with the dialog template, the method includes:

s1500, monitoring a reply mail replied by the sender of the newly added mail based on the conversation template;

after a new mail is replied by a dialogue template, the dialogue template is sent to a corresponding consumer user, the consumer user can reply according to the prompt message given by the dialogue template, the element information corresponding to the element label in the dialogue template is supplemented, and then the mail is resent, so that the customer service system can pull the corresponding reply mail.

Step S1600, extracting the element information submitted corresponding to each element label in the dialogue template, perfecting the conversation information and obtaining complete conversation information;

for the reply e-mail supplemented based on the conversation template, because the information therein is relatively straightforward, it is possible to determine various element information provided in the e-mail content of the reply e-mail based on the e-mail content in the reply e-mail directly by adopting a rule matching manner or by adopting a named entity tagging model of the present application without iterative processing of the processes of the foregoing embodiments of the present application, and supplement the element information to the conversation information that has been previously determined for the newly added e-mail, so that the conversation information can completely correspond to the description information of one customer problem.

Step S1700, predicting a corresponding target intention according to the complete conversation information;

after the complete session information is determined, the intention analysis model of the application can be called again to determine the target intention of the user, and the intention analysis model is expected to obtain a more accurate prediction result due to the fact that the complete session information provides key semantics.

In another embodiment, the mail content in the newly added mail and the complete conversation information obtained according to the reply mail may be merged and then input into the intention analysis model, the mail content provides comprehensive semantics, the complete conversation information provides key semantics, and the corresponding target intention is predicted more accurately by virtue of the reasoning ability of the intention analysis model.

The target intention determined in the above manner may be changed with a very small probability with respect to the target intention predicted solely for the mail content of the newly added mail, and at this time, the process of matching the conversation template may be regressed. However, in most cases, the target intention determined this time serves only to confirm the target intention predicted from the mail content of the newly added mail, and serves to accurately understand the type of the customer service problem.

And S1800, determining a reply text mapped with the target intention according to the target intention, and responding the reply mail by using the reply text.

After the target intention is confirmed, a reply text corresponding to the target intention can be called, and the reply text is sent to the address of the recipient of the newly added mail and is delivered to the consumer user. The reply text, which is usually a description text for guiding the user to solve the corresponding customer service question, may be preset by a person skilled in the art, and in principle, each customer service question may provide a reply text correspondingly.

According to the embodiment, after the customer user perfects the session information, the type of the customer service problem proposed by the user is confirmed by predicting the target intention again, and the preset reply text is called after the confirmation to automatically provide the solution of the customer service problem, so that the automatic response is realized, the digestion efficiency of the customer service problem is improved, and the accuracy and the reliability of automatically solving the customer service problem are improved.

Referring to fig. 12, a customer service email responding apparatus according to an aspect of the present application includes an email obtaining module 1100, an information analyzing module 1200, a template determining module 1300, and an email replying module 1400, where: the mail acquiring module 1100 is configured to acquire the mail content of the newly added mail in the target mailbox; the information analysis module 1200 is configured to determine a target intention corresponding to the mail content and session information, where the session information includes at least one item of element information for defining a customer service question corresponding to the target intention, the customer service question is defined by corresponding complete element information, and the complete element information includes a plurality of element information pointed by a plurality of preset element tags; the template determining module 1300 is configured to determine a corresponding dialog template according to the target intention, where the dialog template includes a preset format text and an element tag corresponding to missing element information of the session information, and the missing element information is determined by the session information relative to complete element information of the customer service problem; the mail replying module 1400 is configured to reply to the new mail with the conversation template.

On the basis of any embodiment of the present application, the information analysis module 1200 includes: the element labeling unit is used for inputting the mail content into a preset named entity labeling model, determining one or more element information in the named entity labeling model and forming session information; and the intention mapping unit is used for inputting the mail content and/or the session information into a preset intention analysis model and predicting a corresponding target intention.

On the basis of any embodiment of the present application, the template determining module 1300 includes: the dialog determining unit is set to determine a corresponding dialog relation tree according to the target intention, the dialog relation tree comprises a plurality of dialog units pre-associated according to a tree structure, and each dialog unit correspondingly provides the dialog template; a deficiency determining unit configured to match element information included in the session information with complete element information of the customer service problem, and determine a deficient element label in the session information; and the template determining unit is used for determining a dialog template completely matched with the element tag according to the missing element tag in the session information.

On the basis of any embodiment of the present application, the template determining unit includes: the platform identification subunit is set to determine a platform identifier according to the receiver mailbox of the newly added mail; the set anchoring subunit is set to determine a dialog template set which is completely matched with the set anchoring subunit according to the missing element labels in the session information; and the template hit subunit is configured to determine a dialog template corresponding to the platform identifier in the dialog template set.

On the basis of any embodiment of the present application, the module 1100, prior to the mail acquiring, includes: the mail pulling module is set to pull at least one newly added mail to the target mailbox according to the preset configuration; the message generating module is set to construct a corresponding mail message for each newly added mail; and the message listing module is set to mark each mail message as a to-be-executed state and add the mail message to the task list for sequential listing and consumption.

On the basis of any embodiment of the present application, the message enlisting module includes: the message consumption module is set to read the mail messages in the task list in a waiting execution state in order for consumption; a completion marking module configured to mark the email message that has been consumed and replied by using the dialogue template as a completed state; and the interface skip module is set to transmit the mail message which is consumed but not replied by adopting the conversation template to the manual customer service interface for processing, and mark the mail message in a manual processing state.

On the basis of any embodiment of the present application, the mail acquiring module 1100 includes: the mail analysis unit is set to respond to the consumption event of the mail message in the state to be executed in the task list and extract the sender mailbox address, the mail subject, the mail body and the mail signature file in the newly added mail; the content packaging unit is used for packaging any items in the mail subject, the mail body and the mail signature file into mail content; and the mailbox determining unit is set to determine the sender mailbox address as a target mailbox address for replying the newly added mail.

On the basis of any embodiment of the present application, the email reply module 1400 includes: the after-treatment tracking module is arranged to monitor a reply email which is replied by a sender of the newly-added email based on the conversation template; the conversation perfecting module is set to extract element information submitted by each element label corresponding to the conversation template, perfects the conversation information and obtains complete conversation information; the intention determining module is used for predicting a corresponding target intention according to the complete conversation information; and the problem solving module is set to determine a reply text mapped with the target intention according to the target intention and answer the reply email with the reply text.

Another embodiment of the application also provides customer service mail response equipment. As shown in fig. 13, the internal structure of the customer service mail answering apparatus is schematically illustrated. The customer service mail answering device comprises a processor, a computer readable storage medium, a memory and a network interface which are connected through a system bus. The computer-readable non-volatile readable storage medium of the customer service mail response device stores an operating system, a database and computer-readable instructions, the database can store information sequences, and the computer-readable instructions can enable a processor to realize a customer service mail response method when being executed by the processor.

The processor of the customer service mail response equipment is used for providing calculation and control capacity and supporting the operation of the whole customer service mail response equipment. The memory of the customer service mail response device can store computer readable instructions, and when the computer readable instructions are executed by the processor, the processor can execute the customer service mail response method. The network interface of the customer service mail response equipment is used for connecting and communicating with the terminal.

It will be understood by those skilled in the art that the structure shown in fig. 13 is a block diagram of only a portion of the structure relevant to the present application, and does not constitute a limitation on the customer service mail answering device to which the present application is applied, and a particular customer service mail answering device may include more or less components than those shown in the figure, or may combine certain components, or have a different arrangement of components.

In this embodiment, the processor is configured to execute specific functions of each module, sub-module, and unit in fig. 12, and the memory stores program codes and various data required for executing the modules or sub-modules. The network interface is used for realizing data transmission between user terminals or servers. The non-volatile readable storage medium in the present embodiment stores program codes and data necessary for executing all modules in the customer service mail answering device of the present application, and the server can call the program codes and data of the server to execute the functions of all modules.

The present application also provides a non-transitory readable storage medium having stored thereon computer readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the customer service mail response method of any of the embodiments of the present application.

The present application also provides a computer program product comprising computer programs/instructions which, when executed by one or more processors, implement the steps of the method as described in any of the embodiments of the present application.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments of the present application may be implemented by hardware related to instructions of a computer program, which may be stored in a non-volatile readable storage medium, and when executed, may include the processes of the embodiments of the methods as described above. The storage medium may be a computer-readable storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a Random Access Memory (RAM).

In summary, the method and the device can guide the user to completely reflect all the element information of the customer service problem through the mail, so that the communication efficiency of the user in the middle process of reflecting the customer service problem can be improved, the complete element information corresponding to the customer service problem can be obtained quickly, the customer service problem can be solved quickly based on the complete element information, the information processing efficiency of the customer service problem is improved, and the customer service satisfaction degree of the business of the cross-border e-commerce is improved.

Claims

1. A customer service mail response method is characterized by comprising the following steps:

acquiring mail content of a newly added mail in a target mailbox;

and replying the new mail with the conversation template.

2. The customer service mail response method according to claim 1, wherein determining the target intention and the session information corresponding to the mail content comprises:

3. The customer service mail answering method according to claim 1, wherein determining its corresponding conversation template based on the target intent comprises:

4. The method of claim 1, wherein determining a completely matching conversation template based on missing element tags in said conversation information comprises:

5. The customer service mail response method according to claim 1, wherein before obtaining the mail content of the newly added mail in the target mailbox, the method comprises:

constructing a corresponding mail message for each newly added mail;

and marking each mail message as a to-be-executed state, and adding the mail message into a task list for listed consumption.

6. The customer service mail response method of claim 5, wherein marking each mail message as pending for addition to the task list after the ordered listing of the consumption, comprises:

marking the email messages which are consumed and replied by adopting the dialogue template as a finished state;

7. The customer service mail response method according to claim 6, wherein the obtaining of the mail content of the newly added mail in the target mailbox comprises:

8. The customer service mail response method according to any one of claims 1 to 7, wherein after replying to the new mail with the conversation template, the method comprises:

and determining a reply text mapped with the target intention according to the target intention, and answering the reply mail with the reply text.

9. A customer service mail answering device, comprising:

and the mail reply module is set to reply the newly-added mail by using the conversation template.

10. A customer service mail answering apparatus comprising a central processor and a memory, wherein the central processor is configured to invoke execution of a computer program stored in the memory to perform the steps included in the method according to any one of claims 1 to 8.

11. A non-transitory readable storage medium storing a computer program in the form of computer readable instructions, the computer program when invoked by a computer performing the steps comprised by the method according to any one of claims 1 to 8.

12. A computer program product comprising computer programs/instructions for performing the steps comprised by the method according to any one of claims 1 to 8 when called upon by a processor.