Dialogue management method, device and equipment

Abstract

The invention provides a dialogue management method, device and equipment. The dialogue management method provided by the invention comprises the following steps: after receiving input information of a user, inputting the input information into a demand identification model, so that the demand identification model performs demand identification on the input information, and outputs a demand identification result corresponding to the input information; inputting the demand identification result into a reply mode identification model, and outputting a reply mode corresponding to the demand identification result by the reply mode identification model; wherein the reply mode comprises manual reply or automatic reply; and replying the input information according to the reply mode. According to the dialogue management method, device and equipment provided by the invention, the flexibility of the dialogue process is improved, the problem solving time is shortened, and the user experience feeling is improved.

Description

Technical Field

[0001] This application relates to the technical field of natural language processing, and in particular, to a dialogue management method, apparatus, and device. Background Art

[0002] Intelligent customer service can automatically reply to input information, greatly reducing the burden on human customer service and improving the efficiency of dialogue answering. For example, on e-commerce platforms such as Taobao and JD.com, first, the intelligent customer service makes a preliminary answer to the questions input by users, and then provides a corresponding interface to transfer to human customer service to handle specific problems of users. However, in the dialogue platform, the existing technology requires that communication must first be carried out with the intelligent customer service, that is, regardless of the questions and requirements, an automatic reply method will be used for answering; then the user himself triggers the transfer to the human customer service according to further requirements, and the allocation is completed according to the idle state of the human customer service. In the above process, regardless of the requirements and questions, a specific process must be passed through, the dialogue method is relatively fixed, and the flexibility is poor; in the case where the automatic reply is far from the requirements, it increases and lengthens the time to solve the problem and reduces the user experience; in the case of complex requirements, the process of switching between automatic reply and human reply increases the complexity of the dialogue. Summary of the Invention

[0003] In view of this, this application provides a dialogue management method, apparatus, and device to solve the problems of cumbersome dialogue process, poor flexibility, long problem-solving time, and poor user experience in the existing technology.

[0004] Specifically, this application is implemented through the following technical solutions:

[0005] The first aspect of this application provides a dialogue management method, and the method includes:

[0006] After receiving the input information of the user, input the input information into a demand recognition model, so that the demand recognition model recognizes the demand of the input information and outputs a demand recognition result corresponding to the input information;

[0007] Input the demand recognition result into a reply method recognition model, so that the reply method recognition model outputs a reply method corresponding to the demand recognition result; wherein, the reply method includes manual reply or automatic reply;

[0008] Reply to the input information according to the reply method.

[0009] The second aspect of this application provides a dialogue management apparatus, and the apparatus includes an identification module and a processing module; wherein,

[0010] The recognition module is configured to input the received input information of the user into a requirement recognition model after receiving the input information of the user, so that the requirement recognition model performs requirement recognition on the input information and outputs a requirement recognition result corresponding to the input information;

[0011] The recognition module is further configured to input the requirement recognition result into a response mode recognition model, so that the response mode recognition model outputs a response mode corresponding to the requirement recognition result; wherein, the response mode includes manual response or automatic response;

[0012] The processing module is configured to respond to the input information according to the response mode.

[0013] A third aspect of the present application provides a dialogue management device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of any of the methods provided in the first aspect of the present application are implemented.

[0014] A fourth aspect of the present application provides a storage medium, on which a program is stored. When the program is executed by a processor, the steps of any of the methods provided in the present application are implemented.

[0015] The dialogue management method, device and equipment provided by the present application, after receiving the input information of the user, input the input information into a requirement recognition model, so that the requirement recognition model performs requirement recognition on the input information and outputs a requirement recognition result corresponding to the input information, and then input the requirement recognition result into a response mode recognition model, so that the response mode recognition model outputs a response mode corresponding to the requirement recognition result; wherein, the response mode includes manual response and automatic response, and thus respond to the input information according to the response mode. In this way, after the user expresses their own requirements and problems to be solved through the input information, the dialogue management party can automatically recognize the requirements in the input information based on the requirement recognition model, and match the corresponding response mode for the requirement through the response mode recognition model, so as to flexibly select automatic response or manual response according to the actual situation of the input information, and respond to the input statement according to the response mode that best matches the actual situation of the input information. From the perspective of the dialogue user, the response process of the dialogue is accelerated and simplified, and the user experience is improved; from the perspective of the dialogue management method, the flexibility of the dialogue is increased, the time of the dialogue is not prolonged, the congestion of multiple dialogue processes is avoided, the management efficiency of the dialogue platform is improved, there is no need to go through a fixed process of switching from automatic response to manual response, the flexibility of the dialogue is increased, the response process of the dialogue is accelerated and simplified, and the user experience is improved. Description of the Drawings

