CN115147132A

CN115147132A - Method and device for generating customer service conversation template

Info

Publication number: CN115147132A
Application number: CN202210775660.1A
Authority: CN
Inventors: 樊艳
Original assignee: Alibaba China Co Ltd
Current assignee: Alibaba China Co Ltd
Priority date: 2022-07-01
Filing date: 2022-07-01
Publication date: 2022-10-04

Abstract

The embodiment of the application provides a method and a device for generating a customer service conversation template, wherein entity recognition processing and association analysis processing are carried out on conversation sentences of multiple rounds of questions and answers, so that entity mining is carried out on the conversation sentences of each round of questions and answers, one or more frequent interaction attributes corresponding to each round of questions and answers are determined, the customer service conversation template can be generated based on the frequent interaction attributes, a merchant can realize rapid configuration of a customer service robot by using the customer service conversation template, and the cost required by the merchant for configuring the customer service robot can be effectively reduced.

Description

Method and device for generating customer service conversation template

Technical Field

The embodiment of the application relates to the technical field of computers, in particular to a method and a device for generating a customer service dialogue template.

Background

With the development of computer technology, it becomes possible to apply the customer service robot to the customer service industry.

In order to enable the customer service robot to provide accurate response service for the customer like real person customer service, the customer service robot needs to be configured in advance for customer service conversation, so that the customer service robot can perform interactive question answering with the customer according to a configured conversation template.

In the existing scenario, a merchant often needs to spend a great deal of labor cost and time cost to write a customer service conversation template, the template generation system is used for a customer service robot, the workload is large, and the template generation cost is high.

Disclosure of Invention

The embodiment of the application provides a method and a device for generating a customer service conversation template, so that the customer service conversation template can be quickly generated, and the workload of a merchant and the cost required by configuring a customer service robot are reduced.

In a first aspect, an embodiment of the present application provides a method for generating a customer service dialog template, including:

determining a plurality of rounds of question-answering dialogue sentences to be processed; carrying out entity recognition processing on the dialogue sentences of each round of question and answer to obtain question and answer categories of each round of question and answer and interaction attributes corresponding to each round of question and answer; processing the question and answer categories of each round of question and answer and the interaction attributes corresponding to each round of question and answer by using a correlation analysis method, and determining one or more frequent interaction attributes corresponding to each question and answer category; and generating a customer service conversation template according to one or more frequent interaction attributes corresponding to each question and answer class.

It can be known that, in the embodiment of the present application, entity recognition processing and association analysis processing are performed on the dialog sentences of multiple rounds of questions and answers, so as to perform entity mining on the dialog sentences of each round of questions and answers, thereby determining one or more frequent interaction attributes corresponding to each round of questions and answers, and based on the frequent interaction attributes, a customer service dialog template can be generated, and a merchant can use the customer service dialog template to implement rapid configuration of a customer service robot, thereby effectively reducing the cost required by the merchant for configuring the customer service robot.

Optionally, the processing of the question-answer category of each round of question-answer and the interaction attribute corresponding to each round of question-answer by using a correlation analysis method includes: clustering the question-answer classes of each round of question-answers and the interaction attributes corresponding to each round of question-answers to obtain an interaction attribute set corresponding to each question-answer class, wherein the interaction attribute set comprises all interaction attributes corresponding to each question-answer class; and performing correlation analysis processing on all the interactive attributes in each interactive attribute set, and determining one or more frequent interactive attributes corresponding to each question and answer class.

Specifically, a specific obtaining mode of frequent interaction attributes is provided in the embodiment of the present application, wherein one or more frequent interaction attributes corresponding to each question and answer class are determined by clustering question and answer classes of each round of questions and answers and interaction attributes of each round of questions and answers, and then performing association analysis processing on an interaction attribute set obtained by clustering.

Optionally, determining one or more frequent interaction attributes corresponding to each question and answer class includes: traversing interaction attribute combination modes of M interaction attributes aiming at M interaction attributes in any interaction attribute set to obtain X interaction attribute combinations; wherein, the first and the second end of the pipe are connected with each other,

n is the number of the interactive attributes in the interactive attribute combination, and is a positive integer greater than or equal to 1; determining the support degree of each interactive attribute combination in the X interactive attribute combinations in an interactive attribute set; determining an increment function value corresponding to each interactive attribute combination according to the value of N corresponding to each interactive attribute combination and the support degree of each interactive attribute combination; and determining one or more frequent interaction attributes of the question and answer classes corresponding to the interaction attribute set according to the increment function values corresponding to the interaction attribute combinations.

It can be known that, in the embodiment of the present application, when performing correlation analysis processing on each interactive attribute, the number and the support degree of the interactive attributes in the interactive attribute combination are used as variables of the incremental function, so that an incremental function value corresponding to each interactive attribute combination can be calculated during the correlation analysis processing, so as to improve the influence degree of the number on the analysis result under the condition that the number of the interactive attributes in the combination is not defined, and further enable one or more frequent interactive attributes of the obtained question and answer classes to be more suitable for generating customer service conversation templates.

Optionally, determining a dialog sentence of multiple rounds of questions and answers to be processed includes: acquiring a plurality of dialogue logs, wherein each dialogue log comprises a plurality of dialogue sentences; classifying and identifying the dialogue behaviors of each dialogue statement in each dialogue log respectively to obtain the dialogue behavior corresponding to each dialogue statement; and determining the dialog sentences of the multiple rounds of questions and answers to be processed according to the dialog behaviors of each dialog sentence.

