CN118113828A - Recommended speaking content generation method and electronic equipment - Google Patents

Abstract

The embodiment of the application discloses a recommended conversation content generation method and electronic equipment, wherein the method comprises the following steps: monitoring dialogue contents generated in the process of dialogue between customer service personnel and a user; if the customer service person performs an operation of sending a solution to the user for a first appeal of the user, constructing a prompt text for interacting with a content generation model, and inputting dialogue content associated with the user into the content generation model, so that the content generation model determines whether the user approves the solution or generates a second appeal/problem, and generates a corresponding recommended speaking; and providing the recommended conversation to the customer service personnel so that the customer service personnel can carry out a dialogue with the user according to the recommended conversation. According to the embodiment of the application, the communication efficiency and the user experience can be improved, and meanwhile, the dependence on skills such as skills of customer service personnel in terms of language can be reduced.

Description

Recommended speaking content generation method and electronic equipment

Technical Field

The application relates to the technical field of content generation, in particular to a recommended conversation content generation method and electronic equipment.

Background

In the commodity information service system, after-sale problems often occur, some problems cannot be agreed between a consumer and a merchant, even when disputes and the like occur, the consumer can complain through the customer service system of the platform side, and then the customer's appeal can be known through an on-line conversation or telephone and other modes by the manual customer service of the platform side, and a solution is provided for the consumer. Of course, in addition to after-market issues, consumers may also initiate consultation with the platform side customer service regarding issues such as marketing campaigns initiated by the platform side, and so on.

In the above-described process of servicing a consumer by a human attendant, the attendant typically needs to have some dialog with the consumer, including understanding the customer's appeal, etc., in addition to providing the consumer with a suitable solution. In the conversation process, it is also important that customer service personnel communicate with consumers by using any conversation technique, and the conversation effect can be improved by using the conversation technique, so that the consumers can be more practically helped to solve the problem. However, this places a relatively high demand on the ability of the customer service personnel, and in addition, the customer service personnel may take a relatively long time to think about how to express, so that the service efficiency is affected, and so on.

In order to help customer service staff to improve service effects, some conversations can be recommended to the customer service staff in some systems, but most of recommended conversations are preconfigured and mainly used for emotion soothing, calling and the like, and the functions of solving user problems and the like are generally difficult to achieve.

Disclosure of Invention

The application provides a recommended conversation content generation method and electronic equipment, which can improve communication efficiency and user experience, and reduce dependence on skills such as skills of customer service staff in terms of using languages.

The application provides the following scheme:

A recommended conversation content generation method, comprising:

monitoring dialogue contents generated in the process of dialogue between customer service personnel and a user;

If the customer service person performs an operation of sending a solution to the user for a first appeal of the user, constructing a prompt text for interacting with a content generation model, and inputting dialogue content associated with the user into the content generation model, so that the content generation model determines whether the user approves the solution or generates a second appeal/problem, and generates a corresponding recommended speaking;

And providing the recommended conversation to the customer service personnel so that the customer service personnel can carry out a dialogue with the user according to the recommended conversation.

Wherein the recommended speaking is used for inquiring/answering to the user aiming at the approval/disapproval condition of the solution or the second appeal/problem of the user.

Wherein before the calling the content generation model, the method further comprises:

And according to feedback information submitted by the user and approved or not approved by the solution or through keywords contained in dialogue content generated after the user receives the solution, prejudging the approval or disapproval condition of the solution by the user, and inputting the feedback information or prejudging result information into the content generation model so that the content generation model can refer to the process of generating the recommended conversation.

Wherein, when constructing a prompt text for interacting with a content generation model according to the dialogue content associated with the user, the dialogue content associated with the user comprises: in the current dialogue process of the user and the customer service personnel, the solution transmits dialogue contents generated before and after the solution and historical dialogue contents generated in the historical consultation process of the user.

Wherein the content generation model is pre-trained by:

Extracting information of solution key nodes in the service flow execution specification;

Acquiring standard questions corresponding to various solutions, enumerating reasons that the solutions may be approved or disapproved, and respectively inquiring/answering to provide corresponding speech samples for various reasons;

collecting historical dialogue record information between a plurality of customer service personnel and a plurality of users;

Inputting the speaking samples and the historical dialogue record information into an artificial intelligent AI large language model, summarizing the speaking samples of the various solutions in various conditions by the AI large language model, carrying out service flow simulation on user problems and information states related in the historical dialogue record information according to the service flow execution specifications, determining reasonable solutions and the following questions/following questions, comparing simulation results with the solutions and the following questions/following questions actually given by the customer in the historical dialogue record, determining whether the solutions and the following questions/following questions actually given by the customer in the historical dialogue record are reasonable or not, and marking;

And pre-training the content generation model by using the marked data.

Wherein, still include:

Before the customer service personnel performs a dialogue with a user, determining core input parameter information required by a content generation model, wherein the core input parameter information comprises: dialogue content generated by the user in a current service session and/or historical dialogue content generated in an associated historical service session;

And constructing a prompt text for dialogue with the content generation mode, inputting the core input parameter information into the content generation model, so that the content generation model extracts key questions according to the core input parameter information, analyzes user appeal, historical processing results and/or disputes of two parties of the dialogue, and generates recommended first-term contents, wherein the recommended first-term contents comprise descriptions of questions possibly required to be asked by the user and back-questions based on the summarized user appeal.

Wherein the content generation model is further to: and carrying out emotion recognition on dialogue content associated with the user, and carrying out concentric color rendering processing on the generated recommended speech.

