CN115934901A - Intelligent conversation method and device, electronic equipment and storage medium - Google Patents
Intelligent conversation method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN115934901A CN115934901A CN202111388664.6A CN202111388664A CN115934901A CN 115934901 A CN115934901 A CN 115934901A CN 202111388664 A CN202111388664 A CN 202111388664A CN 115934901 A CN115934901 A CN 115934901A
- Authority
- CN
- China
- Prior art keywords
- conversation
- dialog
- behavior
- intelligent
- customer service
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 84
- 238000003860 storage Methods 0.000 title claims abstract description 18
- 230000006854 communication Effects 0.000 claims abstract description 65
- 238000004891 communication Methods 0.000 claims abstract description 54
- 238000012545 processing Methods 0.000 claims abstract description 38
- 230000006399 behavior Effects 0.000 claims description 166
- 239000013598 vector Substances 0.000 claims description 28
- 230000008569 process Effects 0.000 claims description 23
- 238000012549 training Methods 0.000 claims description 18
- 230000009471 action Effects 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 230000002457 bidirectional effect Effects 0.000 claims description 4
- 230000007704 transition Effects 0.000 claims description 4
- 238000013473 artificial intelligence Methods 0.000 abstract description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 14
- 238000010801 machine learning Methods 0.000 description 13
- 230000006870 function Effects 0.000 description 8
- 230000000694 effects Effects 0.000 description 6
- 239000007787 solid Substances 0.000 description 6
- 230000008451 emotion Effects 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 239000000428 dust Substances 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000003997 social interaction Effects 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002996 emotional effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000036651 mood Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present disclosure relates to a method, an apparatus, an electronic device and a storage medium for intelligent dialogue, which can be applied to the fields of computers and artificial intelligence, wherein the method comprises: recording the conversation content during the communication process of the conversation participants; inputting the conversation content into a conversation strategy auxiliary model for processing, wherein the output information comprises: recommending information of the conversation behavior at the next moment; and providing a conversation behavior prompt at the next moment for at least one of the conversation participants according to the conversation behavior recommendation information. The conversation behavior prompt is beneficial to improving the communication efficiency, and simultaneously, the prompted current object can communicate with the communication object in a precise, rapid and appropriate mode, so that the satisfaction degree of the communication object to the current object is improved; the method has wide applicability and practical significance, and can be applied to customer service scenes or social scenes and the like.
Description
Technical Field
The present disclosure relates to the field of computer and artificial intelligence technologies, and in particular, to a method and an apparatus for intelligent dialog, an electronic device, and a storage medium.
Background
In various communication conversation scenes, artificial intelligence technology is helpful for improving the efficiency of communication. Taking a consumption purchase scene as an example, in order to meet the needs of pre-sale and post-sale consultation of users, an operator is provided with a corresponding customer service department, and along with the continuous development of deep learning, the semantic analysis effect of a natural language processing model is greatly improved.
Most current customer service systems use deep learning models to semantically analyze a user's question and help generate matching answers by learning a large amount of historical data (e.g., question-answer pairs). From the design concept and function point of view, most customer service systems are configured to answer user questions through deep learning, and still have differences in real person-to-person communication, thereby possibly resulting in low customer satisfaction evaluation on the quality of service of the customer service system; in addition, the customer service system is only limited to provide reference answers during reply for assisting the manual customer service, and has a very limited effect on improving the customer service quality of the manual customer service.
Disclosure of Invention
To solve the technical problem or at least partially solve the technical problem, embodiments of the present disclosure provide a method, an apparatus, an electronic device, and a storage medium for intelligent dialog.
In a first aspect, an embodiment of the present disclosure provides a method of intelligent dialog. The method comprises the following steps: recording the conversation content during the communication process of the conversation participants; inputting the conversation content into a conversation strategy auxiliary model for processing, wherein the output information comprises: conversation behavior recommendation information at the next moment; and providing a conversation behavior prompt of the next moment for at least one of the conversation participants according to the conversation behavior recommendation information.
According to an embodiment of the present disclosure, the dialog policy assistance model includes: an encoding layer, a context prediction layer, and a conversational behavior prediction layer. Wherein, the inputting the conversation content into the conversation strategy auxiliary model for processing comprises: inputting the conversation content into a conversation strategy auxiliary model, coding sentences in the conversation content by the coding layer, and outputting to obtain a sentence vector for representing semantic information; inputting the sentence vectors into the context prediction layer in sequence according to the dialog occurrence sequence of the dialog contents, and processing the sentence vectors by the context prediction layer to generate hidden variables of the dialog at the next time corresponding to each sentence vector; and inputting the hidden variables into the dialogue behavior prediction layer, determining a global optimal solution in the dialogue behaviors corresponding to the hidden variables in a plurality of predefined dialogue behaviors by the dialogue behavior prediction layer, and outputting the global optimal solution in the dialogue behaviors as dialogue behavior recommendation information.
According to an embodiment of the present disclosure, the determining, by the dialogue behavior prediction layer, a global optimal solution in the dialogue behaviors corresponding to the hidden variables in a predefined plurality of dialogue behaviors includes: and taking the hidden variables as the hidden states of the dialogue behavior prediction layer, taking a plurality of predefined dialogue behaviors as the observation states of the dialogue behavior prediction layer, and calculating by the dialogue behavior prediction layer based on a conditional random field algorithm according to the emission probability between the hidden states and the observation states and the transition probability between the observation states to determine the global optimal solution in the dialogue behaviors.
According to an embodiment of the present disclosure, the dialog policy assistance model further includes: and decoding the layer. The above inputting the conversation content into the conversation strategy auxiliary model for processing further includes: and inputting the hidden variable into the decoding layer, and decoding by the decoding layer to obtain recommended dialect information of the next time.
According to an embodiment of the present disclosure, the output information further includes: recommended dialect information at the next time. Wherein, the method further comprises: and providing a speech prompt of the next moment for at least one of the conversation participants according to the recommended speech information.
According to an embodiment of the present disclosure, the method further includes: and performing automatic reply according to the dialogue action prompt and the dialogue operation prompt.
According to an embodiment of the present disclosure, the encoding layer and the decoding layer include: a bidirectional long-short term memory model Bi-LSTM, a denoising self-encoder BART, a unified pre-training language model UniLM or an autoregressive language model GPT; the context prediction layer includes: a bidirectional long-short term memory model Bi-LSTM; the dialog behavior prediction layer includes: conditional random field model CRF.
According to an embodiment of the present disclosure, the recording of the occurring conversation content in the process of communicating between the conversation participants includes: during the communication process of the conversation participants in the target conversation scene, the occurring conversation content is recorded through a recording record or a chatting interface. Wherein the target dialog scenario includes at least one of the following dialog scenarios: a person-to-person conversation scenario, a person-to-machine conversation scenario, said conversation scenario comprising: a customer service consultation scenario or a social scenario.
