CN114428845A

CN114428845A - Intelligent customer service automatic response method and device, equipment, medium and product thereof

Info

Publication number: CN114428845A
Application number: CN202210105938.4A
Authority: CN
Inventors: 许强
Original assignee: Guangzhou Huaduo Network Technology Co Ltd
Current assignee: Guangzhou Huaduo Network Technology Co Ltd
Priority date: 2022-01-28
Filing date: 2022-01-28
Publication date: 2022-05-03

Abstract

The application discloses an intelligent customer service automatic response method and a device, equipment, medium and product thereof, wherein the method comprises the following steps: obtaining a question text submitted by a chat interface of a customer service system of an electricity merchant; acquiring a preset question with the most similar semanteme to the question text from a knowledge base of the e-commerce customer service system as a target question, and determining an answer set mapped by a question set to which the target question belongs as a target answer set; marking standard questions in the question set and preset answers in the answer set by using flow labels in a label pool, wherein each flow label in the label pool represents each business link in the e-commerce ordering flow correspondingly; determining classification probability of the question text mapped to each process label by adopting a classification model, and determining a preset answer with the classification probability of the process label being a relative maximum value from the target answer set as a target answer; and pushing the target answer to the chat interface for display. The application can improve the response accuracy of the e-commerce customer service system.

Description

Intelligent customer service automatic response method and device, equipment, medium and product thereof

Technical Field

The present application relates to the field of intelligent customer service technologies, and in particular, to an intelligent customer service automatic response method, and a corresponding apparatus, computer device, computer-readable storage medium, and computer program product.

Background

At present, in an e-commerce scene, because customers have more consultation problems, general merchants can configure corresponding intelligent customer service robots to assist customer service in answering problems, most intelligent customer service methods in general markets are to establish a set of knowledge base, wherein the knowledge base comprises three parts: standard questions, similar questions, answers. The answer to the reply is generally unique for a particular scenario. Aiming at the unicity of the answer, the improved method generally carries out a plurality of configurations on the answer under the standard question, when the corresponding standard question is triggered, one answer is randomly selected and returned to the client, so that different clients receive different answers, and the intelligent customer service is humanized.

In reality, the customers are grouped according to the service scene, but the answer responses of different grouped customers are not ideal due to the randomness of the answers of the method, so the applicant tries to explore a new idea in order to improve the accurate matching degree between the answers and the questions.

Disclosure of Invention

A primary object of the present application is to solve at least one of the above problems and provide an intelligent customer service automatic response method, and a corresponding apparatus, computer device, computer readable storage medium, and computer program product.

In order to meet various purposes of the application, the following technical scheme is adopted in the application:

the intelligent customer service automatic response method adaptive to one of the purposes of the application comprises the following steps:

obtaining a question text submitted by a chat interface of a customer service system of an electric business;

acquiring preset questions with the most similar semanteme to the question text from the total preset questions in a knowledge base of the e-commerce customer service system as target questions, and determining an answer set mapped by a question set to which the target questions belong as a target answer set; the knowledge base comprises mapping relation data between a question set and an answer set, wherein each question set comprises a plurality of preset questions, and each preset question comprises a standard question and a plurality of similar questions; each answer set comprises a plurality of preset answers, the standard questions and the preset answers are labeled by process labels in a label pool, and each process label in the label pool represents each business link in the e-commerce ordering process correspondingly;

determining a classification probability corresponding to each process label mapped to the label pool by using a classification model trained to a convergence state, and determining a preset answer as a target answer, wherein the classification probability of the labeled process label is a relative maximum value in a range of a target answer set;

and pushing the target answer to a chat interface of the e-commerce customer service system for display.

In a further embodiment, a preset question with the most similar semantic meaning to the question text is obtained from the total preset questions in the knowledge base of the e-commerce customer service system as a target question, and an answer set mapped by a question set to which the target question belongs is determined as a target answer set, including the following steps:

extracting a sentence vector of the question text by adopting a text feature extraction model trained to a convergence state;

calculating similarity data between the sentence vectors of the question text and the sentence vectors of the total preset questions in the knowledge base, which are extracted in advance by the text feature extraction model;

screening out a preset problem of which the similarity data is higher than a preset threshold and is the maximum value, and determining the preset problem as a target problem;

and acquiring an answer set mapped corresponding to the standard questions of the question set to which the target question belongs from a knowledge base as a target answer set.

In a further embodiment, a classification model trained to a convergence state is used to determine a classification probability corresponding to each process label mapped to the label pool, and a preset answer with the classification probability of the labeled process label being a relative maximum value in a range of a target answer set is determined from the target answer set as a target answer, including the following steps:

extracting a sentence vector of the question text by adopting a text feature extraction model in a classification model;

performing classification mapping according to the sentence vector of the question text by adopting a classifier preset in a classification model to obtain the classification probability corresponding to each flow label mapped to the label pool by the question text;

determining one or more classification probabilities corresponding to the preset answers according to one or more process labels carried by the preset answers in the target answer set;

and comparing the classification probability corresponding to each preset answer in the target answer set, determining the maximum classification probability, and taking the preset answer with the maximum classification probability as the target answer.

In a further embodiment, the training process of the classification model is implemented in advance, and includes the following steps:

selecting a preset problem from the problem set in the knowledge base as a training sample, and inputting the preset problem into a text feature extraction model of a classification model to extract a sentence vector;

classifying and mapping the sentence vector through a classifier to obtain the classification probability corresponding to each flow label mapped to the label pool by the sentence vector, and determining the target flow label corresponding to the maximum classification probability;

calculating a loss value of the target process label by taking the process label carried by the standard problem in the selected problem set as a supervision label, and terminating the training if the loss value reaches a preset threshold value and reaches a convergence state; otherwise, gradient updating is implemented, and iterative training is implemented on the classification model by adopting the next training sample.