[0016] Figure 1Flowchart of the first embodiment of the dialogue management method provided by this application;

[0017] Figure 2 Flowchart of the second embodiment of the dialogue management method provided by this application;

[0018] Figure 3 Flowchart of the third embodiment of the dialogue management method provided by this application;

[0019] Figure 4 A hardware structure diagram of the dialogue management device provided by this application in the dialogue management device;

[0020] Figure 5 Schematic structural diagram of the first embodiment of the dialogue management device provided by this application. Detailed implementation manners

[0021] Here, the exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with this application. On the contrary, they are merely examples of devices and methods consistent with some aspects of this application as detailed in the appended claims.

[0022] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms "a", "the", and "said" used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0023] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to a determination".

[0024] This application provides a dialogue management method, device, and equipment to solve the problems of cumbersome dialogue process, poor flexibility, long problem-solving time, and poor user experience in the prior art.

[0025] The dialogue management method, device, and equipment provided by this application, after receiving the user's input information, input the input information into a demand recognition model, so that the demand recognition model recognizes the demand of the input information and outputs a demand recognition result corresponding to the input information. Then, input the demand recognition result into a response method recognition model, so that the response method recognition model outputs a response method corresponding to the demand recognition result; where the response method includes manual response and automatic response, and thus reply to the input information according to the response method. In this way, after the user expresses their own needs and problems to be solved through the input information, the dialogue management party can automatically recognize the demand in the input information based on the demand recognition model, and match a corresponding response method for this demand through the response method recognition model, so as to flexibly select automatic response or manual response according to the actual situation of the input information, and reply to the input statement according to the response method that best matches the actual situation of the input information. From the perspective of the dialogue user, the response process of the dialogue is accelerated and simplified, improving the user experience; from the perspective of the dialogue management method, the flexibility of the dialogue is increased, avoiding lengthening the dialogue time and causing congestion of multiple dialogue processes, improving the management efficiency of the dialogue platform, without going through the fixed process of switching from automatic response to manual response, increasing the flexibility of the dialogue, accelerating and simplifying the response process of the dialogue, and improving the user experience.

[0026] Specific embodiments are given below to introduce the technical solutions of this application in detail.

[0027] Figure 1 It is a flowchart of the first embodiment of the dialogue management method provided by this application. Please refer to Figure 1 , the method includes:

[0028] S101. After receiving the user's input information, input the input information into a demand recognition model, so that the demand recognition model recognizes the demand of the input information and outputs a demand recognition result corresponding to the input information.

[0029] Specifically, the user's input information can be information in text form or information in voice form, and in this embodiment, it is not limited. In addition, the input information can include keywords representing the user's demand. For example, the input information can include a clear object or a pronoun used to indicate the object.

[0030] For example, in one embodiment, the user's input information is: Please help me check how to get to Mall A. At this time, the input information includes a clear reference object (Mall A); for another example, in another embodiment, the user's input information can be: What should I do if the one I checked last time is lost? At this time, the input information includes an indicator word (that) used to refer to the object.

[0031] It should be noted that after the user's input information includes an indicator for indicating an object, after receiving the user's input information, the method may further include: identifying a pronoun in the input information, obtaining the object indicated by the pronoun from the historical conversation record based on the time prompt information, and using the object to replace the pronoun in the input information as the finally received input information.

[0032] Specifically, the time prompt information may be determined from the input information input by the user or be preset time prompt information.