A method of pre-training a content generation model, comprising:

extracting information of solution key nodes in the client service flow execution specification;

Acquiring standard questions corresponding to various solutions, enumerating reasons that the solutions may be approved or disapproved, and respectively inquiring/answering to provide corresponding speech samples for various reasons;

Collecting historical dialogue content information between a customer and a user in a plurality of historical service sessions;

Inputting the speech samples and the historical dialogue content information into an artificial intelligent AI large language model, summarizing the speech samples of the various solutions in various conditions by the AI large language model, carrying out standard service flow simulation on user problems and information states related in the historical dialogue record information according to the service flow execution specifications, determining reasonable solutions and the speech samples of the following questions/following questions, and determining whether the solutions and the speech steps actually given by the customer service personnel in the historical dialogue content are reasonable or not and marking by comparing simulation results with the solutions and the speech steps actually given by the customer service personnel in the historical dialogue content;

and pre-training the content generation model by using the marked data so that after customer service personnel provide a solution for the user aiming at the first appeal of the user, the content generation model determines whether the user approves the solution or generates the second appeal/problem according to the dialogue content associated with the user and generates a corresponding recommended conversation.

A recommended conversation content generation method, comprising:

before customer service personnel conduct a conversation with a user, core input parameter information required by a content generation model is determined, wherein the core input parameter information comprises: dialogue content generated by the user in a current service session and/or historical dialogue content generated in an associated historical service session;

And constructing a prompt text for dialogue with the content generation mode, inputting the core input parameter information into the content generation model, so that the content generation model extracts key questions according to the core input parameter information, analyzes user appeal, historical processing results and/or disputes of two parties of the dialogue, and generates recommended first-term contents, wherein the recommended first-term contents comprise descriptions of questions possibly required to be asked by the user and back-questions based on the summarized user appeal.

A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the method of any of the preceding claims.

An electronic device, comprising:

One or more processors; and

A memory associated with the one or more processors, the memory for storing program instructions that, when read for execution by the one or more processors, perform the steps of the method of any of the preceding claims.

A computer program product comprising computer program/computer executable instructions which, when executed by a processor in an electronic device, implement the steps of any one of the methods of the preceding claims.

According to the specific embodiment provided by the application, the application discloses the following technical effects:

According to the embodiment of the application, in the process of providing the consultation service for the user by the customer service personnel, the dialogue content generated by the two parties can be monitored, if the customer service personnel performs the operation of sending the solution to the user aiming at the first appeal of the user, the prompt text for interacting with the content generation model can be constructed, and the dialogue content associated with the user is input into the content generation model, so that the content generation model determines whether the user approves the solution or whether the second appeal/problem is generated by the user, and the corresponding recommended speaking operation is generated. Further, the recommended utterances may be provided to the attendant so that the attendant can converse with the user according to the recommended utterances. In this way, after the customer service personnel provides a certain solution, whether the user approves the solution or whether the second appeal/problem appeal is generated or not can be inferred through the content generation model, and a recommended conversation can be generated for a specific situation, so that the customer service personnel can use the recommended conversation to carry out further conversation on the user, the communication efficiency and the user experience can be improved, and meanwhile, the dependence on skills such as skills of the customer service personnel in terms of using language is reduced.

Of course, it is not necessary for any one product to practice the application to achieve all of the advantages set forth above at the same time.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of a system architecture provided by an embodiment of the present application;

FIG. 2 is a flow chart of a first method provided by an embodiment of the present application;

FIG. 3 is a schematic diagram of a first interface provided by an embodiment of the present application;

FIG. 4 is a schematic diagram of a second interface provided by an embodiment of the present application;

FIG. 5 is a flow chart of a second method provided by an embodiment of the present application;

FIG. 6 is a flow chart of a third method provided by an embodiment of the present application;

fig. 7 is a schematic diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which are derived by a person skilled in the art based on the embodiments of the application, fall within the scope of protection of the application.

In the embodiment of the application, in order to help customer service personnel provide service efficiency and help users to solve the problem, a corresponding solution is provided. In this scenario, a recommended speaking, i.e. skills, methods used by the language during speaking, etc., can be generated intelligently. High quality speech surgery can more perfectly express its own meaning, increasing the clarity of the transmitted message, and further harvesting good or expected results.

Specifically, considering that a customer service person typically needs to learn about a user's appeal through one or more rounds of dialogue during communication with the user, the user may recognize that the solution may or may not recognize after the solution is presented, or may generate additional appeal or present new questions, the customer service person may need to provide a satisfactory solution to the user by making further questions/answers to the user, and so on. In the above process, what kind of speaking is adopted in the process of asking and answering the user is important, and more factors need to be considered, so that the speaking capability requirement in the process of asking and answering is higher for the customer service staff, and the customer service staff can consume more time when thinking about how to conduct asking and answering.

In view of this situation, in the embodiment of the present application, a recommended conversation for inquiry/response is provided based on AIGC (ARTIFICIAL INTELLIGENCE GENERAL Computationalism, artificial intelligence production content), so that a customer service person can talk to a user based on such recommended conversation, thereby improving service efficiency and facilitating improvement of user satisfaction.

AIGC describes a theoretical framework that considers artificial intelligence systems to implement intelligence by mimicking the intelligent behavior and thought process of humans. AIGC the core idea is that an artificial intelligence system can achieve intelligent behavior by using computational models to simulate the brain and cognitive processes of humans. In other words, the computational model may provide interpretation and simulation of human intelligence, while also being useful in constructing higher-level artificial intelligence systems. AIGC is to build a theoretical framework to better understand and implement artificial intelligence. Based on the above framework, the industry has emerged a variety of generative models, including "Wen Shengwen" models (text-based generation of text), a "meridional map" model (image-based generation of text), a "pictorial text" model (image-based generation of text), and so forth. In an embodiment of the present application, such a large "Wen Shengwen" AI model may be utilized to implement content generation in connection with recommended utterances.

Specifically, conversation content generated during a conversation between a customer service person and a user may be monitored, and if the customer service person sends a solution to the user, prompt text (Prompt) for interacting with a content generation model may be constructed, and conversation content associated with the user (which may include conversation content generated before and after the solution is sent during a current consultation, historical conversation content generated during a last period of time, and the like) may be input into the content generation model, so that the content generation model determines whether the user approves or generates additional second appeal/questions for the currently provided solution, and generates a recommended speaking for a post-query/post-answer for the approval or disapproval condition or the second appeal/questions. In this way, the customer service personnel can talk to the user based on such recommended utterances.