In a preferred embodiment, after the training process of the classification model is completed, the method includes the following steps:

and respectively extracting a sentence vector for each preset problem in the knowledge base by adopting a text feature extraction model in the classification model trained to be in a convergence state, and associating and storing the sentence vector with the preset problem in the knowledge base.

In a specific embodiment, the e-commerce ordering flow comprises a plurality of different stages, each stage comprises one or more business links, a flow label is arranged corresponding to each business link, and a set of flow labels corresponding to all the business links forms the label pool, wherein the different stages comprise a collection stage, a shopping cart stage, a payment stage, a delivery stage and an after-sale stage.

An intelligent customer service automatic answering device adapted to one of the purposes of the application comprises: the system comprises a question response module, a question hit module, an answer hit module and a question response module, wherein the question response module is used for acquiring a question text submitted by a chat interface of the customer service system of the e-commerce; the question hit module is used for acquiring preset questions with the most similar semanteme to the question text from the total preset questions in the knowledge base of the e-commerce customer service system as target questions, and determining an answer set mapped by a question set to which the target questions belong as a target answer set; the knowledge base comprises mapping relation data between a question set and an answer set, wherein each question set comprises a plurality of preset questions, and each preset question comprises a standard question and a plurality of similar questions; each answer set comprises a plurality of preset answers, the standard questions and the preset answers are marked by flow labels in a label pool, and each flow label in the label pool represents each service link in the e-commerce ordering flow correspondingly; the answer hit module is used for determining the classification probability corresponding to each process label mapped to the label pool by adopting a classification model trained to be in a convergence state, and determining a preset answer as a target answer, wherein the classification probability of the labeled process label is a relative maximum value in a target answer set range; and the question answering module is used for pushing the target answer to the chat interface of the e-commerce customer service system for display.

In a further embodiment, the question naming module includes: the vector extraction submodule is used for extracting a sentence vector of the questioning text by adopting a text feature extraction model trained to be in a convergence state; the similarity calculation submodule is used for calculating similarity data between the sentence vectors of the question text and the sentence vectors of the full preset questions in the knowledge base, which are extracted in advance by the text feature extraction model; the target screening submodule is used for screening out a preset problem of which the similarity data is higher than a preset threshold and is the maximum value, and determining the preset problem as a target problem; and the answer selection set submodule is used for acquiring an answer set mapped corresponding to the standard questions of the question set to which the target questions belong from the knowledge base as the target answer set.

In a further embodiment, the answer hit module comprises: the vector extraction submodule is used for extracting the sentence vector of the question text by adopting a text feature extraction model in the classification model; the classification mapping submodule is used for performing classification mapping according to the sentence vector of the question text by adopting a classifier preset in a classification model to obtain the classification probability corresponding to each flow label mapped to the label pool by the question text; the corresponding conversion sub-module is used for determining one or more classification probabilities corresponding to the preset answers according to one or more process labels carried by the preset answers in the target answer set;

and the comparison and optimization module is used for comparing the classification probability corresponding to each preset answer in the target answer set, determining the maximum classification probability in the classification probabilities, and taking the preset answer with the maximum classification probability as the target answer.

In a further embodiment, the intelligent customer service automatic answering device further includes a structure for executing a training process of the classification model, and the structure includes: the sample selection submodule is used for selecting a preset problem from the problem set in the knowledge base as a training sample, inputting the preset problem into a text feature extraction model of a classification model and extracting a sentence vector; the mapping prediction submodule is used for carrying out classification mapping on the sentence vector through a classifier to obtain the classification probability corresponding to each flow label mapped to the label pool by the sentence vector and determining the target flow label corresponding to the maximum classification probability; an iteration decision sub-module, configured to calculate a loss value of the target flow label by using the flow label carried by the standard problem in the selected problem set as a supervision label, and terminate training if the loss value reaches a preset threshold value and reaches a convergence state; otherwise, gradient updating is implemented, and iterative training is implemented on the classification model by adopting the next training sample.

In a preferred embodiment, the intelligent customer service automatic answering device further includes: and the vector preprocessing submodule is used for extracting a sentence vector for each preset problem in the knowledge base by adopting the text feature extraction model in the classification model trained to be in the convergence state, and storing the sentence vector in the knowledge base in association with the preset problem.

The computer device comprises a central processing unit and a memory, wherein the central processing unit is used for calling and running a computer program stored in the memory to execute the steps of the intelligent customer service automatic response method.

A computer-readable storage medium, which stores a computer program implemented according to the intelligent customer service automatic response method in the form of computer-readable instructions, and when the computer program is called by a computer, executes the steps included in the method.

A computer program product, provided to adapt to another object of the present application, comprises computer programs/instructions which, when executed by a processor, implement the steps of the method described in any of the embodiments of the present application.

Compared with the prior art, the application has the following advantages:

firstly, in order to improve the response accuracy of an e-commerce customer service system, the method comprises the steps of labeling a flow label of a corresponding e-commerce order flow business link for a standard question set and a preset answer in a knowledge base in advance according to an e-commerce order flow, matching a preset question with the closest semantic meaning as a target question when a user submits a question text, determining a target answer set corresponding to the standard question associated with the target question, further obtaining a classification probability corresponding to each flow label mapped to a label pool by using a classification model, determining a preset answer carrying the flow label with the highest classification probability as a target answer from all preset answers of the target answer set according to the classification probability, and pushing the target answer to the user for display. The preset answers are labeled according to the process labels corresponding to the business links of the e-commerce ordering process, the questioning text is mapped to the classification probability corresponding to each process label through the classification model to play a role in sequencing reference, therefore, a plurality of preset answers in the target answer set are optimized, the determined target answers correspond to the questioning text in meaning, and the target answers are labeled through the process labels in the e-commerce ordering process in advance and are the maximum classification probability in the process labels of all the preset answers in the target answer set, so that the correlation in the business links can be established between the target answers and the questioning text based on the e-commerce ordering process, the target answers can be matched with the questioning text, and the user experience of the e-commerce customer service system is improved.