[0033] Combined with the second example above, "that" in the above input information is a pronoun, and "last time" is the time prompt information. According to "last time", the object queried most recently in the historical conversation record is found, and it is queried that the object indicated by the pronoun is "headphone". The pronoun is replaced with the referred object to obtain the finally received input sentence "The headphone I queried last time is lost. What should I do?".

[0034] Furthermore, the requirement recognition model may be a large model pre-trained based on the transformer model. The requirement recognition model has a dual-channel encoder, which encodes the input content and output content of requirement recognition respectively, and obtains the context semantic information of the input content and the global semantic information of the standard requirement bidirectionally to obtain an accurate requirement recognition result.

[0035] Optionally, in one embodiment, the requirement recognition model may include a text encoder, a requirement encoder, and a classification module; the text encoder is used to obtain a first encoded representation of the input information; the requirement encoder is used to obtain a second encoded representation of the standard requirement; the classification module is used to obtain a requirement recognition result corresponding to the input information based on the first encoded representation and the second encoded representation.

[0036] When specifically implemented, when the user has corresponding requirements, the corresponding requirements are expressed through the input information. The requirement recognition model performs relevant processing on the user's input information and conducts requirement recognition, and finally obtains a requirement recognition result corresponding to the input information.

[0037] Specifically, the requirement recognition result is used to clarify the reason for the user to initiate a conversation. The requirement recognition result includes at least an object, a state, and / or a question. Among them, the object is used to describe the object targeted by the input information, the state is the current state of the object, and the question is the requirement for initiating the conversation. For example, in one embodiment, for the input information "I don't know how to go to X restaurant", where the object is me, the state is not knowing how to go, and the question is how to go to X restaurant; for another example, in another embodiment, the input information is "The headphone I queried last time is lost. What should I do?", where the object is the headphone, the state is lost, and the question is what to do.

[0038] Optionally, in one embodiment, the input information is input information in the form of speech. The process of the demand recognition model for recognizing the demand of the input information may include:

[0039] (1) Convert the input information in the form of speech into input information in the form of text.

[0040] (2) Recognize the demand of the input information in the form of text to obtain the demand recognition result of the input information.

[0041] In one embodiment, the demand recognition model may include a cascaded speech recognition model and a demand recognition sub-model. When the input information is input information in the form of speech, the input information in the form of speech may be first converted into input information in the form of text by the speech recognition model, and then the demand recognition sub-model may be used to recognize the demand of the input information in the form of text to obtain the demand recognition result of the input information.

[0042] It should be noted that the speech recognition model may be a recurrent neural network, a long short-term memory network, a transcription attention model, etc., which can map speech to text.

[0043] In addition, the demand recognition sub-model is pre-trained according to the demand recognition framework. In this embodiment, it is not limited herein. It should be noted that the demand recognition sub-model is used to process the input information in the form of text output by the speech recognition model for the speech recognition model to obtain the demand recognition result of the input information. Further, the demand recognition sub-model may be a pre-trained large model, and the large model refers to a neural network model with a relatively large number of parameters 。 For example, it may be GPT, BERT, etc.

[0044] For example, in one embodiment, the input information is input information in the form of speech. At this time, converting the input information in the form of speech into input information in the form of text is "I found a flaw in the clothes I bought yesterday. Can I return them?", and further, after recognizing the demand of the input information in the form of text, the object in the demand recognition result of the input information is "clothes", the state is "having a flaw", and the question is "Can I return them?".

[0045] S102. Input the demand recognition result into a reply mode recognition model so that the reply mode recognition model outputs the reply mode corresponding to the demand recognition result; wherein, the reply mode includes manual reply or automatic reply.

[0046] It should be noted that the reply mode recognition model is essentially a classification model, which determines the reply mode by classification.

[0047] Specifically, after inputting the requirement recognition result into the response mode recognition model, the response mode recognition model extracts features from the input information to obtain a feature map, and then determines the confidence levels of the feature map belonging to each preset response mode, and takes the preset response mode with the highest confidence level as the response mode corresponding to the requirement recognition result. Among them, the preset response modes include manual response and automatic response. That is, after recognizing the requirement recognition result, the response mode corresponding to the requirement recognition result is manual response or automatic response.