Wherein the content generation model may also be pre-trained in advance in order to enable the content generation model to have the above-described capabilities, or in order to ensure the quality of the generated recommended utterances. During training, first, information of key nodes of a solution in a service flow execution specification (customer service personnel can provide a specific solution for a user based on the specification) can be extracted, standard questions corresponding to multiple solutions are obtained, reasons that the solution may be approved or not approved are enumerated, and corresponding speech samples are provided when the solution is respectively queried/tracked for multiple reasons. In addition, historical dialogue record information between a plurality of customer service staff and a plurality of users can be collected, and the solutions given by the customer service staff and the questions/answers in the historical dialogue record information are marked reasonably by the AI large language model through summarizing the dialogue samples of the various solutions in various situations and by inputting the dialogue samples and the historical dialogue record information into the AI large language model (which is different from the content generation model, and is mainly used for automatically marking training samples, so that the manual marking process is omitted). After the labeling is performed, the content generation model can be trained by using the training sample with the labeling information, so that the content generation model can generate high-quality recommended speech in various different scenes.

From the view of system architecture, referring to fig. 1, the embodiment of the present application may be applied to a merchandise information service system, and specifically may be a service function for providing a specific intelligent generation recommendation in a workstation system of a customer service person. In particular, when a user sends a consultation request such as complaints to a platform side, customer service personnel in the platform can provide services for the user, in the service process, the embodiment of the application can monitor dialogue contents generated by the two parties, and if the customer service personnel execute an operation of sending a solution to the user, a recommended speaking operation for inquiring/answering the user can be generated through a content generation model. The content generation model can be pre-trained through samples in advance, and the labeling process of a specific sample can be automatically completed by the AI large model on the basis of manually providing part of the samples.

The following describes in detail the specific implementation scheme provided by the embodiment of the present application.

Example 1

First, this embodiment provides a recommended talking content generating method, referring to fig. 2, which may include:

S201: and monitoring dialogue contents generated in the process of conducting dialogue between customer service personnel and a user.

The customer service personnel can comprise manual customer service, customer service robots and the like. Specifically, the user may communicate with customer service personnel of the platform party when he/she needs to complain about disputes and the like generated before the user is in charge of the business, or needs to consult a marketing campaign of the platform party, and the like. In the embodiment of the application, a user can carry out online dialogue with customer service personnel on a platform side through a specific access entrance provided in a client, the user can input contents requiring complaints or consultations such as problems of the user through a chat window and the like provided in an interface, and the contents replied by the customer service personnel can be displayed in the chat window.

S202: if the customer service person performs an operation of sending a solution to the user for a first appeal of the user, a prompt text for interacting with a content generation model is constructed, and dialogue content associated with the user is input into the content generation model, so that the content generation model determines whether the user approves the solution or generates a second appeal/problem/, and generates a corresponding recommended conversation.

Specifically, during the process of the customer service personnel and the user, each action (for example, each sentence input by the two parties, etc.) can call the content generation model to conduct the conversation recommendation. Since the dialog contents specifically generated from the viewpoint of the customer service person may include two kinds, one is to input the specific dialog contents by typing or the like, and the other is to send a solution to the user. In the solution, since the selection can be generally performed from the standard service flow execution specifications, and no specific text input work needs to be performed, the two types of dialogue contents can be distinguished according to different input modes in the process of monitoring the dialogue contents. Further, different conversation content recommendations may be made based on different categories of specific conversation content. Specifically, if the specific action corresponds to only ordinary dialog content and does not involve sending a solution to the user, ordinary general speaking may be recommended. And if the customer service personnel sends a certain solution to the user for the first complaint of the user, a professional speaking recommendation procedure can be entered.

Specifically, in the professional speaking recommendation process, a Prompt text for interacting with the content generation model may be automatically constructed (promt, the content generation model may theoretically generate arbitrary content, but needs to understand what content it needs to generate through "promt"), for example, "what is you about to do you further follow up/ask after you should be about what is you a customer service, a solution has been sent to the user? ", etc. In addition, the dialogue content associated with the user can be input into the content generation model as an input parameter. The dialogue content associated with the user can comprise dialogue content generated between the user and customer service personnel in the current consultation process, historical dialogue content generated by the user in the historical consultation process, and the like.

The dialogue content generated in the current consultation process can comprise the solution information recently sent by customer service personnel, the dialogue content of both parties before and after the solution is sent, and the like. Wherein the dialog content after sending the solution may comprise feedback actions made by the user for the current solution, or dialog content entered further after receiving the solution, etc. Specifically, when the solution is sent to the user, the solution information may be carried in the form of an information card or the like, so that the user may perform operations such as clicking to jump to a receiving page corresponding to the specific solution, and so on. In this case, options such as "approval", "disapproval" and the like, through which the user can make feedback, may also be provided in the information card. Of course, the user may override the operational option and instead continue to talk to the customer service personnel after seeing the solution, in which case it may be expressed whether he or she approves the solution or may also present further additional appeal/questions, etc. Such dialog content, which is further generated after receipt of the solution, may thus also be provided to a content generation model, which determines, by natural language understanding of such dialog content, etc., whether the user approves the current solution, or whether there is some additional second appeal/problem, etc. In a customer service scenario, a problem of a user is generally a different concept from a demand, and the problem is generally a problem before or after sales which is presented by the user for a certain order, etc., and the demand is used for expressing what the user specifically wants to help a customer service person. For example, assume that the dialog content entered by the user is: "how does a bad fruit after receiving a certain order? Can a return for goods or claim? "the user's question is" the goods received by the order have quality problems ", the claim is" return goods or obtain corresponding claims ", etc. It should be noted here that, regarding a specific solution, in the dialogue content input by the user, the appeal may be directly included, or only the problem may be described, and at this time, the appeal that may be required by the user may be predicted by the large model according to the problem input by the user. For example, customer service personnel give solutions about "pay", and the dialogue content further entered by the user is: "punish merchant" can infer that the user has further appeal of "punish merchant" on his own, etc. Or the user may just ask a question after seeing the solution given by the customer service person, e.g. the customer service person gives a solution about "pay", the user may ask: "when does i receive payment for a claim? Is "or" the original return? ", etc., which all pertain to simple challenges, etc.