Secondly, the mapping relation between the questioning text and the process labels is concerned at the semantic level, and then the classification probability of each process label in a label pool corresponding to the questioning text is used as a sequencing basis of a preset answer in a target answer set, so that the determined target answer is necessarily more accurate and fine, the accuracy means that the correspondence between the questioning text and the target answer on meaning expression is realized, and the fineness means that the correspondence of the meaning expression is specifically to the depth of different business links of the e-commerce ordering process, and therefore the intelligent degree of the e-commerce customer service system is comprehensively improved.

In addition, by implementing the technical scheme, a large amount of labor work can be avoided in large-scale customer service scenes such as e-commerce platforms, corresponding implementation cost is saved, and large-scale economic utility is obtained.

Drawings

The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

FIG. 1 is a schematic flow chart diagram of an exemplary embodiment of an intelligent customer service automatic response method of the present application;

FIG. 2 is a flowchart illustrating a process of matching a target answer set according to a quiz text in an embodiment of the present application;

fig. 3 is a schematic flowchart illustrating a process of determining a classification probability of each flow label according to a question text and further determining a target answer from a target answer set according to the classification probability;

FIG. 4 is a flow diagram illustrating a process of training a classification model according to the present application;

FIG. 5 is a functional block diagram of an intelligent customer service auto-answer apparatus according to the present application;

fig. 6 is a schematic structural diagram of a computer device used in the present application.

Detailed Description

Reference will now be made in detail to embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are exemplary only for the purpose of explaining the present application and are not to be construed as limiting the present application.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.

It will be understood by those within the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

As will be appreciated by those skilled in the art, "client," "terminal," and "terminal device" as used herein include both devices that are wireless signal receivers, which are devices having only wireless signal receivers without transmit capability, and devices that are receive and transmit hardware, which have receive and transmit hardware capable of two-way communication over a two-way communication link. Such a device may include: cellular or other communication devices such as personal computers, tablets, etc. having single or multi-line displays or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service), which may combine voice, data processing, facsimile and/or data communication capabilities; a PDA (Personal Digital Assistant) that may include a radio frequency receiver, a pager, internet/intranet access, web browser, notepad, calendar, and/or GPS (Global Positioning System) receiver; a conventional laptop and/or palmtop computer or other device having and/or including a radio frequency receiver. As used herein, a "client," "terminal device" can be portable, transportable, installed in a vehicle (aeronautical, maritime, and/or land-based), or situated and/or configured to operate locally and/or in a distributed fashion at any other location(s) on earth and/or in space. The "client", "terminal Device" used herein may also be a communication terminal, a web terminal, a music/video playing terminal, such as a PDA, an MID (Mobile Internet Device) and/or a Mobile phone with music/video playing function, and may also be a smart tv, a set-top box, and the like.

The hardware referred to by the names "server", "client", "service node", etc. is essentially an electronic device with the performance of a personal computer, and is a hardware device having necessary components disclosed by the von neumann principle such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, an output device, etc., a computer program is stored in the memory, and the central processing unit calls a program stored in an external memory into the internal memory to run, executes instructions in the program, and interacts with the input and output devices, thereby completing a specific function.

It should be noted that the concept of "server" as referred to in this application can be extended to the case of a server cluster. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided, and in physical space, the servers may be independent from each other but can be called through an interface, or may be integrated into one physical computer or a set of computer clusters. Those skilled in the art will appreciate this variation and should not be so limited as to restrict the implementation of the network deployment of the present application.

One or more technical features of the present application, unless expressly specified otherwise, may be deployed to a server for implementation by a client remotely invoking an online service interface provided by a capture server for access, or may be deployed directly and run on the client for access.

Unless specified in clear text, the neural network model referred to or possibly referred to in the application can be deployed in a remote server and used for remote call at a client, and can also be deployed in a client with qualified equipment capability for direct call.

Various data referred to in the present application may be stored in a server remotely or in a local terminal device unless specified in the clear text, as long as the data is suitable for being called by the technical solution of the present application.

The person skilled in the art will know this: although the various methods of the present application are described based on the same concept so as to be common to each other, they may be independently performed unless otherwise specified. In the same way, for each embodiment disclosed in the present application, it is proposed based on the same inventive concept, and therefore, concepts of the same expression and concepts of which expressions are different but are appropriately changed only for convenience should be equally understood.

The embodiments to be disclosed herein can be flexibly constructed by cross-linking related technical features of the embodiments unless the mutual exclusion relationship between the related technical features is stated in the clear text, as long as the combination does not depart from the inventive spirit of the present application and can meet the needs of the prior art or solve the deficiencies of the prior art. Those skilled in the art will appreciate variations therefrom.

The intelligent customer service automatic response method can be programmed into a computer program product, is deployed in a client or a server to run, and is generally deployed in the server to implement, for example, in an e-commerce platform application scenario of the application, so that the method can be executed by accessing an open interface after the computer program product runs and performing human-computer interaction with a process of the computer program product through a graphical user interface.

An exemplary application scenario of the application is an application in an e-commerce platform based on independent stations, each independent station is a business instance of the e-commerce platform, and has an independent access domain name, and an actual owner of the business instance is responsible for issuing and updating commodities.