In a second aspect, embodiments of the present disclosure provide an apparatus for intelligent dialog. The above-mentioned device includes: the device comprises a conversation recording module, a processing module and a prompting module. The conversation recording module is used for recording the conversation content in the process of communication of the conversation participants. The processing module is used for inputting the conversation content into the conversation strategy auxiliary model for processing, and the output information comprises: and recommending information for the conversation behavior at the next moment. The prompt module is used for providing a conversation behavior prompt at the next moment for at least one of the conversation participants according to the conversation behavior recommendation information.
According to an embodiment of the present disclosure, the device includes an intelligent customer service system, and the intelligent customer service system includes the session recording module, the processing module, and the prompt module. Wherein, one of the conversation participants is a manual customer service or a customer service robot; the prompt module is used for providing a conversation behavior prompt at the next moment for the manual customer service or the customer service robot according to the conversation behavior recommendation information.
According to an embodiment of the present disclosure, the apparatus includes an intelligent dialogue system, and the intelligent dialogue system includes the dialogue recording module, the processing module, and the prompt module. Wherein one of the conversation participants is a user using the intelligent conversation system; the prompt module is used for providing dialog behavior prompt at the next moment for the user according to the dialog behavior recommendation information. Or, wherein the device is an intelligent robot, and one of the participants of the conversation is the intelligent robot; the prompt module is used for providing a conversation behavior prompt of the next moment for the intelligent robot according to the conversation behavior recommendation information; and the intelligent robot carries out conversation with the communication object according to the conversation behavior prompt.
In a third aspect, embodiments of the present disclosure provide an electronic device. The electronic equipment comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory finish mutual communication through the communication bus; a memory for storing a computer program; and the processor is used for realizing the intelligent conversation method when executing the program stored in the memory.
In a fourth aspect, embodiments of the present disclosure provide a computer-readable storage medium. The computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the method of intelligent dialog as described above.
The technical scheme provided by the embodiment of the disclosure at least has part or all of the following advantages:
(1) The conversation strategy prompting method comprises the steps that generated conversation contents are input into a conversation strategy auxiliary model to be processed, obtained output information comprises conversation behavior recommendation information at the next moment, and conversation behavior prompts at the next moment are provided for at least one of conversation participants according to the conversation behavior recommendation information at the next moment, so that corresponding replies can be developed by referring to the conversation behavior prompts before a prompted current object carries out conversation, the conversation behavior prompts contribute to improving communication efficiency, and meanwhile, the prompted current object can communicate with a communication object in a precise, rapid and appropriate mode, and the satisfaction degree of the communication object to the current object is promoted;
(2) The method has wide applicability and practical significance, can be applied to customer service scenes or social scenes and the like, and can comprise the following steps in the broad sense of social contact: the communication in the search and rescue task is beneficial to improving the service quality of customer service and the normalization of the service process, and improving the social popularity, the success rate of making friends, the success rate of rescuing the rescued object and the like. For example, when the method is applied to a customer service scene or a social scene, when a and B communicate, for example, a is a consumer and B is a customer service person/a customer service robot; or A and B are two persons who socialize/communicate/make friends, etc.; or A is a person, B is an intelligent robot for carrying out conversation, search and rescue and the like; in each implementation scenario, the device/system/equipment used by B or B itself (in the case of an intelligent robot) can determine, according to the content of the conversation that has occurred, behavior recommendation information to be used for communication at the next time, including one of the following behavior information: the method comprises the following steps of calling, introducing self, soothing the emotion of the opposite side, asking back, expressing thank you, inviting evaluation, actively marketing and the like, and providing conversation behavior prompts for B according to the information, wherein the B can carry out conversation at the next moment by referring to the conversation behavior prompts, so that the communication satisfaction of A to B can be improved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure.
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the related art will be briefly described below, and it is obvious for a person skilled in the art to obtain other drawings without inventive labor.
Fig. 1A and 1B schematically illustrate two exemplary system architectures, respectively, of a method of intelligent dialog suitable for use with embodiments of the present disclosure;
FIG. 2 schematically illustrates a flow diagram of a method of intelligent dialog in accordance with an embodiment of the present disclosure;
FIG. 3 schematically shows a structural diagram of a dialog strategy assistance model according to an embodiment of the present disclosure;
fig. 4 schematically shows a detailed implementation flowchart of operation S202 according to an embodiment of the present disclosure;
fig. 5 schematically shows an implementation process diagram of operation S202 according to an embodiment of the present disclosure;
fig. 6 schematically shows an implementation process schematic of operation S403 according to an embodiment of the present disclosure;
FIG. 7 schematically illustrates a flow diagram of a method of intelligent dialog in accordance with another embodiment of the present disclosure;
FIG. 8 schematically illustrates a block diagram of an apparatus for intelligent dialog according to an embodiment of the present disclosure; and
fig. 9 schematically shows a block diagram of an electronic device provided in an embodiment of the present disclosure.
Detailed Description
The embodiment of the disclosure provides an intelligent conversation method, an intelligent conversation device, electronic equipment and a storage medium. The method comprises the following steps: recording the conversation content during the communication process of the conversation participants; inputting the conversation content into a conversation strategy auxiliary model for processing, wherein the output information comprises: recommending information of the conversation behavior at the next moment; and providing a conversation behavior prompt of the next moment for at least one of the conversation participants according to the conversation behavior recommendation information.
To make the objects, technical solutions and advantages of the embodiments of the present disclosure more clear, the technical solutions of the embodiments of the present disclosure will be described clearly and completely with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are some embodiments of the present disclosure, but not all embodiments. All other embodiments, which can be derived by one of ordinary skill in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.
Fig. 1A and 1B schematically illustrate two exemplary system architectures, respectively, of methods and apparatus for intelligent dialog suitable for use with embodiments of the present disclosure.
Referring to FIG. 1A, an exemplary system architecture 110 suitable for use with the method of intelligent dialog of an embodiment of the present disclosure includes: a customer premises equipment 111 and a means for intelligent dialogue 112. The system architecture 110 may be applied to customer service scenarios.
The customer premise equipment 111 has a customer service interface, through which the user 101a (e.g. a consumer) can consult a question, such as pre-sale or post-sale consultation. Various communication client applications, such as a shopping application, a social platform software, an instant messaging tool, a web browser application, a search application, a mailbox client, etc. (for example only), may also be installed on the client device 111.