[0048] For example, in combination with the above example, the input information is: I don't know how to get to X Restaurant. At this time, after recognizing the requirement recognition result of the input information, it is determined that the response mode corresponding to the requirement recognition result is automatic response.

[0049] The dialogue management method provided in this embodiment can automatically match the best response mode corresponding to the input information according to the requirement recognition result, avoid manual switching, and also avoid the process of defaulting to automatic response, improving the efficiency of the dialogue and the sense of dialogue experience.

[0050] S103. Respond to the input information according to the response mode.

[0051] Specifically, after determining the response mode, respond to the input information according to this response mode. For example, if the determined response mode is manual response, the input information is pushed to the target responder to obtain the response content manually determined by the target responder; for another example, if the determined response mode is automatic response, the response content of the input information is automatically generated and the response content is returned to the dialogue party where the input information is located to complete the current round of dialogue.

[0052] The method provided in this embodiment, after receiving the input information of the user, inputs the input information into a requirement recognition model, so that the requirement recognition model recognizes the requirements of the input information and outputs a requirement recognition result corresponding to the input information. Then, the requirement recognition result is input into a response method recognition model, so that the response method recognition model outputs a response method corresponding to the requirement recognition result; wherein, the response method includes manual response and automatic response, and thus the input information is responded to according to the response method. In this way, after the user expresses their requirements and problems to be solved through the input information, the dialogue management party can automatically recognize the requirements in the input information based on the requirement recognition model, and match a corresponding response method for the requirement through the response method recognition model, so as to flexibly select automatic response or manual response according to the actual situation of the input information, and respond to the input statement according to the response method that best matches the actual situation of the input information. From the perspective of the dialogue user, the response process of the dialogue is accelerated and simplified, and the user experience is improved; from the perspective of the dialogue management method, the flexibility of the dialogue is increased, the time of the dialogue is not prolonged, the congestion of multiple dialogue processes is avoided, the management efficiency of the dialogue platform is improved, there is no need to go through a fixed process from automatic response to manual response, the flexibility of the dialogue is increased, the response process of the dialogue is accelerated and simplified, and the user experience is improved.

[0053] Figure 2 This is the flowchart of the second embodiment of the dialogue management method provided in this application. Please refer to Figure 2 , when the response method is an automatic response, the step of responding to the input information according to the response method may include:

[0054] S201. Input the input information into an intelligent question and answer model, so that the intelligent question and answer model outputs an automatic response content for the input information.

[0055] It should be noted that when the intelligent question and answer model determines the automatic response content based on the input information, it can directly generate a corresponding automatic response content based on the input information. For example, in one embodiment, the intelligent question and answer model may be a generative model constructed based on the transformer model. At this time, the automatic response content can be directly generated based on the intelligent question and answer model.

[0056] In addition, when the intelligent question and answer model determines the automatic response content based on the input information, it can also search for the automatic response content corresponding to the input information from a preset knowledge base. In this embodiment, it is not limited thereto. The preset knowledge base is set according to actual needs and is not limited in this embodiment.

[0057] For example, in a possible implementation, after the input information is input into the intelligent Q&A model, when the intelligent Q&A model determines the automatic reply content, it can calculate the similarity between the demand recognition result and the automatic reply template, and then select the automatic reply template with the highest similarity as the automatic reply content for output.

[0058] Specifically, multiple automatic reply templates can be preset for the intelligent Q&A model to output the automatic reply content based on the automatic reply templates. Among them, the preset multiple automatic reply templates are set according to actual needs, and are not limited in this embodiment. For example, in one embodiment, the preset multiple automatic reply templates may include automatic reply templates of different service types. Among them, the service type includes at least one of the following: for example, pre-sales, after-sales, logistics, etc.

[0059] In specific implementation, the target service type of the demand recognition result can be determined first, and then the automatic reply template with the highest similarity can be determined from multiple alternative automatic reply templates whose service type is the target service type (the multiple alternative automatic reply templates are the automatic reply templates in the preset multiple automatic reply templates whose service type is the target service type) as the automatic reply content for output.