The merchant instance of each independent station can be configured with an e-commerce customer service system provided by an e-commerce platform to realize introduction of an intelligent customer service robot, the e-commerce customer service system is used for providing consultation service for related consumer users, the consumer users enter a chat interface corresponding to the merchant instance to input questions needing consultation as a question text, after receiving the question text, the e-commerce customer service system of the e-commerce platform performs semantic matching on the question text and a question set in a knowledge base configured for the independent station, matches preset questions most similar to the question text in semantics, determines standard questions in the question set to which the e-commerce customer service system belongs according to the preset questions, then calls a pre-stored answer set mapped with the standard questions, determines one preset answer as a target answer by the related technical means of the application, and outputs the preset answer to the chat interface, therefore, the system can respond to the questions of the consumer user and meet the consultation requirements.

When the customer user is used as a questioning user to chat with the intelligent customer service robot, the questioning user is generally allowed to introduce artificial customer service, when the artificial customer service is accessed, the e-commerce customer service system establishes a conversation channel between the questioning user and the artificial customer service user of the independent station, and the two parties continue to carry out artificial conversation, so that the questioning user inputs a questioning text, the artificial customer service user replies an answer text, and chat data are generated in turn.

The chat records generated based on the chat interface chat comprise a question text which is provided by a question user and an answer text which is manually replied by a manual customer service user or automatically replied by a robot, and the question text and the answer text are both filed with speaker characteristic information and stored in a database and can be used for data mining to expand the knowledge base.

Referring to fig. 1, in an exemplary embodiment of the intelligent customer service automatic response method of the present application, the method includes the following steps:

step S1100, obtaining a question text submitted by a chat interface of the E-commerce customer service system:

when a questioning user presents a question to the intelligent customer service robot in a chat interface after the questioning user enters the e-commerce customer service system, inputting corresponding text, namely questioning text, from an input box of the chat interface, and then determining submission. And in response to the event submitted by the user, the client equipment where the chat interface is located submits the question text to a background server which deploys the e-commerce customer service system, the server can perform generalized formatting treatment on the question text according to the requirement by using a service logic realized by visual programming, and the final question text is obtained by removing redundant spaces, expression elements and the like.

Step S1200, acquiring preset questions with the most similar semanteme to the question text from the total preset questions in the knowledge base of the e-commerce customer service system as target questions, and determining answer sets mapped by the question sets to which the target questions belong as target answer sets; the knowledge base comprises mapping relation data between a question set and an answer set, wherein each question set comprises a plurality of preset questions, and each preset question comprises a standard question and a plurality of similar questions; each answer set comprises a plurality of preset answers, the standard questions and the preset answers are labeled by process labels in a label pool, and each process label in the label pool represents each business link in the e-commerce order flow correspondingly:

the e-commerce customer service system is provided with a knowledge base, and the knowledge base is an information set for collecting corresponding relations between questions and answers. The knowledge base comprises a plurality of question sets and a plurality of answer sets, each question set comprises a plurality of preset questions, one preset question in the same question set is a standard question, the other preset questions are similar questions of the standard question, and the similar questions in the same question set are generally consistent or similar in meaning with the standard questions and only slightly different in expression, tone and the like. And establishing a mapping relation between the standard questions in each question set and one answer set, so that the question set and the answer set form a corresponding mapping relation. Each answer set includes a plurality of preset answers, which are used for answering preset questions in the question set mapped with the answer set.

On the basis of the structure of the knowledge base, the method further introduces a flow label to label the standard question and the preset answer. Therefore, a label pool is prepared, the label pool is essentially a label system planned in advance and comprises a plurality of process labels which are made in a one-to-one correspondence mode according to different business links of the e-commerce ordering process, and each process label represents one business link of the e-commerce ordering process. The e-commerce order flow can be divided into a plurality of different stages according to the processing process of e-commerce orders, for example, the stages include a stage of collecting e-commerce products corresponding to a representation order to a user favorite, a stage of adding e-commerce products corresponding to the representation order to a shopping cart corresponding to a user shopping cart, a stage of paying the e-commerce products corresponding to the representation order in a whole payment process, a stage of delivering the e-commerce products corresponding to the representation order to a logistics link, and a stage of delivering the e-commerce products corresponding to the representation order to a user and then delivering the e-commerce products to an after-sales link. Furthermore, for each stage, a plurality of specific business links are subdivided, for example, taking a payment stage as an example, paid business links, unpaid business links and the like can be further subdivided; for another example, taking the after-sale stage as an example, a refund business link, a maintenance business link, and the like can be further subdivided. And the like, various corresponding process labels can be flexibly customized according to different stages, so that the process labels correspond to business links of the e-commerce ordering process.

Of course, different business links of the e-commerce order business process may also correspond to different stages thereof, that is, each stage is regarded as a business link. Therefore, the depth of the flow label corresponding to the e-commerce order flow can be flexibly set by a person skilled in the art according to needs, and the more specific the corresponding business link is, the more precise the node of the e-commerce order flow corresponding to the flow label is, so that the precision of the flow label for marking the standard problem or the preset answer marked by the flow label is exerted by setting the fineness of the business link corresponding to the flow label. Therefore, the process label plays a role in the application, not simply classifying the labeled preset questions and preset answers, but mapping the specific service segmentation links of the e-commerce order process, and playing a key role in guiding response with correct meaning to the questioning text of the user according to different service links.

The standard questions in each question set can be labeled with corresponding flow labels according to the actual relevance of the text content, and similarly, the preset answers in each answer set can be labeled with corresponding flow labels according to the actual relevance of the text content. When the process labels are labeled, one standard question is allowed to be labeled with a plurality of process labels, and similarly, one preset answer can also be labeled with a plurality of process labels. Therefore, the preset answer can be called according to the flow label.