The historical dialog content may be dialog content generated when consultation is performed through a plurality of different consultation channels within a last period of time (e.g., last 7 days, etc.), and if the current consultation is related to a certain order, the historical dialog content related to the order may be screened out from the historical dialog content, etc. Such historical dialog content may assist in generating specific recommended utterances.

After receiving the specific prompt text and the dialogue content, the content generation model can judge whether the user approves the current solution or has further additional appeal/problem according to feedback actions made by the user after receiving the current solution or the input dialogue content and the like. Based on this, a recommended speaking for asking/answering for approval/disapproval cases or the additional appeal/problem can be generated based on dialogue contents, history dialogue contents, and the like generated during the current consultation.

For example, as shown in fig. 3, in the communication process between the user and the customer service personnel, the customer service personnel expresses that his own appeal is "complaint against the business service attitude problem of a certain order", and the customer service personnel knows the situation and then gives a solution that the customer is "affinitive", the problem fed back by the customer is carefully registered, the service attitude problem of the complaint seller will record the assessment index of the seller (including but not limited to warning, deducting, reducing the right and other penalties of the store), and meanwhile the platform also continuously strengthens the supervision, so that the customer brings bad shopping experience to the customer with great sorry; after receiving the solution, the user continues to input the dialogue content as follows: "punishment" the content may be content that the user has asked. After inputting the above information into the content generation model, the content generation model can understand that the user may be approving the currently given "check index of record vendor" scheme by inputting dialogue content to the user after receiving the solution, but emphasizes the part in which "punishment" is done to the merchant, at which time a recommended speaking can be generated for this case. For example, the specific term may be: "a platform attaches great importance to the consumer's interests, we have special departments to supervise merchants, and if a merchant violation is found, we can penalize according to the situation, such as warning, deduction, and even right-down. Please you feel confident that we will try to guarantee your shopping experience.

In the specific implementation, the content generation model can also perform emotion recognition of the user in the process of generating the recommended call, and perform concentric color rendering on the recommended call according to the emotion recognition result. That is, the user's urgency can be determined by condition recognition, and further different emotions of the specific generated speech surgery can be given to the color rendering, so that the specific speech surgery can be enabled to start a concentric impression. For example, if a plurality of question marks or a plurality of exclamation marks are added when the user is asked, the emotion of the user at the time may be excited, or the current problem is heavier or more urgent, at this time, the content of' the emotion and emotion of the user at the place where the user is located, the emotion and emotion of the user can be more expressed when the recommended speech is generated, so that the capability of the customer service staff in terms of transposition thinking, listening ability, expression respect and the like is reflected.

In addition, in order to further improve the quality of the conversation content generated by the content generation model, besides directly inputting the conversation content text associated with the user as a parameter to the content generation model, the conversation content generation model may also pre-judge whether the user approves or disapproves the solution according to feedback information submitted by the user and approved or disapproved by the solution, or through keywords included in the conversation content generated after the user receives the solution, and input the feedback information or pre-judge result information into the content generation model, so that the content generation model refers to the process of generating the recommended conversation.

In particular, in order to enable the content generation model to have the above-mentioned capability of generating the recommended utterances, or enable the generated recommended utterances to be more reasonably and accurately, the content generation model may be pre-trained in advance. That is, although a content generation model of the "Wen Shengwen" class on the market can theoretically generate arbitrary content, if training specific to a certain application scene is not applied, when content generation is performed in that scene, the actually generated content may not necessarily fit the requirements of the scene, the generated content may be disordered, and the like. Therefore, in specific implementation, the content generation model may be pre-trained for scene requirements in embodiments of the present application.

In particular, in pre-training the content generation model, in order to achieve the diversity of generated recommended utterances, historical dialog records in the customer service system may be collected as sample data, so that the content generation model can learn from these data the ability to generate recommended utterances in various situations. However, because the service capabilities of the customer service staff are different, in the process of providing services for users, the provided solutions, the used dialogues and the like may be unreasonable, so that the historical dialogue records also need to be marked, thereby generating positive samples and negative samples, and thus, the learning effect of the content generation model can be improved.

However, because of the large number of history dialogs, if these history dialogs are manually tagged, the effort is very large, and a lot of manpower and time costs are required. Therefore, in the embodiment of the application, an implementation scheme for automatically labeling the history dialogue record through the AI large model is also provided.

In order to achieve the above-mentioned purpose of automatic labeling, first, the key nodes in the service flow execution specification may be extracted, where the service flow execution specification, that is, the standard action execution specification, may specifically exist in the form of a flow tree, etc., where under various standard problems, it is required to confirm which information of the user is respectively, and under various information states, what solutions are respectively corresponding to, or may also include some recommended utterances, where, of course, the recommended utterances are usually manually configured, and belong to the conventional utterances. For example, a standard question (i.e., a question that the user may ask) is "how me would like to return, the first node in the corresponding flow tree may be to confirm what the commodity is, then determine what the price is, time to order, physical state (whether there is shipping, receiving, etc.), whether there is an expiration date, etc. The various states may have corresponding pre-configured solutions that may help customer service personnel quickly determine what solutions should be provided to the user. In the embodiment of the application, the specifically extracted information of the key node can include the information state, the solution, the manual configuration speaking operation and the like corresponding to a specific node.