[0060] It should be noted that the intelligent Q&A model can be a pre-trained large model. A large model generally refers to a deep learning model with a large number of parameters and a deep number of layers. This intelligent Q&A model has hundreds of millions or billions of parameters and performs excellently in understanding language context and processing complex semantic relationships. For example, in one embodiment, the intelligent Q&A model can be a BERT model, GPT, etc.

[0061] S202. Reply the automatic reply content to the user.

[0062] In specific implementation, the automatic reply content can be replied to the user in the form of text or voice, which is not limited in this embodiment.

[0063] Optionally, in a possible implementation, before replying the automatic reply content to the user, the method further includes:

[0064] (1) Obtain the context of the input information and / or the role information of the user.

[0065] Specifically, the context of the input information can be obtained by shifting the input information forward by a preset forward range and backward by a preset backward range. The preset range is set according to actual needs and is not limited in this embodiment. For example, in one embodiment, the content of the upper and lower 5 lines of the input information can be extracted as the context of the input information; for another example, in another embodiment, the content of the upper and lower 3 lines of the input information can be extracted as the context of the input information.

[0066] Of course, in a possible implementation, the role information of the user can also be obtained. It should be noted that, in one embodiment, if the role information is recorded in the user's attribute information, the role information of the user can be directly obtained from the user's attribute information; in another embodiment, if the role information is not recorded in the user's attribute information, the role information of the user can be determined based on the input information and / or the historical conversation record. For example, in one embodiment, the user's input information is "I am very interested in the new features of product Y. Can you provide more detailed information?", and at this time, the role information of the user can be determined as a buyer.

[0067] (2)Polish the automatic reply content according to the context and / or role information.

[0068] Polishing is to make meticulous modifications to the text to improve its quality, fluency, and expressive effect. In specific implementation, for example, in one embodiment, a trained polishing model can be used to polish the automatic reply content (specifically, the polishing model can also be a large model, for example, using GPT for polishing). Another example is that when polishing the automatic reply content based on the context, the context can be subjected to intent recognition to determine the user's intent, and then the automatic reply content can be polished based on the intent to make the automatic reply content better meet the intent. Another example is that when polishing based on both the context and the role information, the purpose of the automatic reply can be clarified according to the context, the audience that the automatic reply content is targeted at can be determined based on the role information, and then the tone of the automatic reply content can be adjusted based on the purpose and the audience.

[0069] Combined with the above example, when the role information of the user is determined to be a buyer, after polishing the automatic reply content at this time, the obtained automatic reply content can be "Of course! Thank you very much for your interest in the new features of product Y. You can obtain more detailed information on our official website or by contacting the customer service. If you have any other questions, we will be happy to help you at any time."

[0070] It should be noted that before replying the automatic reply content to the user, by obtaining the context of the input information and / or the role information of the user, and then polishing the automatic reply content according to the context and / or role information. In this way, the readability, accuracy, etc. of the automatic reply content can be improved, so as to further improve the user experience.

[0071] The method provided in this embodiment, when determining that the reply method is an automatic reply, determines the automatic reply content through an intelligent question-and-answer model, and then replies the automatic reply content to the user. In this way, the workload of manual reply can be reduced, and at the same time, the accuracy and semantic relevance of the automatic reply are improved, further improving the user experience and the efficiency of dialogue management.

[0072] Figure 3 This is the flowchart of the third embodiment of the dialogue management method provided by this application. Please refer to Figure 3 , when the reply mode is manual reply, replying to the input information according to the reply mode may include:

[0073] S301. Generate alternative reply content for the input information based on a generative model.

[0074] It should be noted that alternative reply content for the input information can be generated based on a generative model, so as to provide alternative reply content for the replier as a reference when replying manually, improving the efficiency and accuracy of manual reply.

[0075] In addition, the generative model can be a pre-trained large model. For example, it can be GPT, etc.

[0076] S302. Obtain the context of the input information, and extract the context and the summary of the input information.