In an embodiment, referring to the single arrow in fig. 1A, the apparatus 112 for intelligent dialog includes an intelligent customer service system 1121, wherein the intelligent customer service system 1121 includes a customer service robot; in addition, the manual customer service system and the intelligent customer service system including the customer service robot are independent two systems (not shown in fig. 1A). The customer service robot replies to the user consultation problem in the customer service operation interface based on the preset logic.
Referring to a dialog scenario a1 illustrated in fig. 1A, the dialog scenario a1 occurs between a user 101A and a customer service robot, the user 101A may perform a dialog with the customer service robot through a customer service operation interface on a user end device 111, and an intelligent customer service system 1121 in the device 112 of the above intelligent dialog provides a dialog behavior prompt for the customer service robot at the next time by executing an intelligent dialog method provided by the embodiment of the present disclosure.
In an embodiment, in the session scenario a1, the intelligent customer service system 1121 may be disposed in the customer premise equipment 111, and the corresponding customer service robot is also disposed in the customer premise equipment 111, in which case the device 112 of the intelligent session is the customer premise equipment 111. Or, in another embodiment, in the conversation scenario a1, the apparatus 112 of the intelligent conversation and the user end device 111 are two independent entities, the intelligent customer service system 1121 and the user end device 111 can perform communication connection, and the intelligent customer service system 1121 provides background data support and service for the customer service operation interface on the user end device 111.
In another embodiment, referring to the single and double arrows in FIG. 1A, the intelligent dialog apparatus 112 comprises an intelligent customer service system 1121, and the intelligent customer service system 1121 comprises an artificial customer service system.
In one scenario, the customer service robot replies to the user consultation problem in the customer service operation interface based on a preset logic, and under the condition of triggering the manual condition, the user 101a is switched to the manual customer service system, so that the manual customer service 102a interacts with the manual customer service system or the intelligent customer service system 1121 to communicate with the user 101 a. In another scenario, the user 101a does not perform a session with the customer service robot, but directly initiates a request for manual customer service transfer, and when the manual condition is met, the user 101a may perform a session with the manual customer service 102a, and the manual customer service 102a may perform customer service by referring to a prompt of the manual customer service system or the intelligent customer service system 1121.
Referring to a dialog scenario a2 illustrated in fig. 1A, the dialog scenario a2 occurs between the user 101A and the artificial customer service 102a, the user 101A may perform a dialog with the artificial customer service 102a through the user end device 111, and during the dialog between the artificial customer service 102a and the user 101A, the intelligent customer service system 1121 in the intelligent dialog apparatus 112 provides a dialog behavior prompt for the artificial customer service 102a at the next time by executing the method of intelligent dialog provided by the embodiment of the present disclosure.
In the dialog scenario a2, the intelligent dialog apparatus 112 and the user end device 111 are two independent main bodies, the intelligent customer service system 1121 includes an artificial customer service system, the intelligent dialog apparatus 112 and the user end device 111 can be in communication connection, and the intelligent customer service system 1121 provides background data support and service for a customer service operation interface on the user end device 111 and also provides service operation guidance and assistance for the artificial customer service 102 a.
It is understood that, in some application scenarios, the intelligent customer service system 1121 may also be integrated with a customer service robot and an artificial customer service system at the same time, as shown in fig. 1A.
Referring to the bi-directional single-headed arrow in fig. 1B, in one implementation scenario, an exemplary system architecture 120 suitable for use in the method of intelligent dialog of an embodiment of the present disclosure includes: means for intelligent dialogue 122. The system architecture 120 may be applied to social scenarios.
In this implementation scenario, the intelligent dialog device 122 is an intelligent robot, and the intelligent robot in this case is a broad concept, and any machine device/equipment and the like that has the features of dialog capability and artificial intelligence and can analyze, process and output dialog data can be regarded as an intelligent robot, and examples include but are not limited to: an intelligent voice robot, an intelligent chat robot, an intelligent sound, an intelligent rescue robot, etc., which may be further attached with other functions, such as industrial manufacturing, dancing, etc., and which includes an intelligent dialog system 1221.
Referring to a dialog scenario B1 illustrated in fig. 1B, the dialog scenario B1 occurs between the user 101B and the intelligent robot, the user 101B can directly perform a dialog with the intelligent robot, and the intelligent dialog system 1221 in the intelligent robot provides a dialog behavior prompt at the next time for the intelligent robot by performing the method of intelligent dialog provided by the embodiment of the present disclosure.
The user 101b may be a person having a conversation right to perform a conversation with the intelligent robot, and for example, if the conversation right is opened for all persons, anyone may perform a chat/conversation with the intelligent robot.
In another implementation scenario, referring to the double-headed arrow in fig. 1B, the system architecture 120 includes both the customer premise equipment 121 and the intelligent dialog device 122. The intelligent dialogue device 122 is installed with social software/friend making software/communication software for dialogue, etc. in addition to the intelligent dialogue system 1221.
Referring to an exemplary conversation scenario B2 in fig. 1B, the conversation scenario B2 occurs between the user 101B and the user 102B, the user 101B and the user 102B are users of the social software/friend-making software/communication software, the user 101B and the user 102B may be known people or strangers, and the conversation scenarios between the users may be daily communication, social interaction, acquaintances, friend-making, and the like. The social software/friend making software/communication software or device used by at least one user has an intelligent conversation assistance function, which is implemented by the intelligent conversation system 1221. For example, the device 122 of the intelligent dialog used by the user 102b, executes the method of the intelligent dialog provided by the embodiment of the present disclosure through the intelligent dialog system 1221, and provides the user 102b with a prompt of the dialog behavior at the next time.
Based on the description of the above scenario, it can be known that the method for intelligent dialog provided by the embodiment of the present disclosure has wide applicability and practical significance, and can be applied to a customer service scenario or a social scenario, and under the broad meaning of social contact, the method may include: conversation and exchange in the search and rescue task are beneficial to improving the service quality of customer service and the normalization of the service process, social popularity, friend making success rate, rescue success rate of rescued objects and the like.
Embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
A first exemplary embodiment of the present disclosure provides a method of intelligent dialog.
Fig. 2 schematically shows a flow diagram of a method of intelligent dialogue according to an embodiment of the present disclosure.
Referring to fig. 2, a method for intelligent dialog provided by an embodiment of the present disclosure includes the following operations: s201, S202 and S203. As will be understood from the description of fig. 1A and 1B, operations S201 to S203 may be performed by an electronic device installed with the intelligent customer service system 1121, or by an electronic device installed with the intelligent dialog system 1221.
In operation S201, during the communication between the participants of the conversation, the content of the conversation that occurs is recorded.
The conversation participants refer to objects in conversation, for example, M and N are talking, and the conversation participants are M and N. A. B and C, the conversation participants include a, B and C. The communication scene of the conversation participants can be a scene of communication between people and a machine, and can also be a process of information communication between people and the machine.