After the information on the key node is extracted, the information such as reasons that the specific solution may be approved or disapproved by the user, additional appeal that the user may have, etc. may be enumerated manually, and the corresponding speech techniques for asking/answering the user may be manually configured under the conditions of approval/disapproval, generation of a second appeal/question, etc. These manually configured data may be provided as examples to the AI large model. In addition, the history dialogue record in the system can be input into the AI large model, so that the AI large model can summarize how solutions should be given and how the questions/answers should be asked by using the techniques from understanding the sample information. Furthermore, the AI large model can simulate the service flow according to the service flow execution standard to the user's appeal, the information state on each key node and the like in the history dialogue record, so that the solution and the follow-up/follow-up operation which should be given can be determined. And then, the simulated solution and the conversation process can be compared with the solution and the conversation process actually given by the customer service personnel in the history dialogue record, so that whether the solution and the inquiry/answering process actually given by the customer service personnel are reasonable or not can be determined, and the labeling is carried out. So that whether the solutions, the follow-up questions/answering techniques and the like given by the customer in the specific history dialogue record are reasonable or not can be marked. In this way, automatic labeling of the history dialogue record can be completed.

The large AI model for automatically labeling the history dialogue record and the content generation model for generating the recommended dialogue may be different from each other, and of course, the same model may be used to complete the two tasks in practical applications.

After the marking of the history dialogue record is completed, the history dialogue record with the marking result can be utilized to pretrain the content generation model, so that the content generation model can obtain recommended dialogues for inquiring/answering the user according to various conditions.

S203: and providing the recommended conversation to the customer service personnel so that the customer service personnel can carry out a dialogue with the user according to the recommended conversation.

After the content generation model generates the recommended utterances, such recommended utterances may be provided to the customer service personnel to enable the customer service personnel to interact with the user in accordance with such recommended utterances, including for approval/disapproval of the solution by the user or the second complaint/question, to follow-up/answering to the user, and so forth. For example, as shown in fig. 3, a dialog window interface on the customer service side is shown, and at 31 of the dialog window, recommended speech content generated by the content generation model may be displayed, and in addition, a determination result of whether the user approves the current solution or has an additional second appeal/problem may be displayed. For example, in the example shown in fig. 3, based on the feedback action or the dialogue content input by the user after receiving the solution, it is determined that the second complaint of the user inquiry is "punish the merchant", and then "member inquiry" may be further displayed when displaying the generated recommended dialogue content: the judgment result of whether the user approves the current solution or has additional second appeal/problem, such as requiring punishment merchant, etc., can show the reason of the current recommended conversation, and is convenient for helping customer service personnel confirm whether the recommended conversation can be used for conversation with the user.

After the solution is given to the customer service personnel, according to further feedback or more dialogue contents input by the user, whether the user approves or disapproves the solution or whether the user has further additional second appeal/problem or the like is judged, and a recommended speaking operation is provided for the customer service personnel for inquiring/answering the user in the case that the user approves or disapproves or has the second appeal/problem or the like. In practical applications, besides the difficulty of the inquiry/answering link for the customer service personnel or the relatively high capability requirement for the customer service personnel, the first question (which may be simply referred to as the first question) presented by the customer service personnel to the user after the user enters the line to initiate the consultation request is also important. In a conventional manner, a "first question" question is typically a question to the user, which the user is presented with. However, in the case of such a consultation to the platform customer service, a large proportion may be that the incoming line is transferred from other consultation processes, for example, after the incoming line is transferred, the customer may first talk with the customer through the customer service robot, and the customer has described the problem, but the customer service robot cannot solve the problem, and when the customer service needs to be transferred to the manual customer service, the current customer service personnel can only transfer the incoming line, and so on. Or the user has previously consulted by incoming line, but has not solved the problem or left part of the problem, etc., and then re-enters line to initiate consultation, etc. For the above situation, if the "first question" of the customer service personnel is still for the user to describe the problem, the feeling for the user is repeatedly queried, and then the user needs to repeatedly describe the problem, so the efficiency is low, and the user experience is poor. Of course, in one way, the history dialogue record of the user may be presented to the current customer service personnel, and the current customer service personnel may roughly learn the problem that the user may need to solve by reviewing the history dialogue record, but if the history dialogue record is relatively long, the process of reviewing is relatively time-consuming, which may cause a long waiting time of the user, and so on.

Therefore, in the embodiment of the application, for the situation of switching incoming lines or repeating incoming lines, a recommendation of "first question" can also be generated through a content generation model. Specifically, after the user enters the line, core parameter information (including history dialogue record information generated in the current incoming line service or history incoming line service process) and order information and the like required by reasoning can be obtained. Then, key questions can be extracted from the content generation model, user appeal, historical processing results, disputes of both parties and the like can be analyzed, and on the basis, a recommended first question dialogue can be generated, and the dialogue can be used for carrying out back questions on the basis of the analysis. For example, it is analyzed what the current incoming line of the user may need to ask, and then the question is described in the "first question" session, and in addition, the user may be asked in return for help that the user may need customer service personnel for the question. Therefore, under the condition that the reasoning is correct, the user only needs to answer yes and the like, and the user does not need to repeatedly describe own questions and requirements, so that the communication efficiency can be effectively improved, and the user experience is improved.

For example, as shown in fig. 4, assuming that a user has been incoming a few days ago, it is described that the own problem is "do return do nothing? ", customer service personnel confirm the problem: "is a parent, please ask is that the merchant has not agreed to return? But then the user does not reply any more, the customer service system ends the session because the user has not received the user response for a long time. When the user re-enters the line, it is likely to continue to consult the last question, since the last question may not have been resolved, and so it may be inferred by the content generation model and a recommended "first question" call may be generated, for example, as shown at 41 in fig. 4: "do you see you say that the return merchant does not deal with, ask you to prompt you to agree to return? ", etc. The recommended call can be displayed in a dialogue window at the customer service personnel side, and operation options such as copying to an input box are provided, if the customer service personnel really needs to use the recommended call, the recommended call content can be copied to the input box through the options and clicked to be sent, and the customer service personnel is not required to review the historical dialogue record by themselves, so that the efficiency can be improved, and the waiting time of a user can be shortened. In addition, if the user does need to continue consulting the return question, the user can directly reply to "yes" without repeatedly describing his own questions and appeal. After determining the user's questions and complaints, if the customer service person gives a certain solution, the customer service person may be provided with a recommended conversation for asking/answering the user for a certain feedback situation of the above-mentioned solution, etc. based on the above-mentioned solution, by whether the user approves or disapproves the solution, or whether there is an additional second complaint/question, etc.

In summary, through the embodiment of the application, in the process that the customer service personnel provides services such as consultation, complaint and the like for the user, dialogue contents generated by the two parties can be monitored, if the customer service personnel performs an operation of sending a solution to the user, a prompt text for interacting with a content generation model can be constructed, and dialogue contents associated with the user are input into the content generation model, so that the content generation model can judge whether the user approves the solution or generates a second complaint/problem, and generate a corresponding recommended speaking. Further, the recommended utterances may be provided to the attendant so that the attendant can converse with the user according to the recommended utterances. In this way, after the customer service personnel provides a certain solution, whether the user approves the solution or whether the second appeal/problem is generated or not can be inferred through the content generation model, and the recommended speaking operation can be generated according to specific situations, so that the customer service personnel can use the recommended speaking operation to conduct further dialogue with the user, the communication efficiency and the user experience can be improved, and meanwhile, the dependence on skills such as skills in terms of using languages of the customer service personnel is reduced.

Example two

The second embodiment provides a pre-training method of a content generation model for a pre-training process of the content generation model, referring to fig. 5, the method may include:

s501: extracting information of solution key nodes in the client service flow execution specification, acquiring standard questions corresponding to various solutions, enumerating reasons that the solutions may be approved or not approved, and providing corresponding speaking samples when the solution is respectively queried/answered for various reasons;

s502: collecting historical dialogue content information between a customer and a user in a plurality of historical service sessions;

S503: the speech samples and the historical dialogue content information are input into an artificial intelligent AI large language model, the AI large language model summarizes the speech samples of the various solutions in various cases when the various solutions are queried/answered, standard service flow simulation is carried out on user problems and information states related in the historical dialogue record information according to the service flow execution specifications, reasonable solutions and speech samples for the query/answering are determined, and the simulation results are compared with the solutions and the query/answering speech actually given by customer service personnel in the historical dialogue content to determine whether the solutions and the query/answering speech actually given by the customer service personnel in the historical dialogue content are reasonable or not and mark;

s504: and pre-training the content generation model by using the marked data so that after customer service personnel provide a solution for the user aiming at the first appeal of the user, the content generation model determines whether the user approves the solution or has the second appeal/problem according to the dialogue content associated with the user and generates a corresponding recommended speaking.

Example III

The third aspect of this embodiment provides a method for generating content of a recommended conversation for generating a first-query recommended conversation for a customer service person, referring to fig. 6, the method may include:

S601: before customer service personnel conduct a conversation with a user, core input parameter information required by a content generation model is determined, wherein the core input parameter information comprises: dialogue content generated by the user in a current service session and/or historical dialogue content generated in an associated historical service session;

s602: and constructing a prompt text for dialogue with the content generation mode, inputting the core input parameter information into the content generation model, so that the content generation model extracts key questions according to the core input parameter information, analyzes user appeal, historical processing results and/or disputes of two parties of the dialogue, and generates recommended first-term contents, wherein the recommended first-term contents comprise descriptions of questions possibly required to be asked by the user and back-questions based on the summarized user appeal.

For the details of the second and third embodiments, which are not described in detail, reference may be made to the description of the first embodiment and other portions of the present specification, and the details are not repeated here.

It should be noted that, in the embodiment of the present application, the use of user data may be involved, and in practical application, the user specific personal data may be used in the solution described herein within the scope allowed by the applicable legal regulations in the country under the condition of meeting the applicable legal regulations in the country (for example, the user explicitly agrees to the user to notify practically, etc.).

Corresponding to the first embodiment, the embodiment of the application also provides a recommended conversation content generating device, which may include:

the dialogue content monitoring unit is used for monitoring dialogue content generated in the process of dialogue between customer service personnel and a user;

A recommended conversation generation unit configured to construct a prompt text for interacting with a content generation model and input dialogue content associated with the user into the content generation model if the customer service person performs an operation of sending a solution to the user for a first appeal of the user, so that the content generation model determines whether the user approves the solution or generates a second appeal/problem, and generates a corresponding recommended conversation;

And providing the recommended conversation to the customer service personnel so that the customer service personnel can carry out a dialogue with the user according to the recommended conversation.

Specifically, the recommended speaking is used for inquiring/answering to the user for the approval/disapproval condition of the solution or the second appeal/problem of the solution.

In particular, the apparatus may further include:

And the prejudging unit is used for prejudging the condition of the approval or disapproval of the solution by the user according to feedback information submitted by the user and approved or disapproval of the solution or keywords contained in dialogue content generated after the user receives the solution before the call of the content generating model, and inputting the feedback information or prejudging result information into the content generating model so that the content generating model refers to the process of generating the recommended speech.

Wherein, when constructing a prompt text for interacting with a content generation model according to the dialogue content associated with the user, the dialogue content associated with the user comprises: in the current dialogue process of the user and the customer service personnel, the solution transmits dialogue contents generated before and after the solution and historical dialogue contents generated in the historical consultation process of the user.

Specifically, the content generation model may be pre-trained by:

Extracting information of solution key nodes in the service flow execution specification;

Acquiring standard questions corresponding to various solutions, enumerating reasons that the solutions may be approved or disapproved, and respectively inquiring/answering to provide corresponding speech samples for various reasons;

collecting historical dialogue record information between a plurality of customer service personnel and a plurality of users;

Inputting the speaking samples and the historical dialogue record information into an artificial intelligent AI large language model, summarizing the speaking samples of the various solutions in various conditions by the AI large language model, carrying out service flow simulation on user problems and information states related in the historical dialogue record information according to the service flow execution specifications, determining reasonable solutions and the following questions/following questions, comparing simulation results with the solutions and the following questions/following questions actually given by the customer in the historical dialogue record, determining whether the solutions and the following questions/following questions actually given by the customer in the historical dialogue record are reasonable or not, and marking;

And pre-training the content generation model by using the marked data.

In addition, the apparatus may further include:

An input parameter information determining unit, configured to determine core input parameter information required by a content generation model before the customer service personnel performs a dialogue with a user, where the core input parameter information includes: dialogue content generated by the user in a current service session and/or historical dialogue content generated in an associated historical service session;

and the first question and first question recommended operation generating unit is used for constructing a prompt text for carrying out a dialogue with the content generating mode, inputting the core input parameter information into the content generating model, so that the content generating model extracts key questions according to the core input parameter information, and generates recommended first question and first question operation contents after analyzing user appeal, historical processing results and/or disputed points of the two parties of the dialogue, wherein the recommended first question and first question operation contents comprise descriptions of questions possibly required to be asked by the user and back questions based on the summarized user appeal.

Wherein the content generation model is further to: and carrying out emotion recognition on dialogue content associated with the user, and carrying out concentric color rendering processing on the generated recommended speech.

Corresponding to the embodiment, the embodiment of the application also provides a pretraining device of the content generation model, which can comprise:

The information extraction unit is used for extracting the information of the solution key nodes in the client service flow execution specification;

The speech surgery sample providing unit is used for acquiring standard questions corresponding to various solutions, enumerating reasons that the solutions may be approved or disapproved, and respectively inquiring/answering to provide corresponding speech surgery samples for various reasons;

A history dialogue content collection unit for collecting history dialogue content information between the customer and the user in a plurality of history service sessions;

The labeling unit is used for determining reasonable solutions and answering/answering operation samples by inputting the speaking samples and the historical dialogue content information into an artificial intelligent AI large language model, summarizing the speaking samples of the various solutions for answering/answering under various conditions by the AI large language model, carrying out standard service flow simulation on user problems and information states related in the historical dialogue record information according to the service flow execution specifications, determining whether the solutions and the answering/answering operation samples are reasonable or not by comparing simulation results with the solutions and the answering/answering operations actually given by customer service personnel in the historical dialogue content, and determining whether the solutions and the answering/answering operations actually given by the customer service personnel in the historical dialogue content are reasonable or not and labeling;

And the training unit is used for pre-training the content generation model by using the marked data so that after the customer service personnel provide the solution for the user aiming at the first appeal of the user, the content generation model determines whether the user approves the solution or generates the second appeal/problem according to the dialogue content associated with the user and generates a corresponding recommended conversation.

Corresponding to the embodiment, the embodiment of the application also provides a recommended conversation content generating device, which can comprise:

an input parameter information determining unit, configured to determine core input parameter information required by a content generation model before a customer service person performs a dialogue with a user, where the core input parameter information includes: dialogue content generated by the user in a current service session and/or historical dialogue content generated in an associated historical service session;

and the first question and first question recommended operation generating unit is used for constructing a prompt text for carrying out a dialogue with the content generating mode, inputting the core input parameter information into the content generating model, so that the content generating model extracts key questions according to the core input parameter information, and generates recommended first question and first question operation contents after analyzing user appeal, historical processing results and/or disputed points of the two parties of the dialogue, wherein the recommended first question and first question operation contents comprise descriptions of questions possibly required to be asked by the user and back questions based on the summarized user appeal.

In addition, the embodiment of the present application also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps of the method described in the foregoing method embodiment.

And an electronic device comprising:

One or more processors; and

A memory associated with the one or more processors for storing program instructions that, when read for execution by the one or more processors, perform the steps of the methods described in the foregoing method embodiments.

A computer program product comprising computer program/computer-executable instructions which, when executed by a processor in an electronic device, implement the steps of the method of the preceding method embodiments.

Fig. 7 illustrates an architecture of an electronic device, which may include a processor 710, a video display adapter 711, a disk drive 712, an input/output interface 713, a network interface 714, and a memory 720, among others. The processor 710, the video display adapter 711, the disk drive 712, the input/output interface 713, the network interface 714, and the memory 720 may be communicatively connected via a communication bus 730.

The processor 710 may be implemented by a general-purpose CPU (Central Processing Unit) or a microprocessor, an Application SPECIFIC INTEGRATED Circuit (ASIC), or one or more integrated circuits, etc. for executing related programs to implement the technical solution provided by the present application.

The Memory 720 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory ), static storage, dynamic storage, etc. The memory 720 may store an operating system 721 for controlling the operation of the electronic device 700, and a Basic Input Output System (BIOS) for controlling the low-level operation of the electronic device 700. In addition, a web browser 723, a data storage management system 724, a recommended-speech content generation system 725, and the like may also be stored. The recommended session content generation system 725 may be an application program that specifically implements the operations of the foregoing steps in the embodiment of the present application. In general, when the technical solution provided by the present application is implemented by software or firmware, relevant program codes are stored in the memory 720 and invoked by the processor 710 for execution.

The input/output interface 713 is used to connect with an input/output module to enable information input and output. The input/output module may be configured as a component in a device (not shown) or may be external to the device to provide corresponding functionality. Wherein the input devices may include a keyboard, mouse, touch screen, microphone, various types of sensors, etc., and the output devices may include a display, speaker, vibrator, indicator lights, etc.

The network interface 714 is used to connect communication modules (not shown) to enable communication interactions of the device with other devices. The communication module may implement communication through a wired manner (such as USB, network cable, etc.), or may implement communication through a wireless manner (such as mobile network, WIFI, bluetooth, etc.).

Bus 730 includes a path to transfer information between various components of the device (e.g., processor 710, video display adapter 711, disk drive 712, input/output interface 713, network interface 714, and memory 720).

It should be noted that although the above devices illustrate only the processor 710, the video display adapter 711, the disk drive 712, the input/output interface 713, the network interface 714, the memory 720, the bus 730, etc., the device may include other components necessary to achieve proper operation in an implementation. Furthermore, it will be appreciated by those skilled in the art that the apparatus may include only the components necessary to implement the present application, and not all of the components shown in the drawings.