Optionally, the classifying and identifying the dialogue behavior of each dialogue statement in each dialogue log is performed respectively to obtain the dialogue behavior corresponding to each dialogue statement, and the classifying and identifying includes: processing each dialogue statement according to a pre-trained dialogue behavior classification model to obtain a dialogue behavior of each dialogue statement; the dialogue behavior of the dialogue sentences belongs to one of a client answer behavior, a client intention behavior, a customer service question-asking behavior and a customer service answer behavior.

It can be known that, in the embodiment of the present application, classification and recognition processing of dialog statements in a large number of dialog logs is implemented by using a classification and recognition algorithm of dialog behaviors, so as to obtain corresponding dialog behaviors. By utilizing the dialogue behavior of each dialogue statement, the dialogue statements of multiple rounds of questions and answers to be processed can be quickly determined from a large number of dialogue logs so as to be subjected to subsequent processing.

Optionally, the entity recognition processing is performed on the dialog sentences of each round of question and answer to obtain question and answer categories of each round of question and answer and interaction attributes corresponding to each round of question and answer, and the method includes: carrying out entity extraction processing on the dialogue sentences of each round of question answering and carrying out alignment processing on the extracted entity results, obtaining entity information corresponding to each dialogue statement, wherein the entity information comprises a plurality of entity types and entity values corresponding to each entity type; and according to the type of the entity type, taking the entity value corresponding to the class as the question-answer class of each round of question-answers, and taking the attribute and the entity value corresponding to the attribute as the interaction attribute corresponding to each round of question-answers.

Optionally, the entity extraction processing is performed on the dialog sentences of each round of question answering, and the extracted entity results are aligned, so as to obtain entity information corresponding to each dialog sentence, including: aiming at any round of question and answer, performing entity extraction processing on each dialogue statement in the round of question and answer by using a preset named entity recognition model to obtain an entity type, an entity logic and an entity value of each dialogue statement; according to the entity logic of each dialogue statement, carrying out alignment judgment on the logic relationship between the entity type and the entity value of each dialogue statement, and selecting the entity type with the preset logic relationship and the corresponding entity value for each dialogue statement according to the judgment result; and judging whether each dialogue sentence of the round of question answering comprises the same or similar entity type, wherein the same or similar entity type and the corresponding entity value form the entity information of the round of question answering.

Optionally, the entity extraction processing is performed on each dialogue statement in the round of question and answer by using a preset named entity recognition model, and the processing includes: for each conversational sentence in each turn of question-answer sentences, and sequentially carrying out vector coding processing, logic extraction processing and entity extraction processing on the dialogue statement by using a preset named entity recognition model.

It can be known that entity extraction processing can be performed on the spoken sentence by using the named entity recognition model, so that entities with actual meanings in the conversational sentence can be extracted, attribute alignment processing is performed on the basis of the extracted entities, question and answer categories of each round of question and answer and interaction attributes included in each round of question and answer are obtained, and subsequent further processing is facilitated to obtain frequent interaction attributes.

Optionally, the customer service dialog template includes a customer service question template and a customer service answer template; generating a customer service dialogue template according to one or more frequent interaction attributes corresponding to each question and answer class, wherein the generation comprises the following steps: configuring a customer service question template according to one or more frequent interaction attributes corresponding to each question and answer class; and configuring a customer service answer template according to one or more frequent interaction attributes corresponding to each question and answer class.

In a second aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

at least one processor; and

a memory;

the memory stores computer-executable instructions;

the at least one processor executes computer-executable instructions stored by the memory, causing the at least one processor to perform a method as in the first aspect.

In a third aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when a processor executes the computer-executable instructions, the method according to the first aspect is implemented.

In a fourth aspect, embodiments of the present application provide a computer program product comprising computer instructions which, when executed by a processor, implement the method according to the first aspect.

The embodiment of the application provides a method and a device for generating a customer service conversation template, wherein entity identification processing and association analysis processing are carried out on conversation sentences of multiple rounds of questions and answers, so that entity mining is carried out on the conversation sentences of each round of questions and answers, one or more frequent interaction attributes corresponding to each round of questions and answers are determined, the customer service conversation template can be generated based on the frequent interaction attributes, a merchant can utilize the customer service conversation template to realize rapid configuration of a customer service robot, and the scheme can effectively reduce the cost required by the merchant for configuring the customer service robot.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure.

FIG. 1 is a schematic flow diagram illustrating the configuration and use of a customer service robot;

FIG. 2 is a drawing of a block diagram of a cell based on which the present application is based a schematic diagram of a network architecture;

fig. 3 is a flowchart illustrating a method for generating a customer service dialog template according to an embodiment of the present application;

FIG. 4 is a diagram illustrating a named entity recognition model according to an embodiment of the present disclosure;

FIG. 5 is a schematic view of a scenario for configuring intelligent customer service based on a customer service dialog template according to an embodiment of the present application;

fig. 6 is a schematic diagram of a hardware structure of an electronic device provided in the present application.

With the foregoing drawings in mind, certain embodiments of the disclosure have been shown and described in more detail below. The drawings and written description are not intended to limit the scope of the disclosed concepts in any way, but rather to illustrate the disclosed concepts to those skilled in the art by reference to specific embodiments.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the exemplary embodiments below do not represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

In the technical scheme of the application, the processing of collecting, storing, using, processing, transmitting, providing, disclosing and the like of various related information accords with the regulations of related laws and regulations, and does not violate the common customs of public sequences.

The following describes a scheme based on the present application, taking a customer service robot of a merchant carried on an e-commerce platform as an example. Fig. 1 is a schematic flow chart illustrating a configuration and a use of a customer service robot, as shown in fig. 1, a customer may establish a customer service communication channel with a merchant through a customer service interface provided by an e-commerce platform, and enjoy customer service provided by the merchant through the customer service communication channel. As described above, in order to enable the customer service robot to implement the intelligent customer service function, a merchant generally needs to configure a customer service dialog for the customer service robot in advance, so that the customer service robot can perform an interactive question and answer with a customer according to a configured dialog template.

As shown in fig. 1, for a merchant, in order to enable a customer service robot to provide accurate customer service to customers like a real person customer service, a corresponding dialog template needs to be configured for the customer service robot. In the customer service scenario, a conversation template refers to a conversational template that may be used to interactively respond with a customer.

For example, for a customer-initiated question: "how can you recommend a 200ML facial cleanser product for I? "the customer service robot will determine the answer words from the dialogue template according to the key information such as" facial cleanser "and" 200ML ", for example," ask what are your skin? ".

Then, when the client answers: when the 'my skin is oily skin', the customer service robot can determine corresponding answering words again from the dialogue template according to key information such as 'skin' and 'oily' in the 'skin', such as 'recommend product A to you'. Based on the mechanism, the customer service conversation robot finishes interactive question answering with the customer to realize the customer service answering function.

However, prior art customer service dialog templates are typically mined by the merchant itself. At present, no related scheme in the field of intelligent customer service can assist a merchant in extracting and mining a customer service conversation template required by an intelligent customer service robot, so that the merchant needs a large amount of time cost and labor cost to create the customer service conversation template when using related functions of the customer service robot, the workload is high, and the template generation cost is high.

With respect to the scheme shown in fig. 1, the embodiment of the present application provides a method and an apparatus for generating a customer service dialog template, where the scheme includes determining dialog statements of multiple rounds of questions and answers to be processed; carrying out entity recognition processing on the dialogue sentences of each round of question and answer to obtain question and answer categories of each round of question and answer and interaction attributes corresponding to each round of question and answer; processing the question and answer categories of each round of question and answer and the interaction attributes corresponding to each round of question and answer by using a correlation analysis method, and determining one or more frequent interaction attributes corresponding to each question and answer category; and generating a customer service conversation template according to the one or more frequent interaction attributes corresponding to each question and answer class. According to the scheme, entity recognition processing and association analysis processing are carried out on the conversation sentences of multiple rounds of questions and answers, so that entity mining is carried out on the conversation sentences of each round of questions and answers, one or more frequent interaction attributes corresponding to each round of questions and answers are determined, a customer service conversation template can be generated based on the frequent interaction attributes, a merchant can utilize the customer service conversation template to realize rapid configuration of a customer service robot, and the cost required by the merchant for configuring the customer service robot can be effectively reduced.

Referring to fig. 2, fig. 2 is a schematic diagram of a network architecture on which the present application is based, and the network architecture shown in fig. 2 may specifically include a server 21, a merchant terminal 22, and a customer service terminal 23.

In the framework shown in fig. 2, the server 21 may specifically be a server cluster arranged in the cloud, where the server 21 is configured to perform corresponding processing on a plurality of rounds of question-answering conversation sentences according to the method for generating a customer service conversation template provided in the present application, so as to obtain the customer service conversation template.

The merchant terminal 22 and the customer terminal 23 may be hardware devices with network communication functions and user interaction functions, which include but are not limited to smart phones, tablet computers, desktop computers, internet of things devices, and the like. The merchant can obtain the customer service dialogue template generated by the server 21 through the merchant terminal 22, and configure the customer service robot carried in the server 21 by using the customer service dialogue template. The customer can establish a customer service communication channel with the server 21 through the customer terminal 23, and can perform customer service question answering with the configured customer service robot of the merchant through the customer service communication channel.

The following describes in detail a method and an apparatus for generating a customer service interaction template according to the present application with specific embodiments. The following embodiments may be combined with each other and may not be described in detail in some embodiments for the same or similar concepts or processes.

It should be noted that the execution subject of the generation method of the customer service dialog template provided in this embodiment is the server mentioned in fig. 2, and fig. 3 is a flowchart of the generation method of the customer service dialog template provided in this embodiment. As shown in fig. 3, the method for generating the customer service dialog template may include the following steps:

step 301, determining the dialog sentences of the multiple rounds of questions and answers to be processed.

It should be noted that the dialogue sentences of the multiple rounds of questions and answers to be processed in the present embodiment may be specifically offline data stored in the server in advance. The dialogue log is used for recording question and answer sentences generated by customer service and customer service dialogue. Wherein the at least one question dialog sentence and the at least one answer sentence together constitute a round of questions and answers. Meanwhile, in order to ensure the integrity of the generated customer service dialogue template, the questions and answers to be processed in the embodiment of the application are in multiple rounds, that is, the embodiment performs subsequent processing on multiple groups of questions and answers formed by at least one question dialogue sentence and at least one answer sentence.

On the basis of the present embodiment, in order to obtain more rounds of dialog sentences of questions and answers, in an alternative embodiment, a large number of existing dialog logs may be analyzed to determine the dialog sentences of the multiple rounds of questions and answers to be processed in the dialog logs.

Specifically, the server may first obtain a plurality of dialog logs, and each dialog log includes a plurality of dialog statements.

Then, the classification and recognition of the dialogue behaviors can be respectively performed on each pair of the spoken sentences in each dialogue log, so as to obtain the dialogue behaviors corresponding to each dialogue sentence.

The conversation behavior can be specifically divided into a conversation behavior of a client and a conversation behavior of a customer service. Illustratively, the customer service dialogue behavior includes a customer service question-asking behavior and a customer service answer behavior, and the customer service dialogue behavior includes a customer answer behavior and a customer intention behavior.

After each dialog sentence in each dialog log is determined, the server determines the dialog sentences of multiple rounds of questions and answers as the dialog sentences of the multiple rounds of questions and answers to be processed according to the dialog behaviors of each dialog sentence in each dialog log, so that the dialog sentences of each round of questions and answers comprise two adjacent dialog sentences in the same dialog log, and the dialog behaviors of the two adjacent dialog sentences are customer service question-asking dialog behaviors and customer answer dialog behaviors in sequence.

Illustratively, one of the dialog logs may be recorded as: "

Customer: i had bought product a in your house before and felt well.

Customer: can you recommend another product today?

Customer service: ask what kind of skin is you?

Customer the method comprises the following steps: both my cheeks and the T-shaped region are prone to oil.

Customer service: asking you what effect they need this time?

Customer: resisting aging and removing speckle, and recently, there is some wrinkles on face.

Customer service: good relatives recommend you to use product B. "

By performing behavior classification analysis on the dialog sentences, the obtained dialog behaviors of the dialog sentences are as follows in sequence: the system comprises a client intention behavior, a customer service question-asking behavior, a client answering behavior and a customer service answering behavior.

By analyzing each dialogue behavior corresponding to the dialogue log, two rounds of question and answer dialogue sentences can be selected from the dialogue logs, namely customer service: ask what kind of skin is you? Customer: both cheek and tee areas of my are prone to oil. "and" customer service: asking you what effect they need this time? Customer: resisting aging and removing speckle, and recently, there is some wrinkles on face. "

On the basis of the above embodiment, in an alternative embodiment, the classification and recognition of the dialog behavior for each spoken sentence respectively can be implemented based on a pre-trained dialog behavior classification model. Specifically, the server processes each dialogue statement according to a pre-trained dialogue behavior classification model to obtain a dialogue behavior of each dialogue statement; the dialogue behavior of the dialogue sentences belongs to one of a customer answer behavior, a customer intention behavior, a customer service question-asking behavior and a customer service answer behavior.

In particular, the dialog behavior classification model described above may specifically employ a BERT model, which is implemented based on a bidirectional encoder representation technique of a transformer, and may be used for text encoding and classification. In order to effectively classify the dialogue behavior of the dialogue sentences by using the model, optionally, an unsupervised training mode based on the dialogue rules can be used for training the model. During training, conversation rules corresponding to various types of conversation behaviors can be determined according to various pairs of speaking sentences in each conversation log; then, according to each dialogue rule, carrying out sample extraction processing on each dialogue sentence in each dialogue log to obtain a positive and negative sample set corresponding to each type of dialogue behavior; and then, training a pre-constructed dialogue behavior classification model by utilizing each positive and negative sample set to obtain a trained dialogue behavior classification model, wherein the trained dialogue behavior classification model is used for performing behavior classification and recognition processing on dialogue sentences to be processed to obtain dialogue behaviors corresponding to each dialogue sentence.

Illustratively, the dialog rules include: a scene class rule, and/or a regular expression class rule, and/or a negative example keyword class rule. The scene type rule can represent the type of an intention scene in which the dialog behavior appears, and the scene type rule corresponding to the dialog behavior of the wishful graph can comprise a consultation scene; regular expression class rules may represent regular expressions used by this type of dialog behavior to characterize keyword patterns or key sentence patterns, such as a regular expression class rule for question dialog behavior may include "how do your … is …? "; the negative example keyword class rule may indicate a keyword or a keyword sentence that is not used by the dialog behavior, for example, the negative example keyword class rule corresponding to the question dialog behavior includes: "if you have other questions".

Of course, on the basis of the above manner, in an optional embodiment, in the process of training the model, the positive and negative sample sets corresponding to each type of dialogue behavior may also be verified; and optimizing the dialogue rules corresponding to the dialogue behaviors of various types according to the verification result, and determining an optimized positive and negative sample set according to the optimized dialogue rules, wherein the optimized positive and negative sample set is used for training the dialogue behavior classification model.

Step 302, performing entity identification processing on the dialogue sentences of each turn of question and answer to obtain question and answer categories of each turn of question and answer and interaction attributes included in each turn of question and answer, wherein the interaction attributes include attribute types and attribute values.

In the present embodiment, after the dialogue acts of each dialogue sentence are determined and the question and answer turns are determined based on the dialogue acts, and performing entity recognition processing on each round of question and answer to determine the question and answer category and the corresponding interaction attribute for each round of question and answer.

Step 302 may include step 3021 and step 3022.

Step 3021, performing entity extraction on the dialog sentences of each question and answer turn and aligning the extracted entity results to obtain entity information corresponding to each dialog sentence, where the entity information includes multiple entity types and entity values corresponding to each entity type.

Step 3022, according to the type of the entity type, taking the entity value corresponding to the category as the question-answer category of each round of question-answer, and taking the attribute and the entity value corresponding to the attribute as the interaction attribute corresponding to each round of question-answer.

For any round of question answering, step 3021 may specifically include: firstly, a preset named entity recognition model is utilized to perform entity extraction processing on each dialogue statement in the round of question answering, and an entity type, an entity logic and an entity value of each dialogue statement are obtained. Then, according to the entity logic of each dialogue statement, carrying out alignment judgment on the logic relationship between the entity type and the entity value of each dialogue statement, and selecting the entity type with the preset logic relationship and the corresponding entity value for each dialogue statement according to the judgment result; and finally, judging whether each dialogue statement of the round of question answering comprises the same or similar entity type, wherein the same or similar entity type and the corresponding entity value form entity information of the round of question answering.

Named Entity Recognition (NER), also called "proper name Recognition", refers to a technique that can be used to recognize entities having a specific meaning in a text. Generally, entities may include names of people, places, organizations, proper nouns, and so on.

The embodiment realizes the entity extraction processing of the dialogue sentences by utilizing a pre-trained named entity recognition model. It can be known that, since the application is faced with the processing of related information in the field of intelligent customer service, and there are many industries involved in the field, such as the cosmetic industry, the digital industry, and the like, the named entity identification model needs to have the expansibility from one industry to another.

Based on this, the named entity recognition model can be Based on a Domain & Profile Based NER (DPNer) model proposed by a migratable dialog state generator (TRADE), which enables the model parameters to be shared between different domains by introducing Domain knowledge in the model for context enhancement, thereby expanding migratable industry knowledge to other industries. That is, because the model has strong compatibility with industry knowledge, when the named entity recognition model is used for processing dialogue sentences of dialogue logs of different industries, the model can be realized without further training.

Based on such a model structure, in an optional implementation manner, a preset named entity recognition model may be further used to sequentially perform vector coding processing, logic extraction processing, and entity extraction processing on each dialogue statement in each round of question and answer statements. Specifically, fig. 4 is a schematic diagram of a named entity recognition model provided in this embodiment, and in order to implement the vector encoding process, the logic extraction process, and the entity extraction process, as shown in fig. 4, the named entity recognition model may specifically include: a text encoder (query encoder) 401, a slot gate (slot gate) 402, and a slot value generator (slot value decoder) 403.

The text Encoder (query Encoder) 401 adopts a trained Albert Encoder structure, and the structure can encode the sentence text of the dialog sentence into a sequence of fixed-length vectors.

Slot gate 402 is a multi-classification block that may be used to identify the logic of the extraction entity, and includes a linear layer. By this linear layer, a slot gate (slot gate) 402 maps a sequence of fixed-length vectors obtained by encoding a dialog by a text encoder into a distribution of one ' none ', ' = ', ' <', ' < ' = ', ' > = '. Wherein ' none ' is used to indicate the physical logic of "negative", "' = ' is used to indicate the physical logic of" positive "," < "> is used to indicate the physical logic of" less than "," < = ' is used to indicate the physical logic of "not more than/less than or equal to", "greater than", and ' > = ' is used to indicate the physical logic of "not less than/greater than or equal to".

A slot value generator (slot value decoder) 403 is a pointer generator that employs a soft-gated structure that generates a single output vocabulary distribution by combining two distributions of text and vocabulary. The implicit state of each word vector is determined when the coded vector is decoded, and an attention mechanism is combined to predict the distribution of the codes. And finally obtaining the entity type and the entity value corresponding to the entity type.

Illustratively, if a question and answer turn includes a dialog sentence "customer service asking about 200 pieces of money for a facial cleanser product, do you accept? The expected delivery time is about 2 months and 3 days. Customer: my skin is not greasy, and a facial cleanser product at this price may be as good as express.

Two dialog statements are processed in the above-described manner to obtain the following results:

for the first dialog statement, the following are available: [ price, '=',200 blocks ], [ product, '=', facial cleanser ], [ delivery time, '=',2 months, 3 days ]; for the second dialog statement, the following are available: [ skin type, 'none', oil skin ] [ price, '=', price ], [ product, '=', facial cleanser ], [ express aging, '=', appropriate ].

After the entity extraction processing result of each dialog statement is obtained, the alignment processing is performed as described in the foregoing step 3021. Specifically, the alignment process is a process of determining which entity types and entity values belong to the necessary entity types and which entity types and entity values are redundant or erroneous according to the entity information identified from the dialog statement. In the embodiment of the application, entity types and entity values of 'questioning and answering' involved in the question and answer process need to be aligned and retained, and entity types and entity values of 'questioning and answering' or 'answering without question' are filtered. At the same time, entity types and corresponding entity values for some entities that do not satisfy the preset entity logic will also be filtered.

Therefore, for any round of question and answer, the alignment judgment can be performed on the logical relationship between the entity type and the entity value of each dialogue statement according to the entity logic of each dialogue statement, so that the entity type with the preset logical relationship and the corresponding entity value can be selected for each dialogue statement according to the judgment result. For example, the ' = ' in this embodiment is a preset logical relationship, but other entity logics are not preset logical relationships, and based on this, the [ skin, skin ', oil skin ] corresponding to each second pair of sentences will be filtered out.

And finally, judging whether each pair of spoken sentences of the round of question answering comprises the same or similar entity type, wherein the same or similar entity type and the corresponding entity value form entity information of the round of question answering. Since the question-answer includes both a question dialog sentence and an answer dialog sentence, this step will retain similar or identical entity types and corresponding entity values in both dialog sentences.

Illustratively, the [ price, '=',200 blocks ] of the first conversational sentence and the [ price, '=', price ] of the second conversational sentence are the same entity type (both are the entity type of "price"), and then the [ price, (price, 200 blocks) ] can be obtained as the entity information.

Illustratively, the [ delivery time, '=',2 month 3 day ] of the first dialogue sentence and the [ express delivery, '=', appropriate ] of the second dialogue sentence are similar entity types, and then [ (delivery time, express delivery), (2 month 3 day, appropriate) ] can be obtained as the entity information. The similar entity types can be specifically judged through the mapping relation among the entity types, for example, the delivery time and the express delivery have the mapping relation, and for example, the skin color and the color have the mapping relation. It is to be understood that this mapping relationship is constructed in advance, and the present embodiment is not limited thereto.

Through the entity extraction processing and the alignment processing, entity information corresponding to each dialogue statement can be obtained. Taking the above example as an example, the obtained entity information of the question and answer is: [ price, (price, 200 pieces) ], [ product, facial cleanser ], [ (time to delivery, express), (2 months 3 days, right) ].

In step 3022, the question and answer classes and the interaction attributes of the round of question and answer are also determined according to the entity information obtained in step 3021, where the interaction attributes include attribute types and attribute values.

For the entity type, the entity type may specifically include an item type and an attribute type, where the item type represents the goods or products surrounded by the turn of question and answer, and the attribute type represents the information surrounded by the turn of question and answer and related to the goods attribute or other related attributes.

Taking the foregoing example as an example, the "product" corresponds to a category type, and the corresponding "facial cleanser" can be used as a question-answer category corresponding to the question-answer. The "price" and "(delivery time, express)" can be used as the attribute type corresponding to the question and answer, and the corresponding "(price, 200 pieces)" and "(2 months, 3 days, proper)" are the attribute values.

To this end, the question and answer class of each round of question and the interaction attribute included in each round of question and answer are obtained, and [ question and answer class-interaction attribute ] in the above example can be expressed as [ facial cleanser, ([ price (price, 200 pieces) ], [ (time to delivery, express), (2 months and 3 days, appropriate) ]) ].

Step 303, processing the question-answer classes of each round of question-answer and the interaction attributes corresponding to each round of question-answer by using a correlation analysis method, and determining one or more frequent interaction attributes corresponding to each question-answer class.

In step 303, the question-answer classes of each round of answers and the interaction attributes corresponding to each round of answers are clustered to obtain an interaction attribute set corresponding to each question-answer class, where the interaction attribute set includes all the interaction attributes corresponding to each question-answer class.

Different rounds of questions and answers of different conversation logs are generally corresponding to different question and answer classes, so that the generation of the customer service conversation template is more comprehensive and accurate. In this embodiment, each interactive attribute is clustered based on the question and answer categories to form an interactive attribute set corresponding to each question and answer category, such as an interactive attribute set corresponding to a facial cleanser, an interactive attribute set corresponding to a shampoo, and the like.

In addition, the interaction attribute set should include the interaction attribute of the question-answer class. For example, the following interaction attributes exist for the question-answer class of facial cleanser: [ facial cleanser, [ skin (oil skin) ] ] and [ facial cleanser, ([ skin (dry skin) ], [ price, (100 pieces) ]) ], in which case the set of attributes of the facial cleanser is expressed as: ([ skin type, (dry skin, oily skin) ], [ price, (100 pieces) ]).

After the interaction attribute sets of the question and answer classes are obtained, all interaction attributes in each interaction attribute set are subjected to correlation analysis processing, and one or more frequent interaction attributes corresponding to each question and answer class are determined.

In order to adapt the customer service dialog template to a sufficient number of customer service configuration strategies, in the embodiment, the most frequent interaction attribute combination in the question-answer class is mined from all the combinations by randomly combining the interaction attributes.

In one of the alternative embodiments, the association analysis process may be implemented based on the classical frequent item set mining algorithm Apriori algorithmm.

In another alternative embodiment, the association analysis process may incorporate the number of combinations as additional variables based on the classical frequent item set mining algorithm Apriorialgorithm, so as to achieve the purpose of finding a set of attribute combinations that are as complete and high-frequency as possible at the same time.

Specifically, for M interaction attributes in any interaction attribute set, the interaction attribute combination mode of the M interaction attributes is traversed to obtain X types of interaction attributesA sexual combination; wherein the content of the first and second substances,

That is, taking the total number M =3 of the interaction attributes in the interaction attribute set as an example, when the number N of the interaction attributes is 1, 2, and 3, respectively, X kinds of interaction attribute combinations can be constructed, respectively. Then, the support degree of each interactive attribute combination with N being 1, the support degree of each interactive attribute combination with N being 2, and the support degree of each interactive attribute combination with N being 3 are calculated by a support degree algorithm. And finally, calculating the increment function value of each interactive attribute combination based on the N value and the support degree of each interactive attribute combination. Generally, at least one interactive attribute combination with a higher increment function value will be the processing result, and the interactive attribute in the interactive attribute combination with the higher increment function value will be the frequent interactive attribute of the question-answer class.