The communication mode can be a voice mode, a text mode, etc., and correspondingly, the dialogue content also comprises a text mode and/or a voice mode.
Communication includes face-to-face communication as well as remote communication. For example, in remote communication, communication can be performed through a communication phone, a network phone, or a chat interface in instant messaging software, a voice call, a video call, or the like.
When recording the dialog content, the identification of the respective object is recorded in the conversation of each object, i.e. it is clear which object expresses what information.
In operation S202, the dialog contents are input to the dialog policy support model for processing, and the output information includes: and recommending information for conversation action at the next moment.
The input of the dialogue strategy assistance model is the dialogue content which is in the form of voice or text and occurs, and the output information comprises: and recommending information for conversation action at the next moment.
For example, during the communication process of the conversation participants, the conversation contents are input into the conversation strategy assistance model in real time while the conversation contents are recorded.
The dialog behavior recommendation information includes, but is not limited to, one or more of the following information: call, self-introduction, mood appeasing, question and search, thank you, invitation evaluation, active marketing, etc.
When the dialog behavior recommendation information is various, relative priority or relative importance among the behavior recommendation information is provided.
In operation S203, a dialog behavior prompt at the next time is provided to at least one of the participants of the dialog according to the dialog behavior recommendation information.
The information output by the conversation strategy auxiliary model can cover behavior recommendation information of all conversation participants, and can provide conversation behavior prompt at the next moment for one of the conversation participants in a use scene. For example, one of the above-described conversation participants may be: a customer service person using an electronic device installed with the intelligent customer service system 1121, or a customer service robot in the intelligent customer service system 1121, or a user using an electronic device installed with the intelligent dialog system 1221, or an intelligent robot installed with the intelligent dialog system 1221. And providing a dialog behavior prompt of the next moment for the multiple parties in the dialog participants. For example, two persons who both use the intelligent conversation device 122 illustrated in fig. 1B and are not good at social contact are expected to communicate smoothly by the aid of the auxiliary function provided by the intelligent customer service system 1221, so as to improve the communication satisfaction of the other person to the other person.
Based on the operations S201 to S203, the generated dialog content is input to the dialog policy assistance model for processing, the obtained output information includes the dialog behavior recommendation information at the next time, and a dialog behavior prompt at the next time is provided for at least one of the dialog participants according to the dialog behavior recommendation information at the next time, so that a corresponding reply can be developed with reference to the dialog behavior prompt before the prompted current object performs a dialog, and the dialog behavior prompt helps to improve the communication efficiency, and at the same time, the prompted current object can communicate with the communication object in an accurate, rapid and appropriate manner, which helps to improve the satisfaction degree of the communication object to the current object.
When the method for intelligent dialogue provided by the embodiment of the disclosure is applied to a customer service scene, compared with the existing intelligent customer service system, the existing intelligent customer service system can only provide assistance in semantic aspect, that is, help to provide correct answers corresponding to user questions. However, conversation exchange in a real scene is not only passive answer questions, but also various conversation behaviors such as active inquiry, emotional appeasing, thank you for invitation and the like. Because the concept provided by the disclosure focuses on global decision of the conversation strategy, the optimal conversation behavior strategy is generated through prediction in the context of the conversation text, the manual customer service/customer service robot is helped to adopt the most appropriate conversation strategy, namely the answer is answered, the question is asked, and the recommendation is recommended, so that the interaction form is more natural, commodity recommendation and sale can be better carried out, the requirement of the customer can be better met, and the like, and the improvement of the customer service quality is facilitated.
According to an embodiment of the present disclosure, in the above operation S201, during the process of communicating between the session participants, the recording of the occurring session content includes: during the communication process of the conversation participants in the target conversation scene, the conversation content is recorded through recording or recorded in a chat interface.
Wherein the target dialog scenario includes at least one of the following dialog scenarios: a conversation scene between people and a machine, the conversation scene comprising: a customer service consultation scenario or a social scenario.
For example, when the method is applied to a customer service scene or a social scene, when a and B communicate, for example, a is a consumer and B is a customer service person/a customer service robot; or A and B are two persons who socialize/communicate/make friends, etc.; or A is a person, B is an intelligent robot for carrying out conversation, search and rescue and the like; in each implementation scenario, the device/system/equipment used by B or B itself (in the case of an intelligent robot) can determine, according to the content of the conversation that has occurred, that the behavior recommendation information to be used for next-time communication includes one or more of the following behavior information: the method comprises the following steps of calling, introducing self, appeasing the emotion of the other party, asking questions, replying, expressing thank you, inviting evaluation, initiatively marketing and the like, and providing conversation behavior prompts for B accordingly, wherein the B can carry out conversation at the next moment by referring to the conversation behavior prompts, so that the communication satisfaction degree of the A to the B can be improved.
It should be noted that, in the broad sense of social contact, for example, the following may be included: the intelligent conversation method is implemented in the scenes of customer service, social interaction and the like, and is beneficial to improving the service quality of the customer service, the normalization of the service process, the social popularity, the friend making success rate, the rescue success rate of the rescued object and the like.
Fig. 3 schematically shows a structural diagram of a dialogue strategy assistance model according to an embodiment of the present disclosure.
Referring to the solid box illustration in fig. 3, the dialog strategy assistance model 300 includes, according to an embodiment of the present disclosure: an encoding layer 301, a context prediction layer 302, and a dialog behavior prediction layer 303.
Referring to fig. 3, the dialog strategy assistance model 300 includes, according to another embodiment of the present disclosure, a solid-line block and a dashed-line block: an encoding layer 301, a context prediction layer 302, a dialog behavior prediction layer 303, and a decoding layer 304.
Fig. 4 schematically shows a detailed implementation flowchart of operation S202 according to an embodiment of the present disclosure. Fig. 5 schematically shows an implementation process diagram of operation S202 according to an embodiment of the present disclosure. In fig. 5, a solid line part represents a corresponding implementation scenario when the dialog policy auxiliary model includes the encoding layer 301, the context prediction layer 302 and the dialog behavior prediction layer 303, and a dotted line part represents an implementation process diagram that may be included in the further operation S202 when the dialog policy auxiliary model further includes the decoding layer 304.
Referring to fig. 4, the dialogue strategy assistance model 300 includes: based on the embodiments of the encoding layer 301, the context prediction layer 302, and the conversation activity prediction layer 303, in the operation S202, the inputting of the conversation content into the conversation policy assistance model for processing includes the following operations: s401, S402 and S403.
In operation S401, the dialog content is input into the dialog policy support model, the sentence in the dialog content is encoded by the encoding layer, and a sentence vector for representing semantic information is output.