[0077] Referring to the previous description, the context of the input information can be obtained by shifting the input information forward by a preset forward range and backward by a preset backward range.

[0078] Specifically, the context and the input information can be regarded as a whole. For the convenience of description, the overall text composed of the context and the input information is denoted as the target text. For example, in one embodiment, keywords and / or key sentences of the target text are extracted, and then a summary is constructed based on the keywords and / or key sentences. For another example, topic modeling technology can be used to determine the topic of the target text, and then a summary is generated according to the topic. For still another example, the summary of the target text can be extracted based on a summary extraction model.

[0079] S303. Select a target replier from multiple repliers.

[0080] Specifically, there are multiple repliers provided in the dialogue platform, and one replier can be selected from the multiple repliers as the target replier. For example, in one embodiment, the replier with the least amount of replies to be processed among the multiple repliers can be used as the target replier; for another example, in another embodiment, the repliers can also be sorted according to their praise rates, and the replier with the highest praise rate can be selected as the target replier; for still another example, a replier can be randomly selected from the multiple repliers as the target replier.

[0081] It should be noted that specifically, selecting a target replier from multiple repliers includes:

[0082] (1) Determine the service type corresponding to the summary.

[0083] As can be seen from the above description, the input information entered by the user can correspond to different service types. For example, in one embodiment, the summary of the input information obtained from the user is: Where is it? (The complete one is: Where is the express delivery?), and the service type corresponding to this summary is logistics. For example, in another embodiment, the summary of the input information obtained from the user is: There is damage (The complete one is: I have received the item and found that there is damage), and the service type corresponding to this summary is after-sales. For example, in yet another embodiment, the summary of the input information obtained from the user is: What size should I buy? (The complete one is: I am XXX tall. What size should I buy?), and the service type corresponding to this summary is pre-sales.

[0084] (2) According to the reply type tags of the multiple respondents, find multiple candidate respondents that match the service type from the multiple respondents.

[0085] The reply type tag is used to represent the service type corresponding to the respondent, that is, what service type of conversation the respondent can handle. In specific implementation, each respondent can be set to reply to conversations of at least one service type. For example, in one embodiment, there are multiple respondents provided in the conversation platform. Specifically, it includes respondent A, respondent B, respondent C, respondent D, and respondent E. Among them, the reply type tag of respondent A is pre-sales, the reply type tag of respondent B is logistics, the reply type tag of respondent C is after-sales, the reply type tag of respondent D is pre-sales, and the reply type tag of respondent E is pre-sales.

[0086] Combined with the above third example (the summary of the user's input information is: What size should I buy?), when it is determined that the service type corresponding to the summary is pre-sales, in this step, it is determined that respondent A, respondent D, and respondent E are candidate respondents.

[0087] (3) Determine the first respondent among the multiple candidate respondents as the target respondent; where the first respondent is the candidate respondent with the least number of messages to be replied among the multiple candidate respondents.

[0088] It should be noted that the conversation platform will count the number of messages to be replied for each respondent. Among them, the number of messages to be replied is the number of input information that has been assigned to the respondent and is waiting for the respondent to reply. In this way, the candidate respondent with the least number of messages to be replied can be selected as the target respondent according to the counted number of messages to be replied of the candidate respondents.

[0089] Combined with the example in step (2), after it is determined that respondent A, respondent D, and respondent E are candidate respondents, for example, in one embodiment, the number of messages to be replied of respondent A is 10, the number of messages to be replied of respondent D is 15, and the number of messages to be replied of respondent E is 13. At this time, respondent A is determined as the target respondent.

[0090] S304: Push the summary and the alternative reply content to the target reply party, so that the target reply party determines the manual reply content of the input information based on the summary and the alternative reply content, and replies to the user according to the manual reply content.

[0091] Specifically, after the summary, target respondent, and alternative reply content are determined, the summary and alternative reply content are pushed to the target respondent. In this way, after receiving this information, the target respondent can determine whether the alternative reply content is appropriate based on the summary, and modify the alternative reply content to obtain manual reply content, or manually edit the manual reply content based on the summary and alternative reply content.

[0092] Furthermore, after obtaining the manual reply content, the user is replied according to the manual reply content, for example, the manual reply content is presented to the user in the form of text or voice.

[0093] For example, in one embodiment, the summary is: damaged. The alternative responses are: Dear customer, we are very sorry that you received a damaged product. We understand that this has caused you trouble and we are deeply sorry for this; we are very sorry that the product you received is damaged. In order to better help you, please provide some details, such as the specific damaged parts and photos, so that we can deal with this matter quickly; we attach great importance to your feedback. We have recorded your problem and will take measures to resolve it as soon as possible. We will ensure that you get a perfect product.

[0094] At this point, the target responding party determines the manual response content of the input information based on the summary and alternative response content, which can be: Dear customer, we are very sorry that you received a damaged product. In order to better help you, please provide some details, such as the specific damaged parts and photos. In order to solve this problem, we will resend you a complete product, or refund you according to your wishes, or we are willing to provide you with some additional benefits or discounts. We hope that you can experience our better service again.

[0095] The method provided in this embodiment, when determining that the reply method is manual reply, can preferentially generate alternative reply content for the input information based on the generative model, obtain the context of the input information, extract the context, and the summary of the input information, and then select a target reply party from multiple reply parties, and finally push the summary and the alternative reply content to the target reply party. In this way, when the reply method is manual reply, by determining the summary and the alternative reply content, and then pushing the determined summary and the alternative reply content to the reply party, a reference can be provided for the reply party, so that the reply party can determine the manual reply content based on this, which can not only improve the efficiency of manual reply, but also further improve the user experience.

[0096] Corresponding to the foregoing embodiments of a dialogue management method, the present application also provides an embodiment of a dialogue management device.

[0097] The embodiment of the dialogue management device of the present application can be applied to a dialogue management device. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of the dialogue management device where it is located reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. From the hardware level, as Figure 4 shown, it is a hardware structure diagram of the dialogue management device where the dialogue management device provided by the present application is located. In addition to Figure 4 the shown processor, memory, network interface, and non-volatile memory, the dialogue management device where the device is located in the embodiment usually includes other hardware according to the actual functions of the dialogue management device, which will not be elaborated here.

[0098] Figure 5 Please refer to Figure 5 for the structural schematic diagram of the first embodiment of the dialogue management device provided by the present application. The device includes an identification module 510 and a processing module 520; wherein,

[0099] The identification module 510 is configured to, after receiving the input information of the user, input the input information into a demand identification model, so that the demand identification model performs demand identification on the input information and outputs a demand identification result corresponding to the input information;

[0100] The identification module 510 is further configured to input the demand identification result into a reply mode identification model, so that the reply mode identification model outputs a reply mode corresponding to the demand identification result; wherein, the reply mode includes manual reply or automatic reply;

[0101] The processing module 520 is configured to reply to the input information according to the reply mode.

[0102] The device provided in this embodiment can be used to execute the steps of the method shown in Figure 1 , and its implementation principle and implementation process are similar to those described above, which will not be elaborated here.

[0103] Optionally, when the input information is input information in voice form, the process of the demand identification model performing demand identification on the input information includes:

[0104] Converting the input information in voice form into input information in text form;

[0105] Perform requirement recognition on the input information in text form to obtain the requirement recognition result of the input information.

[0106] Optionally, when the reply mode is an automatic reply, the processing module 520 is configured to input the input information into an intelligent question and answer model, so that the intelligent question and answer model outputs the automatic reply content of the input information;

[0107] The processing module 520 is further configured to reply the automatic reply content to the user.

[0108] Optionally, before replying the automatic reply content to the user, the processing module 520 is further configured to obtain the context of the input information and / or the role information of the user;

[0109] The processing module 520 is further configured to polish the automatic reply content according to the context and / or role information.

[0110] Optionally, when the reply is a manual reply, the processing module 520 is further configured to generate alternative reply content of the input information based on a generative model;

[0111] The processing module 520 is further configured to obtain the context of the input information, and extract the context and the summary of the input information;

[0112] The processing module 520 is further configured to select a target responder from multiple responders;

[0113] The processing module 520 is further configured to push the summary and the alternative reply content to the target responder, so that the target responder determines the manual reply content of the input information based on the summary and the alternative reply content, and replies to the user according to the manual reply content.

[0114] Optionally, the processing module 520 is further configured to determine the service type corresponding to the summary;

[0115] The processing module 520 is further configured to search for multiple candidate responders matching the service type from the multiple responders according to the reply type tags of the multiple responders;

[0116] The processing module 520 is further configured to determine the first responder among the multiple candidate responders as the target responder; wherein, the first responder is the candidate responder with the least amount of pending replies among the multiple candidate responders.

[0117] Please continue to refer to Figure 4, this application also provides a dialogue management device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the methods provided in the first aspect of this application.

[0118] This application also provides a storage medium, on which a program is stored. When the program is executed by a processor, it implements the steps of any of the methods provided in this application.

[0119] The implementation process of the functions and roles of each unit in the above device can be specifically referred to the implementation process of the corresponding steps in the above method, which will not be elaborated here.

[0120] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial descriptions of the method embodiments. The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0121] The above are only the preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this application shall be included within the scope of protection of this application.

Claims

1. A dialogue management method, characterized in that, The method includes: After receiving the input information of the user, input the input information into a demand recognition model, so that the demand recognition model recognizes the demand of the input information and outputs a demand recognition result corresponding to the input information; Input the demand recognition result into a response mode recognition model, so that the response mode recognition model outputs a response mode corresponding to the demand recognition result; wherein, the response mode includes manual response or automatic response; Reply to the input information according to the response mode.

2. The method according to claim 1, characterized in that, The input information is input information in voice form. The process of the demand recognition model recognizing the demand of the input information includes: Convert the input information in voice form into input information in text form; Recognize the demand of the input information in text form to obtain a demand recognition result of the input information.

3. The method according to claim 1, wherein When the response mode is automatic response, the step of replying to the input information according to the response mode includes: Input the input information into an intelligent question and answer model, so that the intelligent question and answer model outputs an automatic response content of the input information; Reply the automatic response content to the user.

4. The method according to claim 3, characterized in that, Before replying the automatic response content to the user, the method further includes: Obtain the context of the input information and / or the role information of the user; Polish the automatic response content according to the context and / or role information.

5. The method according to claim 1, characterized in that, When the response mode is manual response, the step of replying to the input information according to the response mode includes: Generate alternative response content of the input information based on a generative model; Obtain the context of the input information, and extract the context and the summary of the input information; Select a target responder from multiple responders; Push the summary and the alternative response content to the target responder, so that the target responder determines the manual response content of the input information based on the summary and the alternative response content, and replies to the user according to the manual response content.

6. The method according to claim 5, wherein The step of selecting a target responder from multiple responders includes: Determine the service type corresponding to the summary; According to the response type tags of the multiple responders, find multiple candidate responders that match the service type from the multiple responders; Determine the first responder among the multiple candidate responders as the target responder; wherein, the first responder is the candidate responder with the least number of responses to be replied among the multiple candidate responders.

7. A dialogue management device, characterized in that, The device includes an identification module and a processing module; wherein, The identification module is configured to, after receiving the input information of the user, input the input information into a demand recognition model, so that the demand recognition model recognizes the demand of the input information and outputs a demand recognition result corresponding to the input information; The identification module is further configured to input the demand recognition result into a response mode recognition model, so that the response mode recognition model outputs a response mode corresponding to the demand recognition result; wherein, the response mode includes manual response or automatic response; The processing module is configured to reply to the input information according to the response mode.

8. The device according to claim 7, characterized in that, When the input information is input information in voice form, the process of the demand recognition model for recognizing the demand of the input information includes: Converting the input information in voice form into input information in text form; Performing demand recognition on the input information in text form to obtain a demand recognition result of the input information.

9. A dialogue management device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method according to any one of claims 1-6 are implemented.

10. A storage medium, on which a program is stored, characterized in that, When the program is executed by the processor, the steps of the method according to any one of claims 1-6 are implemented.

Images