On the basis of the given knowledge base, the question text submitted by the user can be further semantically matched with the total number of preset questions in the knowledge base so as to determine the preset question which is semantically most matched with the question text, namely the preset question which is semantically most similar.

When semantic matching is carried out, deep semantic information of the question text can be extracted by adopting a convolutional neural network model, then the preset question closest to the data distance of the question text is determined according to the data distance between the deep semantic information of the question text and the deep semantic information of each preset question in a knowledge base, and the determined preset question is the target question which is most matched with the question text. Specific embodiments for implementing this matching process will be further given in the subsequent embodiments of the present application, and the table is not pressed here for the time being. The convolutional neural network model can be a convolutional neural network model realized based on CNN and RNN, including but not limited to TextCNN, LSTM, Transformer, Bert, Albert, Electrora and the like, and is suitable for representing and learning texts to obtain a basic model of corresponding sentence vectors. Preferably, the convolutional neural network model can also be constructed as a twin network so as to improve the calculation efficiency of the similarity matching. Those skilled in the art can select a suitable basic model in the prior art to extract a sentence vector representing deep semantic information of the question text according to the principles disclosed in the present application.

After the target question is determined, a question set in which the target question is located is also determined, so that a standard question can be determined, an answer set mapped with the standard question can be determined according to the mapping relation between the answer set and the question set, and the answer set can be used as the target answer set. As mentioned above, the target answer set includes a plurality of preset answers, and each preset answer is labeled with one or more process labels in the label pool.

Step S1300, determining a classification probability corresponding to each flow label mapped to the label pool by using a classification model trained to a convergence state, and determining a preset answer as a target answer, where the classification probability of the labeled flow label is a relative maximum value in a range of a target answer set, from the target answer set:

in order to implement the method, a classification model is prepared, and the classification model is trained to be in a convergence state and then used in the method and is mainly used for determining the classification probability of each flow label mapped to the label pool for the questioning text.

The classification model mainly includes a text feature extraction model and a classifier, where the text feature extraction model may be the same model as the model used to extract the question text and the sentence vectors of the preset questions in step S1200, so that after the classification model is trained to converge, the text feature extraction model therein may be used to extract the sentence vectors for the question text and the preset questions, or may be different models independently equipped, and how to implement the classification model is specific, and those skilled in the art can implement the method flexibly according to the disclosure herein. The classifier is used for performing classification mapping according to the sentence vector obtained by the text feature extraction model so as to calculate the classification probability of the sentence vector mapped to a plurality of classification labels of the classification space. Thus, the classifier can be a multi-classifier constructed by the softmax function, with the obtained classification probabilities normalized to a numerical space of (0, 1), and the sum of the classification probabilities for all classification tags being 1.

The classification model is trained to a convergence state in advance through a training process before being put into use, and the steps of the training process are exemplarily and deeply disclosed in a subsequent embodiment, which is not shown here. After the classification model is trained, the classification mapping capability according to a given question text is learned, and the question text is mapped to each classification label in a preset classification space, namely to each process label in the label pool, so that the classification probability corresponding to each process label is obtained.

Therefore, each flow label in the label pool corresponds to the question text and has a corresponding classification probability, obviously, the classification probability has a function of being used as a sequencing index, the preset answers in the target answer set are associated and labeled with the flow label, and the corresponding classification probability can be determined through the flow label, so that the classification probability can be used for comparing the matching degree of all the preset answers in the target answer set.

Specifically, mapping relation data between each flow label and the classification probability thereof actually generated according to the question text can be regarded as a table, then the table is queried according to the flow labels carried by preset answers in a target answer set to determine the corresponding classification probability, if one preset answer contains a plurality of flow labels, the classification probabilities corresponding to the plurality of flow labels can be respectively determined, and the flow label with the highest classification probability in the preset answer is taken as the target classification probability for comparing the preset answer with other preset answers. Therefore, each preset answer in the target answer set obtains a target classification probability and the target classification probability, the preset answers in the target answer set can be ranked and compared according to the target classification probability, the preset answer with the maximum target classification probability is determined to be the target answer, the target answer is in the target answer set, and the labeled process label classification probability is the preset answer with the maximum value in the range of the target answer set. The target answer is theoretically an appropriate answer corresponding to a question included in the quiz text in meaning and corresponding to a business link associated with the quiz text in business link.

Step S1400, the target answer is pushed to a chat interface of the e-commerce customer service system to be displayed:

after the target answer is obtained, the server can push the target answer to the user side terminal equipment submitting the question text, and the user side terminal equipment displays the target answer to a chat interface of the e-commerce customer service system of the user side terminal equipment after receiving the target answer, so that the user can obtain a more accurate target answer.

From the above disclosure of the exemplary embodiment and its modified embodiments, it can be known that the technical solution of the present application has various positive effects, including but not limited to the following aspects:

Referring to fig. 2, in a further embodiment, the step S1200, obtaining a preset question with a semantic most similar to the question text from the total preset questions in the knowledge base of the e-commerce customer service system as a target question, and determining an answer set mapped to a question set to which the target question belongs as a target answer set, includes the following steps:

step S1210, extracting a sentence vector of the question text by adopting a text feature extraction model trained to a convergence state:

and extracting sentence vectors of the question text by adopting a text feature extraction model trained to be in a convergence state. The text feature extraction model can be a convolutional neural network model realized based on CNN and RNN, including but not limited to TextCNN, LSTM, Transformer, Bert, Albert, Electrora, and the like, and is suitable for representing and learning texts to obtain a basic model of corresponding sentence vectors. The skilled person can train itself according to various training modes to learn how to represent and learn the text information to obtain the sentence vector representing the deep semantic information of the text information.