Referring to fig. 5, as indicated by a solid arrow, the encoding layer 301 encodes a sentence in the dialog content recorded in the above-described operation S201 that occurs.
In some embodiments, the specific process of encoding may include: the sentences in the dialogue content are divided according to morphemes by adopting a preprocessing network layer, then each morpheme is input into the coding layer 301, and a corresponding sentence vector is obtained through output. The meaning of the sentence here means: the content generated by the speaking object in each turn of conversation can be 1 sentence or a plurality of sentences.
For example, referring to the solid arrow in fig. 5, taking the example that the conversation content occurs between a and B, the recorded conversation content has the conversation occurrence order and the production target of the speaking content, for example, the recording form of the conversation content includes: {10: "sentence 1";10, B: "sentence 2";10, A says "sentence 3" }.
Correspondingly splitting sentence 1 into morphemes W 1i Correspondingly splitting sentence 2 into morphemes W 2j Correspondingly splitting the sentence 3 into morphemes W 3k (ii) a Wherein, i, j and k are respectively the serial numbers 1, 2 and 3 of each morpheme 8230, and the maximum value of i, j and k is the total number of the morphemes. Each morpheme W 1i Inputting the sentence vector into the coding layer 301 to obtain a sentence vector U 1 (ii) a Each morpheme W 2j Inputting the sentence vector into the coding layer 301 to obtain a sentence vector U 2 (ii) a Each morpheme W 3k Inputting the sentence vector into the coding layer 301 to obtain a sentence vector U 3 . The above sentence vector U 1 、U 2 And U 3 In the form of vectors characterizing the semantics of the respective sentences. For example, the coding layer 301 may be a Bi-directional long-short term memory model (Bi-LSTM), a denoising auto-encoder (BART), a unified pre-training language model (UniLM), or an autoregressive language model (GPT), etc. to code the dialog features.
Illustratively, sentence 1 is: "you good, i have a question to consult"; sentence 2 is: "you are good, i.e. customer service with number x, happy to serve you, and what you specifically have needs of consultation"; sentence 3 is: "the efficiency is too low when the user is transferred to the personal customer service after waiting for a long time; however, we want to ask how the specific functions of the vacuum cleaner product are, and do it easy for a novice to operate.
Morpheme W after sentence 1 is split 1i Respectively as follows: w 11 ～W 15 The correspondence is respectively: you, me, have, question, consult. Morpheme W after sentence 2 is split 1i Respectively as follows: w 21 ～W 2(14) (14 in brackets represents a whole number, if the understanding is not actually influenced, the brackets can be removed; and the subsequent same representation modes can be analogized) are respectively as follows: you, i, y, number x, customer service, happy, serve you, specific, what, need, consult. Morpheme W after sentence 3 is split 1i Respectively as follows: w 31 ～W 3(23) The user can ask about the specific function, how, right, novice, whether, convenient and operation of the product of the dust collector.
In operation S402, the sentence vectors are sequentially input to the context prediction layer according to the dialog occurrence order of the dialog contents, and the context prediction layer processes the sentence vectors to generate hidden variables of the dialog at the next time corresponding to each sentence vector.
Referring to the solid arrow in FIG. 5, the sentence vector U is divided according to the order of occurrence of the dialog contents 1 、U 2 And U 3 Sequentially input into the context prediction layer 302, and the context prediction layer 302 processes the sentence vector U 1 、U 2 And U 3 Processing to generate sentence vector U 1 Corresponding hidden variable V of next time conversation 1 Sentence vector U 2 Hidden variable V of corresponding next moment conversation 2 Sentence vector U 3 Corresponding hidden variable V of next time dialogue 3 。
In operation S403, the hidden variable is input to the dialogue behavior prediction layer, a global optimal solution of the dialogue behaviors corresponding to the hidden variable is determined among a plurality of predefined dialogue behaviors by the dialogue behavior prediction layer, and the global optimal solution of the dialogue behaviors is output as dialogue behavior recommendation information.
Referring to FIG. 5, shown by the solid arrow, the hidden variable V 1 ，V 2 And V 3 Input into the dialogue behavior prediction layer 303, and the dialogue behavior prediction layer 303 determines a global optimal solution in the dialogue behavior corresponding to the hidden variable, namely, the hidden variable V in the example, among a plurality of predefined dialogue behaviors 1 The global optimal solutions in the corresponding dialogue behavior are respectively: behavior A 13 Hidden variable V 2 The global optimal solutions in the corresponding dialogue behavior are respectively: behavior A 22 Hidden variable V 3 The global optimal solutions in the corresponding dialog behavior are respectively: behavior A 31 . Behavior A 13 Behavior A 22 And action A 31 For example, respectively: self-introduction, reply, appeasing the emotion of the other party and ask for search.
The words of A in the second turn actually correspond to sentences under two semantics, so the behavior A obtained after the processing of the dialogue behavior prediction layer 31 Including appeasing the emotion of the other party and exploring the two conversational behaviors back and forth. Accordingly, in operation S203, according to the dialog behavior recommendation information, a dialog behavior prompt at the next time may be provided to B as follows: the emotion of the other party is pacified and then the opposite party is asked and searched.
Illustratively, in an existing dialog: after ase:Sub>A-B-ase:Sub>A, B may reply according to the above dialog behavior prompt when replying to ase:Sub>A in the second round, for example, reply as follows: "I can understand your experience now very much, because the resources of the artificial customer service are limited, I shows sorry for too long waiting time; asking how large the area of the house is, where the main cleaning area of the dust collector is, asking for providing the information, and providing relevant recommendations for the user.
In other embodiments, when encoding is performed in operation S401, a preprocessing network layer may be used to perform sentence semantic division on the dialog content according to a speaking object, semantics, a pause relation, and the like, divide the divided sentences according to morphemes, and then perform corresponding encoding operation on each sentence to obtain a corresponding sentence vector. After the operations S402 and S403 are performed, the hidden variable of the dialog at the next time corresponding to each sentence vector is processed by the CRF model, and the dialog behavior recommendation information at the next time is output to be single.
Fig. 6 schematically shows an implementation process diagram of operation S403 according to an embodiment of the present disclosure.
According to an embodiment of the present disclosure, in the operation S403, determining, by the dialogue activity prediction layer, a global optimal solution in the dialogue activities corresponding to the hidden variables in a predefined plurality of dialogue activities includes: and taking the hidden variables as the hidden states of the dialogue behavior prediction layer, taking a plurality of predefined dialogue behaviors as the observation states of the dialogue behavior prediction layer, and calculating by the dialogue behavior prediction layer based on a Conditional Random Field (CRF) algorithm according to the emission probability between the hidden states and the observation states and the transition probability between the observation states to determine a global optimal solution in the dialogue behaviors.