Illustratively, the frequent interaction attributes of a facial cleanser obtained by screening can be expressed as ([ skin type, (dry skin, oil skin) ]).

And step 304, generating a customer service dialogue template according to one or more frequent interaction attributes corresponding to each question and answer class.

The customer service interaction template can comprise a customer service question template and a customer service answer template, the customer service robot can realize the interaction function of asking questions to the customer based on the customer service question template, and the customer service robot can realize the interaction function of responding and answering questions asked by the customer service based on the customer service answer template.

Fig. 5 is a scene schematic diagram of configuring intelligent customer service based on a customer service conversation template according to an embodiment of the present application, and as shown in fig. 5, a merchant may select a customer service conversation template of a question and answer class to be configured on a current interface, where the selected customer service conversation templates are obtained by the method for generating a customer service conversation template. And after the customer service dialogue template is selected, the customer service dialogue template is loaded to a configuration page to be output to the intelligent customer service robot for use. Of course, in a possible implementation, the customer service dialog template may be presented in the format of a mind map for easy viewing by the merchant.

As shown in the left frame of the lower diagram of fig. 5, the frequent interaction attributes of the article a include three attributes, which are: skin (oil skin, dry skin, mixed skin, sensitive skin) ], [ volume (30 ML, 50ML, 100ML, 200 ML) ], [ after-market (7 days unprovoked, after-market risk, claims to break) ]. Each frequent interaction attribute comprises an attribute type and a corresponding attribute value, wherein the skin type, the capacity and the after-sale are respectively the attribute types. The 'oily skin, dry skin, mixed skin and sensitive skin' are attribute values corresponding to 'skin type'; "30ML, 50ML, 100ML, 200ML" are attribute values corresponding to "capacity"; the 7-day unproblematic, after-sales insurance and damage payment are attribute values corresponding to the after-sales.

And as shown in the right side box of the lower diagram of fig. 5, through the corresponding relationship between the attribute types and the attribute values in the frequent interaction attributes, a corresponding thinking diagram can be established for the merchant to view, and at this time, the merchant inputs a corresponding dialect according to the interaction logic in the thinking diagram.

And attribute types such as "skin type", "capacity" and "after sale" can be used to configure the customer service questioning template, for example, a merchant can configure "what is your skin type" for the attribute type "skin type? "questioning. The attribute value corresponding to each attribute type can be used for configuring a customer service answer template, for example, a merchant can configure an answer saying that the attribute value of 'oil skin' suggests you select the item of merchandise.

That is, considering that the frequent interaction attributes include attribute types and attribute values, the customer service interaction template of each question and answer class can be determined according to the frequent interaction attributes corresponding to each question and answer class when the customer service interaction template is generated; namely, the attribute type in the frequent interaction attribute can be used for configuring the customer service questioning template, and the attribute value in the frequent interaction attribute can be used for configuring the customer service answering template.

The embodiment of the application provides a method for generating a customer service conversation template, wherein entity recognition processing and association analysis processing are carried out on a plurality of rounds of question-answer conversation sentences, so that entity mining is carried out on the conversation sentences of each round of question-answer, one or more frequent interaction attributes corresponding to each round of question-answer are determined, the customer service conversation template can be generated based on the frequent interaction attributes, a merchant can utilize the customer service conversation template to realize rapid configuration of a customer service robot, and the cost required by the merchant for configuring the customer service robot can be effectively reduced.

Fig. 6 is a schematic diagram of a hardware structure of an electronic device provided in the present application, and as shown in fig. 6, an embodiment of the present application provides an electronic device, where a memory of the electronic device may be used to store at least one program instruction, and a processor is used to execute the at least one program instruction, so as to implement the technical solution of the foregoing method embodiment. The implementation principle and technical effect are similar to those of the embodiments related to the method, and are not described herein again.

The embodiment of the application provides a chip. The chip comprises a processor for calling a computer program in a memory to execute the technical solution in the above embodiments. The principle and technical effects are similar to those of the related embodiments, and are not described herein again.

The embodiment of the present application provides a computer program product, which, when the computer program product runs on an electronic device, enables the electronic device to execute the technical solutions in the above embodiments. The principle and technical effects are similar to those of the related embodiments, and are not described herein again.

The embodiment of the present application provides a computer-readable storage medium, on which program instructions are stored, and when the program instructions are executed by an electronic device, the electronic device is enabled to execute the technical solutions of the above embodiments. The principle and technical effects are similar to those of the related embodiments, and are not described herein again.

The above embodiments are provided to explain the purpose, technical solutions and advantages of the present application in further detail, and it should be understood that the above embodiments are merely illustrative of the present application and are not intended to limit the scope of the present application, and any modifications, equivalent substitutions, improvements and the like made on the basis of the technical solutions of the present application should be included in the scope of the present application.

Claims

1. A method for generating a customer service dialogue template is characterized by comprising the following steps:

determining a plurality of rounds of question-answering dialogue sentences to be processed;

carrying out entity recognition processing on the dialogue sentences of each round of question and answer to obtain question and answer categories of each round of question and answer and interaction attributes corresponding to each round of question and answer;

processing the question and answer categories of each round of question and answer and the interaction attributes corresponding to each round of question and answer by using a correlation analysis method, and determining one or more frequent interaction attributes corresponding to each question and answer category;

and generating a customer service dialogue template according to the one or more frequent interaction attributes corresponding to each question and answer class.

2. The generation method according to claim 1, wherein the processing the question-answer class of each round of question-answers and the interaction attribute corresponding to each round of question-answers by using the association analysis method includes:

clustering the question-answer classes of each round of question-answers and the interaction attributes corresponding to each round of question-answers to obtain an interaction attribute set corresponding to each question-answer class, wherein the interaction attribute set comprises all interaction attributes corresponding to each question-answer class;

and performing correlation analysis processing on all the interactive attributes in each interactive attribute set, and determining one or more frequent interactive attributes corresponding to each question and answer class.

3. The method according to claim 2, wherein the determining one or more frequent interaction attributes corresponding to each question and answer class comprises:

traversing interaction attribute combination modes of M interaction attributes in any interaction attribute set to obtain X interaction attribute combinations; wherein the content of the first and second substances,

n is the number of the interactive attributes in the interactive attribute combination, and is a positive integer greater than or equal to 1;

determining the support degree of each interaction attribute combination in the X interaction attribute combinations in the interaction attribute set;

determining an increment function value corresponding to each interactive attribute combination according to the value of N corresponding to each interactive attribute combination and the support degree of each interactive attribute combination;

and determining one or more frequent interaction attributes of the question and answer classes corresponding to the interaction attribute set according to the increment function values corresponding to the interaction attribute combinations.

4. The generation method according to claim 1, wherein the determining of the dialog sentences of the multiple rounds of question answering to be processed comprises:

acquiring a plurality of dialogue logs, wherein each dialogue log comprises a plurality of dialogue sentences;

classifying and identifying the dialogue behaviors of each dialogue statement in each dialogue log respectively to obtain the dialogue behavior corresponding to each dialogue statement;

and determining the dialog sentences of the multiple rounds of questions and answers to be processed according to the dialog behaviors of each dialog sentence.

5. The generation method of claim 4, wherein the classifying and identifying the dialogue acts for each dialogue statement in each dialogue log respectively to obtain the dialogue act corresponding to each dialogue statement comprises:

processing each dialogue statement according to a pre-trained dialogue behavior classification model to obtain a dialogue behavior of each dialogue statement;

wherein the dialogue action of the dialogue statement belongs to one of a customer answering action, a customer intention action, a customer service questioning action and a customer service answering action.

6. The generation method according to claim 1, wherein the performing entity recognition processing on the dialogue sentences in each round of question and answer to obtain question and answer categories of each round of question and answer and interaction attributes corresponding to each round of question and answer includes:

performing entity extraction processing on the dialogue sentences of each round of question answering and aligning the extracted entity results to obtain entity information corresponding to each dialogue sentence, wherein the entity information comprises a plurality of entity types and entity values corresponding to each entity type;

and according to the type of the entity type, taking the entity value corresponding to the category as the question and answer category of each round of question and answer, and taking the attribute and the entity value corresponding to the attribute as the interaction attribute corresponding to each round of question and answer.

7. The generation method according to claim 6, wherein the performing entity extraction processing on the dialogue sentences of each round of question answering and performing alignment processing on the extracted entity results to obtain entity information corresponding to each dialogue sentence comprises:

aiming at any round of question and answer, performing entity extraction processing on each dialogue statement in the round of question and answer by using a preset named entity recognition model to obtain an entity type, an entity logic and an entity value of each dialogue statement;

according to the entity logic of each dialogue statement, carrying out alignment judgment on the logic relationship between the entity type and the entity value of each dialogue statement, and selecting the entity type with the preset logic relationship and the corresponding entity value for each dialogue statement according to the judgment result;

and judging whether each dialogue statement of the round of question answering comprises the same or similar entity type, wherein the same or similar entity type and the corresponding entity value form entity information of the round of question answering.

8. The method for generating customer service dialog templates according to claim 7, wherein the performing entity extraction processing on each dialog statement in the turn of question answering by using a preset named entity recognition model comprises:

and aiming at each dialogue statement in each round of question-answering statements, sequentially carrying out vector coding processing, logic extraction processing and entity extraction processing on the dialogue statement by using a preset named entity recognition model.

9. The method for generating customer service dialog templates of any of claims 1-8, wherein the frequent interaction attributes comprise attribute types and attribute values, and the customer service dialog templates comprise customer service question templates and customer service answer templates;

generating a customer service dialogue template according to the one or more frequent interaction attributes corresponding to each question and answer class, wherein the generation comprises the following steps:

configuring the customer service questioning template according to the attribute type of one or more frequent interaction attributes corresponding to each question and answer class;

and configuring the customer service answer template according to the attribute values of one or more frequent interaction attributes corresponding to each question and answer class.

10. An electronic device, comprising:

at least one processor; and

a memory;

the memory stores computer-executable instructions;

the at least one processor executing the computer-executable instructions stored by the memory causes the at least one processor to perform the method of claims 1-9.

11. A computer-readable storage medium having computer-executable instructions stored thereon which, when executed by a processor, implement the method of claims 1-9.

12. A computer program product comprising computer instructions, characterized in that the computer instructions, when executed by a processor, implement the method according to claims 1-9.