Step S1220, calculating similarity data between the sentence vector of the question text and the sentence vectors extracted in advance by the text feature extraction model for the full-scale preset questions in the knowledge base:

in order to realize the similar matching between the question text and the total number of preset questions in the knowledge base, the sentence vectors of each preset question in the knowledge base are extracted in advance by adopting the text feature extraction model trained to the convergence state, and the sentence vectors and the corresponding preset questions are associated and mapped and stored in the knowledge base, so that the question text and the corresponding preset questions can be conveniently called in the step.

Further, a data distance algorithm is adopted, including any one of a cosine similarity algorithm, an Euclidean distance algorithm, a Pearson correlation coefficient algorithm, a Jacard algorithm and the like, similarity data between the sentence vector of the question text and the sentence vector of each preset problem in the knowledge base is calculated, and the similarity data are characterized according to the similarity that the higher the numerical value is, the more the similarity is, and the lower the numerical value is, the dissimilarity is, so that a similarity data sequence is obtained, and the similarity data sequence is convenient to operate and can be sequenced according to needs. Here, as an alternative, a twin network synchronization may be adopted to perform an instant matching between the question text and each preset question, so as to determine similarity data between the two questions.

Step S1230, a preset problem that the similarity data is higher than a preset threshold and the similarity data is the maximum is screened out, and is determined as a target problem:

a preset threshold is used, which may be an empirical threshold or an experimental threshold, for determining whether the determined similarity data satisfies the most basic similarity requirement. Therefore, the similarity data corresponding to each preset problem in the knowledge base can be sequenced to determine the similarity data of the maximum value and the corresponding preset problem, then the similarity data of the preset problem is compared with the preset threshold, if the similarity data of the preset problem is higher than the preset threshold, the preset problem can be determined as a target problem, otherwise, the fact that the matched preset problem does not exist in the knowledge base can be judged, and an entrance for jumping the artificial customer service is pushed to the user, so that the user can communicate with the artificial customer service user through the entrance.

Step S1240, obtaining an answer set mapped correspondingly to the standard question of the question set to which the target question belongs from a knowledge base as a target answer set:

after the target question is determined, the problem set to which the target question belongs, namely the problem set containing the target question, can be determined. After the set of questions is determined, the standard questions therein may be determined. After the standard question is determined, the answer set mapped with the standard question can be determined, and the answer set is the target answer set.

In the embodiment, through a specific similarity matching method, the corresponding target answer set is determined for the question text, so that the preset answer serving as the target answer is further preselected in the target answer set subsequently, the matched preset question is screened, and the data range of the subsequently determined target answer is narrowed.

Referring to fig. 3, in a further embodiment, in the step S1300, determining, by using a classification model trained to a convergence state, a classification probability corresponding to each process label mapped to the label pool of the question text, and determining, as a target answer, a preset answer with the classification probability of the labeled process label being a relative maximum value in a range of the target answer set, the method includes the following steps:

step S1310, extracting a sentence vector of the question text by using a text feature extraction model in the classification model:

in this embodiment, the classification model includes a text feature extraction model and a classifier, which are trained to a convergence state in advance to be used, so that the text feature extraction model therein is suitable for extracting a sentence vector for the question text, and then the sentence vector is classified and mapped by the classifier. Therefore, the question text is input into the classification model, and the sentence vector of the question text is extracted by the classification model. Similarly, the text feature extraction model may be a convolutional neural network model implemented based on CNN and RNN, including but not limited to TextCNN, LSTM, Transformer, Bert, Albert, electrora, and the like, and is suitable for performing representation learning on a text to obtain a basic model of a corresponding sentence vector.

Step S1320, performing classification mapping according to the sentence vector of the question text by using a classifier preset in the classification model, and obtaining the classification probability corresponding to each flow label mapped to the label pool by the question text:

the sentence vectors of the questioning text enter the classifier through the full-connection layer, and the classifier calculates the classification probability corresponding to each flow label mapped to the label pool according to the sentence vectors of the questioning text. As mentioned above, the classification probability set obtained by the classifier can be regarded as a query table, which contains the mapping relationship between the process label and its classification probability.

Step S1330, determining one or more classification probabilities corresponding to each preset answer according to one or more process labels carried by the preset answers in the target answer set:

the first determined target answer set comprises a plurality of preset answers, each preset answer carries one or more process labels, and the process labels can determine the corresponding classification determination by inquiring the classification probability set. For each preset answer in the range of the target answer set, no matter how many process labels are provided, only the process label with the largest classification probability can be reserved, and the classification probability of the process label is used as the target classification probability.

Step S1340, comparing the classification probabilities corresponding to the preset answers in the target answer set, determining the maximum classification probability, and taking the preset answer with the maximum classification probability as the target answer:

and comparing the classification probability corresponding to each preset answer in the target answer set, comparing the target classification probability of each preset answer under the condition that the target classification probability of each preset answer is screened out, sequencing each preset answer in the target answer set according to the target classification probability from large to small, wherein the preset answer with the highest sequencing can be determined as the target answer.

In this embodiment, the classification probability obtained by classifying the sentence vectors of the quiz text is used as a sorting basis to sort the preset answers in the target answer set, and finally the preset answer corresponding to the flow label with the maximum classification probability is determined, and accordingly, the determined preset answer is semantically matched with one preset question in the question set which is semantically mapped with the answer set in which the preset answer is located, so that the preset answer has a semantic correspondence relationship with the quiz text, and the preset answer corresponding to the flow label with the maximum classification probability obtained by classifying the sentence vectors of the quiz text is also used, so that the preset answer corresponds to the quiz text in a business link, so that the target answer determined for the quiz text is more accurate and fine, and the quiz corresponding to the e order is taken as an example, so that the answer text can be provided closer to the business link suggested or indicated by the quiz text, the user experience of the e-commerce customer service system can be obviously improved.

Referring to fig. 4, in a further embodiment, in order to make the classification model learn the sentence vector of the extracted question text and the capability of performing classification mapping according to the sentence vector, the training process of the classification model is implemented in advance, and includes the following steps:

step S2100, selecting a preset problem from the problem set in the knowledge base as a training sample, inputting a text feature extraction model of a classification model, and extracting a sentence vector:

as described above, the problem set in the knowledge base establishes a mapping relationship with the process labels in the label pool by pre-labeling the standard problems therein, so that the preset problems in the problem set also have a mapping relationship with the process labels of the standard problems by the standard problems, and accordingly, a preset problem and a process label labeled by the corresponding standard problem can be constructed as the training data for training the classification model. A plurality of pairs of such training data may be organized according to this principle to form a training data set for training the classification model.

When the classification model is trained, a preset problem in the training data is selected as a training sample, and the training sample is input into a text feature extraction model in the classification model to extract deep semantic information so as to obtain a corresponding sentence vector.

Step S2200, classifying and mapping the sentence vector through a classifier to obtain the classification probability corresponding to each flow label mapped by the sentence vector to the label pool, and determining the target flow label corresponding to the maximum classification probability:

the sentence vectors of the training samples are further mapped to a classification space corresponding to the classifier through a full connection layer, the classification space is the classification space defined by the label pool, and therefore the classification space comprises classification labels corresponding to the full flow labels of the label pool one by one, so that the classifier can calculate the classification probability corresponding to the flow labels mapped to the flow labels based on the sentence vectors, and then the flow label corresponding to the maximum classification probability is determined to be the target flow label.

Step S2300, taking the flow label carried by the standard problem in the selected problem set as a supervision label, calculating the loss value of the target flow label, and terminating the training if the loss value reaches a preset threshold value and reaches a convergence state; otherwise, implementing gradient updating, and implementing iterative training on the classification model by adopting the next training sample:

because a flow label is carried as a preset problem of the training sample in the training data of the training sample, the flow label can be directly used as a supervision label of the classification model, the loss value of the target flow label relative to the supervision label is calculated under the action of a cross entropy loss function, then whether the loss value reaches a preset threshold value required by the training of the classification model is judged, if the loss value reaches the preset threshold value, the classification model is trained to a convergence state, the training of the classification model can be terminated, and the classification model is put into normal use. Otherwise, if the preset threshold is not reached, it indicates that the classification model has not been trained to the convergence state, so that gradient update can be performed on the classification model according to the loss value, the weight parameters of each link of the classification model are modified by back propagation so as to further approach the loss function of the classification model to convergence, and then, the training sample in the next training data is called continuously from step S2100 to perform loop iteration, so that the classification model is continuously iteratively trained, and the classification model finally reaches the convergence state.

The text feature extraction model in the classification model obtained by training in this embodiment can be used as a sentence vector for extracting question texts submitted by users and preset questions of a knowledge base in the technical scheme of the present application, so that the text feature extraction model required in step S1200 and the specific steps thereof and step S1300 can be obtained only by training the classification model, and the implementation cost can be saved compared with the case of respectively training two different models.

In the embodiment, the classification model is trained by adopting the data in the knowledge base, so that a training data set is avoided being additionally constructed, the training implementation cost can be saved, the classification model can more accurately obtain the sentence vector of the quiz text, the classification probability set corresponding to the sentence vector can be more accurately obtained, and the preset answers in the target answer set are guided to be effectively sequenced to determine the target answer matched with the quiz text.

step S2400, extracting a sentence vector for each preset problem in the knowledge base respectively by using a text feature extraction model in the classification model trained to be in a convergence state, associating the sentence vector with the preset problem and storing the sentence vector in the knowledge base:

after the classification model is trained to be in a convergence state, the text feature extraction model can be adopted to extract a sentence vector for each preset problem in the knowledge base, and the extracted sentence vector and the preset problem are associated and stored in the knowledge base, so that the sentence vector of the preset problem can be directly quoted subsequently, the data access rate of the e-commerce customer service system in response can be improved, and the response speed of the questioning text can be improved. Therefore, a new text feature extraction model does not need to be trained for extracting the sentence vector with the preset problem, and the deployment cost of the technical scheme of the application can be saved naturally.

Referring to fig. 5, an intelligent automatic customer service response device adapted to one of the objectives of the present application is a functional implementation of the intelligent automatic customer service response method of the present application, and the device includes: the method comprises the following steps: the system comprises a question response module 1100, a question hit module 1200, an answer hit module 1300 and a question response module 1400, wherein the question response module 1100 is used for acquiring a question text submitted by a chat interface of a customer service system of an e-commerce; the question hit module 1200 is configured to obtain a preset question with the most similar semantic meaning to the question text from the full preset questions in the knowledge base of the e-commerce customer service system as a target question, and determine an answer set mapped to a question set to which the target question belongs as a target answer set; the knowledge base comprises mapping relation data between a question set and an answer set, wherein each question set comprises a plurality of preset questions, and each preset question comprises a standard question and a plurality of similar questions; each answer set comprises a plurality of preset answers, the standard questions and the preset answers are labeled by process labels in a label pool, and each process label in the label pool represents each business link in the e-commerce ordering process correspondingly; the answer hit module 1300 is configured to determine, by using a classification model trained to a convergence state, a classification probability corresponding to each process label mapped to the label pool in the question text, and determine, from the target answer set, a preset answer as a target answer, where the classification probability of the labeled process label is a relative maximum value in a range of the target answer set; the question answering module 1400 is configured to push the target answer to a chat interface of the e-commerce customer service system for display.