Referring to FIG. 6, the above-mentioned hidden variable V is shown 1 ，V 2 And V 3 As a hidden state of the dialogue behavior prediction layer 303, a plurality of predefined dialogue behaviors are set as an observation state of the dialogue behavior prediction layer, for example, a dialogue behavior a 11 、 A 12 And A 13 Is a hidden variable V 1 Of the observed state, dialog behavior A 21 、A 22 And A 23 Is a hidden variable V 2 Of the observed state, conversational behaviour A 31 、A 32 And A 33 Is a hidden variable V 3 The observation state of (1). In fig. 6, a dotted arrow is used to illustrate a path corresponding to the emission probability between the hidden state and the observed state, a solid arrow is used to illustrate a path corresponding to the transition probability between the observed states, and when a conditional random field algorithm is used for calculation, a global optimal solution is obtained through successive multiplications of probabilities, where the global optimal solution corresponds to: a. The 13 、A 22 And A 33 By way of example, the connections are schematically illustrated in FIG. 6 by long dashed arrowsThe path corresponding to the emission probability of the global optimal solution is distinguished from the paths indicated by other short-dashed arrows; in addition, fig. 6 also illustrates an optimal path corresponding to the global optimal solution, which is shown with reference to the path marked by the five-pointed star in fig. 6.
In the embodiment including operations S401 to S403, processing is performed based on the sentence-level model, then processing is performed based on the dialogue-level model, then based on the dialogue-level model, a CRF algorithm is used to generate a corresponding dialogue policy, and at least one of the dialogue participants is guided from the dialogue behavior level.
According to an embodiment of the present disclosure, the dialogue strategy assistance model 300 described above includes: based on the coding layer 301, the context prediction layer 302, the dialog behavior prediction layer 303, and the decoding layer 304, in operation S202, the dialog contents are input to the dialog policy assistant model for processing, and operation S404 is also included in addition to operations S401, S402, and S403.
In operation S404, the hidden variable is input to the decoding layer 304, and the decoding layer 304 decodes the hidden variable to obtain recommended dialect information at the next time, as shown by the dotted arrow and the dotted frame in fig. 5.
For example, the hidden variable V may be set 3 Input to the decoding layer 304, and decoded by the decoding layer 304 to obtain predicted morphemes at the next time
So that the predicted word is picked up by the above-mentioned predictor>
And obtaining recommended dialect information, wherein the value of m is the serial number 1, 2, 3 \8230, 8230and the maximum value of m is the total number of the predicted morphemes. For example, the recommended dialogistic information is: "I understand your feeling, for which I denotes a sorry; you can provide the using environment information of the dust collector so that the relevant recommendation can be made by the user according to the information.
Fig. 7 schematically shows a flow diagram of a method of intelligent dialog according to another embodiment of the present disclosure.
Referring to fig. 7, the dialogue strategy assistance model 300 includes: based on the coding layer 301, the context prediction layer 302, the dialog behavior prediction layer 303, and the decoding layer 304, the dialog strategy assistance model 300 outputs information further including: for the sake of distinction, the recommended speech information at the next time includes, in fig. 7, information schematically output in operation S202 a: and operation steps of recommending the dialogue behavior information at the next moment and recommending the dialogue information at the next moment. On the basis, the method of the intelligent dialog includes operation S701 in addition to the operations S201 to S203.
In operation S701, a speech technology prompt at the next time is provided to at least one of the participants of the conversation based on the recommended speech technology information.
In some embodiments, the method of intelligent dialog may further include, on the basis of the operations S201 to S203 and S701, the step of: and performing automatic reply according to the dialogue action prompt and the dialogue operation prompt. The operation can be triggered when the user clicks the dialogue behavior prompt and the dialogue prompt, or can be automatically triggered to be executed according to the preset authorization setting of the user. For example, the above automatic reply mode may be applied to a scenario in which the customer service robot has a conversation with the user, or a scenario in which the intelligent robot has a conversation with a person, and the like.
In the embodiment including operations S201 to S203 and S701, where operation S202 includes operations S401 to S404, in addition to the advantages of the above embodiment including operations S401 to S403, since the dialog policy can be output and the suggestion related to the dialog technique can be output, when the embodiment is applied to a customer service scene or a social scene, not only can users such as an artificial customer service, a customer service robot, an intelligent robot, and a user with social assistance need be assisted to adopt a relatively appropriate dialog behavior, but also references for the dialog technique can be provided for the users.
According to embodiments of the present disclosure, the encoding layer 301 and the decoding layer 304 include, but are not limited to: a Bi-directional long-short term memory model (Bi-LSTM), a denoising autocoder (BART), a unified pre-training language model (UniLM), or an autoregressive language model (GPT); the context prediction layer 202 includes, but is not limited to: a bidirectional long-short term memory model (Bi-LSTM); the dialog behavior prediction layer 203 includes but is not limited to: conditional Random Field (CRF) models.
According to the embodiment of the present disclosure, the building process of the dialog policy assistance model is as follows: and taking the training corpus in the conversation scene as the input of a machine learning model, wherein the output of the machine learning model is the behavior prediction information at the next moment, and the structure of the machine learning model is the same as that of the conversation strategy auxiliary model, except that the parameters in the machine learning model are initial parameters and are not optimized. And performing multiple training on the machine learning model by using the fact of the conversation corresponding to the next moment in the training corpus as a training label serving as the behavior prediction information, and adjusting parameters of the machine learning model in the training process to ensure that a loss function representing the difference between the behavior prediction information and the training label is converged, wherein the training is considered to be finished, and the trained machine learning model is the conversation strategy auxiliary model.
For example, taking application to a customer service scene as an example, the corpus in the conversation scene may use conversation content between high-quality customer service (i.e., customer service with a high service quality score in the conversation process) and the client as the corpus, perform behavior tagging on the conversation content actually occurring in the high-quality customer service, and use the tagged behavior as a label during training. While the behavior annotation of a single object is illustrated according to the actual requirement, in other embodiments, the behavior annotation of a plurality of objects may be performed.