From the above description of embodiments, it will be apparent to those skilled in the art that the present application may be implemented in software plus a necessary general hardware platform. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the embodiments or some parts of the embodiments of the present application.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for a system or system embodiment, since it is substantially similar to a method embodiment, the description is relatively simple, with reference to the description of the method embodiment being made in part. The systems and system embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

The above description of the method and the electronic device for generating recommended speech content provided by the present application applies specific examples to illustrate the principles and embodiments of the present application, and the above examples are only used to help understand the method and the core idea of the present application; also, it is within the scope of the present application to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the application.

Claims (12)

1. A recommended speech content generation method, characterized by comprising:

monitoring dialogue contents generated in the process of dialogue between customer service personnel and a user;

If the customer service person performs an operation of sending a solution to the user for a first appeal of the user, constructing a prompt text for interacting with a content generation model, and inputting dialogue content associated with the user into the content generation model, so that the content generation model determines whether the user approves the solution or generates a second appeal/problem, and generates a corresponding recommended speaking;

And providing the recommended conversation to the customer service personnel so that the customer service personnel can carry out a dialogue with the user according to the recommended conversation.

2. The method of claim 1, wherein the step of determining the position of the substrate comprises,

The recommended speaking is used for inquiring/answering the user for the approval/disapproval condition of the solution or the second appeal/problem of the solution.

3. The method of claim 1, wherein the step of determining the position of the substrate comprises,

Before the call to the content generation model is made, the method further comprises the following steps:

And according to feedback information submitted by the user and approved or not approved by the solution or through keywords contained in dialogue content generated after the user receives the solution, prejudging the approval or disapproval condition of the solution by the user, and inputting the feedback information or prejudging result information into the content generation model so that the content generation model can refer to the process of generating the recommended conversation.

4. The method of claim 1, wherein the step of determining the position of the substrate comprises,

When a prompt text for interacting with a content generation model is constructed according to the dialogue content associated with the user, the dialogue content associated with the user comprises: in the current dialogue process of the user and the customer service personnel, the solution transmits dialogue contents generated before and after the solution and historical dialogue contents generated in the historical consultation process of the user.

5. The method of claim 1, wherein the step of determining the position of the substrate comprises,

The content generation model is pre-trained by:

Extracting information of solution key nodes in the service flow execution specification;

Acquiring standard questions corresponding to various solutions, enumerating reasons that the solutions may be approved or disapproved, and respectively inquiring/answering to provide corresponding speech samples for various reasons;

collecting historical dialogue record information between a plurality of customer service personnel and a plurality of users;

Inputting the speaking samples and the historical dialogue record information into an artificial intelligent AI large language model, summarizing the speaking samples of the various solutions in various conditions by the AI large language model, carrying out service flow simulation on user problems and information states related in the historical dialogue record information according to the service flow execution specifications, determining reasonable solutions and the following questions/following questions, comparing simulation results with the solutions and the following questions/following questions actually given by the customer in the historical dialogue record, determining whether the solutions and the following questions/following questions actually given by the customer in the historical dialogue record are reasonable or not, and marking;

And pre-training the content generation model by using the marked data.

6. The method as recited in claim 1, further comprising:

Before the customer service personnel performs a dialogue with a user, determining core input parameter information required by a content generation model, wherein the core input parameter information comprises: dialogue content generated by the user in a current service session and/or historical dialogue content generated in an associated historical service session;

And constructing a prompt text for dialogue with the content generation mode, inputting the core input parameter information into the content generation model, so that the content generation model extracts key questions according to the core input parameter information, analyzes user appeal, historical processing results and/or disputes of two parties of the dialogue, and generates recommended first-term contents, wherein the recommended first-term contents comprise descriptions of questions possibly required to be asked by the user and back-questions based on the summarized user appeal.

7. The method according to claim 1 or 6, wherein,

The content generation model is further for: and carrying out emotion recognition on dialogue content associated with the user, and carrying out concentric color rendering processing on the generated recommended speech.

8. A method of pre-training a content generation model, comprising:

extracting information of solution key nodes in the client service flow execution specification;

Acquiring standard questions corresponding to various solutions, enumerating reasons that the solutions may be approved or disapproved, and respectively inquiring/answering to provide corresponding speech samples for various reasons;

Collecting historical dialogue content information between a customer and a user in a plurality of historical service sessions;

Inputting the speech samples and the historical dialogue content information into an artificial intelligent AI large language model, summarizing the speech samples of the various solutions in various conditions by the AI large language model, carrying out standard service flow simulation on user problems and information states related in the historical dialogue record information according to the service flow execution specifications, determining reasonable solutions and the speech samples of the following questions/following questions, and determining whether the solutions and the speech steps actually given by the customer service personnel in the historical dialogue content are reasonable or not and marking by comparing simulation results with the solutions and the speech steps actually given by the customer service personnel in the historical dialogue content;

and pre-training the content generation model by using the marked data so that after customer service personnel provide a solution for the user aiming at the first appeal of the user, the content generation model determines whether the user approves the solution or generates the second appeal/problem according to the dialogue content associated with the user and generates a corresponding recommended conversation.

9. A recommended speech content generation method, characterized by comprising:

before customer service personnel conduct a conversation with a user, core input parameter information required by a content generation model is determined, wherein the core input parameter information comprises: dialogue content generated by the user in a current service session and/or historical dialogue content generated in an associated historical service session;

And constructing a prompt text for dialogue with the content generation mode, inputting the core input parameter information into the content generation model, so that the content generation model extracts key questions according to the core input parameter information, analyzes user appeal, historical processing results and/or disputes of two parties of the dialogue, and generates recommended first-term contents, wherein the recommended first-term contents comprise descriptions of questions possibly required to be asked by the user and back-questions based on the summarized user appeal.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 9.

11. An electronic device, comprising:

One or more processors; and

A memory associated with the one or more processors for storing program instructions that, when read for execution by the one or more processors, perform the steps of the method of any of claims 1 to 9.

12. A computer program product comprising computer program/computer-executable instructions which, when executed by a processor in an electronic device, implement the steps of the method of any one of claims 1 to 9.