In a further embodiment, the answer hit module 1300 includes: the vector extraction submodule is used for extracting the sentence vector of the question text by adopting a text feature extraction model in the classification model; the classification mapping submodule is used for performing classification mapping according to the sentence vector of the question text by adopting a classifier preset in a classification model to obtain the classification probability corresponding to each flow label mapped to the label pool by the question text; the corresponding conversion sub-module is used for determining one or more classification probabilities corresponding to the preset answers according to one or more process labels carried by the preset answers in the target answer set;

In order to solve the technical problem, an embodiment of the present application further provides a computer device. As shown in fig. 6, the internal structure of the computer device is schematically illustrated. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected by a system bus. The computer readable storage medium of the computer device stores an operating system, a database and computer readable instructions, the database can store control information sequences, and the computer readable instructions, when executed by the processor, can enable the processor to realize an intelligent customer service automatic response method. The processor of the computer device is used for providing calculation and control capability and supporting the operation of the whole computer device. The memory of the computer device may have stored therein computer readable instructions that, when executed by the processor, may cause the processor to perform the intelligent customer service auto-answer method of the present application. The network interface of the computer device is used for connecting and communicating with the terminal. Those skilled in the art will appreciate that the architecture shown in fig. 6 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, 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 and its sub-module in fig. 5, and the memory stores program codes and various data required for executing the modules or the sub-modules. The network interface is used for data transmission to and from a user terminal or a server. The memory in this embodiment stores program codes and data required for executing all modules/sub-modules in the intelligent customer service automatic response device of the present application, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

The present application also provides a storage medium storing computer-readable instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of the intelligent customer service automatic 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 according to any embodiment 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 can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when the computer program is executed, the processes of the embodiments of the methods can be included. 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).

To sum up, the intelligent degree of the e-commerce customer service system can be comprehensively improved, answers which are matched with specific business links in an e-commerce ordering process are determined for user questions, and the response accuracy of the e-commerce customer service system is improved.

Those of skill in the art will appreciate that the various operations, methods, steps in the processes, acts, or solutions discussed in this application can be interchanged, modified, combined, or eliminated. Further, other steps, measures, or schemes in various operations, methods, or flows that have been discussed in this application can be alternated, altered, rearranged, broken down, combined, or deleted. Further, steps, measures, schemes in the prior art having various operations, methods, procedures disclosed in the present application may also be alternated, modified, rearranged, decomposed, combined, or deleted.

The foregoing is only a partial embodiment of the present application, and it should be noted that, for those skilled in the art, several modifications and decorations can be made without departing from the principle of the present application, and these modifications and decorations should also be regarded as the protection scope of the present application.

Claims

1. An intelligent customer service automatic response method is characterized by comprising the following steps:

2. The automatic intelligent customer service response method according to claim 1, wherein a preset question with the most similar semantic meaning to the question text is obtained from the total preset questions in the knowledge base of the e-commerce customer service system as a target question, and an answer set mapped to the question set to which the target question belongs is determined as a target answer set, comprising the following steps:

and acquiring an answer set mapped corresponding to the standard question of the question set to which the target question belongs from a knowledge base as a target answer set.

3. The method of claim 1, wherein a classification probability corresponding to each process label mapped to the label pool of the question text is determined by using a classification model trained to a convergence state, and a preset answer with the classification probability of the labeled process label being a relative maximum value in a range of a target answer set is determined from the target answer set as a target answer, comprising the following steps:

adopting a text feature extraction model in a classification model to extract a sentence vector of the question text;

4. The intelligent customer service automatic response method according to claim 3, wherein the training process of the classification model is implemented in advance, and comprises the following steps:

5. The intelligent customer service automatic response method according to claim 4, wherein after the training process of the classification model is completed, the method comprises the following steps:

6. The method according to any one of claims 1 to 5, wherein the e-commerce ordering flow comprises a plurality of different stages, each stage comprises one or more business links, a flow label is set corresponding to each business link, and a set of flow labels corresponding to all business links forms the label pool, wherein the different stages comprise a collection stage, a shopping cart stage, a payment stage, a delivery stage and an after-sale stage.

7. An intelligent customer service automatic answering device, comprising:

the question response module is used for acquiring a question text submitted by a chat interface of the electricity merchant customer service system;

the question hit module is used for acquiring preset questions with the most similar semanteme to the question text from the total preset questions in the knowledge base of the e-commerce customer service system as target questions, and determining an answer set mapped by a question set to which the target questions belong as a target answer set; the knowledge base comprises mapping relation data between a question set and an answer set, wherein each question set comprises a plurality of preset questions, and each preset question comprises a standard question and a plurality of similar questions; each answer set comprises a plurality of preset answers, the standard questions and the preset answers are labeled by process labels in a label pool, and each process label in the label pool represents each business link in the e-commerce ordering process correspondingly;

the answer hit module is used for determining the classification probability corresponding to each process label mapped to the label pool by adopting a classification model trained to be in a convergence state, and determining a preset answer as a target answer, wherein the classification probability of the labeled process label is the relative maximum value in the range of a target answer set, and the preset answer is used as the target answer;

and the question answering module is used for pushing the target answer to the chat interface of the e-commerce customer service system for display.

8. A computer device comprising a central processor and a memory, characterized in that the central processor is adapted to invoke execution of a computer program stored in the memory to perform the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that it stores, in the form of computer-readable instructions, a computer program implemented according to the method of any one of claims 1 to 6, which, when invoked by a computer, performs the steps comprised by the corresponding method.

10. A computer program product comprising computer program/instructions, characterized in that the computer program/instructions, when executed by a processor, implement the steps of the method as claimed in any one of claims 1 to 6.