In an embodiment, during training, the input corpus needs to be limited in length, for example, the number of dialog turns is limited to 3 rounds (the computational power of a machine of a specific training model can be changed by the limited length), then if there are 6 rounds of dialog contents between a high-quality customer service and a client in one corpus (similar to the case of multiple corpora), the 1 st round of dialog contents can be input into a machine learning model, the behavior prediction result of each object (user and high-quality customer service) in the 2 nd round of dialog is output, and the actual behavior of the selected object (for example, high-quality customer service) in the 2 nd round of dialog is used as a training tag to train the machine learning model; in order to implement generalization of the model, the 1 st round of dialog and the 2 nd round of dialog contents can be further input into the machine learning model, the behavior prediction results of each object in the third round of dialog are output, and the actual behavior of the selected object (such as high-quality customer service) in the 3 rd round of dialog is used as a training label to train the machine learning model; similarly, the 1 st to 3 rd round conversation contents may be input into the machine learning model, the behavior prediction results of each object in the 4 th round conversation are output, the actual behavior of the selected object (for example, high-quality customer service) in the 4 th round conversation is used as a training label, the machine learning model is trained again, and then the 2 nd to 4 th and 3 rd to 5 th round conversation contents may be input as a training corpus in the same manner.
Similarly, when the method is applied to a social scene, for example, the dialog content in the search and rescue success case may be used as the corpus, the dialog content in the chat scene corresponding to the high user evaluation satisfaction may be used as the corpus, and the like.
A second exemplary embodiment of the present disclosure provides an apparatus for intelligent dialog.
Fig. 8 schematically shows a block diagram of an apparatus for intelligent dialogue according to an embodiment of the present disclosure.
Referring to fig. 8, an apparatus 800 for intelligent dialog provided by an embodiment of the present disclosure includes: a dialog logging module 801, a processing module 802 and a prompt module 803.
The conversation recording module 801 is configured to record conversation contents occurring during communication between conversation participants.
The processing module 802 is configured to input the dialog content into the dialog policy auxiliary model for processing, where the output information includes: and recommending information for the conversation behavior at the next moment.
Referring to fig. 3, the dialog strategy assistance model 300 includes: an encoding layer 301, a context prediction layer 302, and a dialog behavior prediction layer 303. The processing module 802 includes functional modules or sub-modules for implementing the operations S401 to S403.
The prompt module 803 is configured to provide a dialog behavior prompt at the next time for at least one of the dialog participants according to the dialog behavior recommendation information.
According to another embodiment of the present disclosure, referring to fig. 3, the dialogue strategy assistance model 300 includes: an encoding layer 301, a context prediction layer 302, a dialog behavior prediction layer 303, and a decoding layer 304. Correspondingly, the processing module 802 may include, in addition to the functional modules or sub-modules for implementing the operations S401 to S403, the functional modules or sub-modules for implementing the operation S404, and the obtained output information includes: besides the recommended speech behavior information at the next moment, the recommended speech operation information at the next moment is also included. The prompt module 803 is further configured to provide a speech prompt at a next time for at least one of the conversation participants according to the recommended speech information.
According to an embodiment of the present disclosure, the apparatus 800, besides including the dialog recording module 801, the processing module 802, and the prompt module 803, may further include: and an automatic reply module. The automatic reply module is used for carrying out automatic reply according to the conversation behavior prompt and the conversation prompt.
According to an actual scenario of the present disclosure, for example, as shown in fig. 1A, the apparatus 800 of the intelligent dialog (which may correspond to the apparatus 112 of the intelligent dialog in fig. 1A) includes an intelligent customer service system 1121, and the intelligent customer service system 1121 includes the dialog recording module 801, the processing module 802, and the prompting module 803. Wherein, one of the conversation participants is an artificial customer service or a customer service robot; the prompt module is configured to provide a dialog behavior prompt at a next time for the manual customer service or the customer service robot according to the dialog behavior recommendation information, which is shown in dialog scenes a1 and a2 in fig. 1A.
According to another implementation scenario of the present disclosure, for example, as shown in fig. 1B, the apparatus 800 of the intelligent dialog (which may correspond to the apparatus 122 of the intelligent dialog in fig. 1B) includes an intelligent dialog system 1221, and the intelligent dialog system 1221 includes the dialog recording module 801, the processing module 802, and the prompting module 803. Wherein one of the conversation participants is a user using the intelligent conversation system; the prompt module is configured to provide a dialog behavior prompt at the next time for the user according to the dialog behavior recommendation information, which is shown in a dialog scene B2 in fig. 1B. Or, wherein the device is an intelligent robot, and one of the conversation participants is the intelligent robot; the prompt module is used for providing a conversation behavior prompt of the next moment for the intelligent robot according to the conversation behavior recommendation information; the intelligent robot makes a conversation with the communication object based on the conversation behavior presentation, as shown in a conversation scene B1 in fig. 1B.
Any number of the above-described dialog recording module 801, processing module 802 and prompt module 803 may be combined into one module to be implemented, or any one of them may be split into a plurality of modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of the other modules and implemented in one module. At least one of the dialog logging module 801, the processing module 802 and the prompting module 803 may be implemented at least partially as a hardware circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented in hardware or firmware in any other reasonable manner of integrating or packaging a circuit, or in any one of three implementations of software, hardware and firmware, or in any suitable combination of any of them. Alternatively, at least one of the dialog logging module 801, the processing module 802 and the prompt module 803 may be at least partially implemented as a computer program module which, when executed, may perform a corresponding function.
A third exemplary embodiment of the present disclosure provides an electronic apparatus.
Fig. 9 schematically shows a block diagram of an electronic device provided by an embodiment of the present disclosure.
Referring to fig. 9, an electronic device 900 provided in the embodiment of the present disclosure includes a processor 901, a communication interface 902, a memory 903, and a communication bus 904, where the processor 901, the communication interface 902, and the memory 903 are in communication with each other through the communication bus 904; a memory 903 for storing computer programs; the processor 901 is configured to implement the method of intelligent dialogue as described above when executing the program stored in the memory.
A fourth exemplary embodiment of the present disclosure also provides a computer-readable storage medium. The computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the method of intelligent dialog as described above.
The computer-readable storage medium may be contained in the apparatus/device described in the above embodiments; or may be present alone without being assembled into the device/apparatus. The computer-readable storage medium carries one or more programs which, when executed, implement methods according to embodiments of the disclosure.
According to embodiments of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
It is noted that, in this document, relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the language "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
The foregoing are merely exemplary embodiments of the present disclosure, which enable those skilled in the art to understand or practice the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the disclosure. Thus, the present disclosure is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (13)
1. A method of intelligent dialog, comprising:
recording the conversation content during the communication process of the conversation participants;
inputting the conversation content into a conversation strategy auxiliary model for processing, wherein the output information comprises: recommending information of the conversation behavior at the next moment; and
and providing a conversation behavior prompt at the next moment for at least one of the conversation participants according to the conversation behavior recommendation information.
2. The method of claim 1, wherein the conversation strategy assistance model comprises: the device comprises a coding layer, a context prediction layer and a conversation behavior prediction layer;
wherein, the inputting the conversation content into the conversation strategy auxiliary model for processing comprises:
inputting the conversation content into a conversation strategy auxiliary model, coding sentences in the conversation content by the coding layer, and outputting to obtain a sentence vector for representing semantic information;
inputting the sentence vectors into the context prediction layer in sequence according to the dialog occurrence sequence of the dialog contents, and processing the sentence vectors by the context prediction layer to generate hidden variables of the dialog at the next time corresponding to each sentence vector; and
and inputting the hidden variables into the conversation behavior prediction layer, determining a global optimal solution in conversation behaviors corresponding to the hidden variables in a plurality of predefined conversation behaviors by the conversation behavior prediction layer, and outputting the global optimal solution in the conversation behaviors as conversation behavior recommendation information.
3. The method of claim 2, wherein determining, by the dialog behavior prediction layer, a globally optimal solution in the dialog behavior corresponding to the hidden variable among a predefined plurality of dialog behaviors comprises:
and taking the hidden variables as the hidden states of the dialogue behavior prediction layer, taking a plurality of predefined dialogue behaviors as the observation states of the dialogue behavior prediction layer, and calculating by the dialogue behavior prediction layer based on a conditional random field algorithm according to the emission probability between the hidden states and the observation states and the transition probability between the observation states to determine the global optimal solution in the dialogue behaviors.
4. The method of claim 2, wherein the conversation strategy assistance model further comprises: a decoding layer;
the inputting the conversation content into the conversation strategy auxiliary model for processing further comprises:
and inputting the hidden variable into the decoding layer, and decoding by the decoding layer to obtain recommended dialogues information at the next moment.
5. The method of claim 1 or 4, wherein the outputted information further comprises: recommended dialect information at the next moment;
wherein the method further comprises:
and providing a dialect prompt of the next moment for at least one of the conversation participants according to the recommended dialect information.
6. The method of claim 5, further comprising: and automatically replying according to the dialogue action prompt and the dialogue prompt.
7. The method of claim 4, wherein the encoding layer and the decoding layer comprise: a bidirectional long-short term memory model, a denoising autoencoder, a unified pre-training language model, or an autoregressive language model; the context prediction layer includes: a two-way long-short term memory model; the dialog behavior prediction layer includes: a conditional random field model.
8. The method of claim 1, wherein recording the conversation content occurring during the communication between the conversation participants comprises:
in the process of communication of conversation participants in a target conversation scene, recording the conversation content through recording or recording a chat interface;
wherein the target dialog scenario comprises at least one of the following dialog scenarios: a conversation scene between people and machines, the conversation scene comprising: a customer service consultation scenario or a social scenario.
9. An apparatus for intelligent dialog, comprising:
the conversation recording module is used for recording the conversation content in the communication process of the conversation participants;
the processing module is used for inputting the conversation content into the conversation strategy auxiliary model for processing, and the output information comprises: conversation behavior recommendation information at the next moment; and
and the prompt module is used for providing a conversation behavior prompt at the next moment for at least one of the conversation participants according to the conversation behavior recommendation information.
10. The apparatus of claim 9, wherein the apparatus comprises a smart customer service system, the smart customer service system comprising the conversation recording module, the processing module, and the prompting module;
wherein one of the conversation participants is an artificial customer service or a customer service robot; and the prompting module is used for providing a conversation behavior prompt at the next moment for the artificial customer service or the customer service robot according to the conversation behavior recommendation information.
11. The apparatus of claim 9, wherein the apparatus comprises an intelligent dialog system, the intelligent dialog system comprising the dialog recording module, the processing module, and the prompt module;
wherein one of the conversation participants is a user using the intelligent conversation system; the prompt module is used for providing a dialog behavior prompt at the next moment for the user according to the dialog behavior recommendation information; or,
the device is an intelligent robot, and one of the conversation participants is the intelligent robot; the prompting module is used for providing conversation behavior prompt at the next moment for the intelligent robot according to the conversation behavior recommendation information; and the intelligent robot carries out conversation with the communication object according to the conversation behavior prompt.
12. An electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus;
a memory for storing a computer program;
a processor for implementing the method of any one of claims 1 to 8 when executing a program stored on a memory.
13. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111388664.6A CN115934901A (en) | 2021-11-22 | 2021-11-22 | Intelligent conversation method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111388664.6A CN115934901A (en) | 2021-11-22 | 2021-11-22 | Intelligent conversation method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115934901A true CN115934901A (en) | 2023-04-07 |
Family
ID=86554489
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111388664.6A Pending CN115934901A (en) | 2021-11-22 | 2021-11-22 | Intelligent conversation method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115934901A (en) |
-
2021
- 2021-11-22 CN CN202111388664.6A patent/CN115934901A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10262195B2 (en) | Predictive and responsive video analytics system and methods | |
| CN111028827B (en) | Interaction processing method, device, equipment and storage medium based on emotion recognition | |
| TWI698830B (en) | Method and device for transferring robot customer service to manual customer service, computer equipment and computer readable storage medium | |
| CN108595436B (en) | Method and system for generating emotional dialogue content and storage medium | |
| CN106133825A (en) | Broad sense phrase in automatic speech recognition system | |
| KR101615848B1 (en) | Method and computer program of recommending dialogue sticker based on similar situation detection | |
| US10950223B2 (en) | System and method for analyzing partial utterances | |
| CN110069612B (en) | Reply generation method and device | |
| TW201919042A (en) | Voice interactive device and voice interaction method using the same | |
| Wilks et al. | A prototype for a conversational companion for reminiscing about images | |
| CN116049360A (en) | Intelligent voice dialogue scene conversation intervention method and system based on client image | |
| CN111462726B (en) | Method, device, equipment and medium for answering out call | |
| Sabharwal et al. | Developing Cognitive Bots Using the IBM Watson Engine: Practical, Hands-on Guide to Developing Complex Cognitive Bots Using the IBM Watson Platform | |
| CN110489519B (en) | Session method based on session prediction model and related products | |
| US9667786B1 (en) | Distributed coordinated system and process which transforms data into useful information to help a user with resolving issues | |
| CN110827040A (en) | Consumer appeal solution method and system | |
| CN114239607A (en) | Conversation reply method and device | |
| CN110931002B (en) | Man-machine interaction method, device, computer equipment and storage medium | |
| CN112069830A (en) | Intelligent conversation method and device | |
| Mohammed et al. | Chatbot system architecture | |
| CN116561284A (en) | Intelligent response method, device, electronic equipment and medium | |
| Hung et al. | Context‐Centric Speech‐Based Human–Computer Interaction | |
| CN115934901A (en) | Intelligent conversation method and device, electronic equipment and storage medium | |
| US11798538B1 (en) | Answer prediction in a speech processing system | |
| CN110765242A (en) | Method, device and system for providing customer service information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |