CN114490994B - Conversation management method and device - Google Patents
Conversation management method and device Download PDFInfo
- Publication number
- CN114490994B CN114490994B CN202210308618.9A CN202210308618A CN114490994B CN 114490994 B CN114490994 B CN 114490994B CN 202210308618 A CN202210308618 A CN 202210308618A CN 114490994 B CN114490994 B CN 114490994B
- Authority
- CN
- China
- Prior art keywords
- information
- dialog
- dialogue
- submodel
- target information
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Abstract
The invention provides a conversation management method and a device, wherein the method comprises the following steps: acquiring target information; inputting the target information into a dialogue management model, and acquiring response information corresponding to the target information output by the dialogue management model; the target information is information input by the user in the current conversation process; the dialogue management model is obtained by training based on the sample information and the sample response information corresponding to the sample information; the dialogue management model is used for updating the dialogue state representation corresponding to the dialogue at this time based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; the dialog states are represented as one-dimensional vectors. The conversation management method and the conversation management device provided by the invention can realize automatic maintenance of the man-machine conversation state without manual definition, can express based on the one-dimensional conversation state, improve the calculation speed, and can realize more efficient and more accurate conversation management with lower cost investment.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a conversation management method and device.
Background
With the development of science and technology, human-machine conversation, especially task-oriented human-machine conversation of task type, is widely used in various fields, for example: personal assistant, intelligent customer service, intelligent house, etc. The smooth and high-quality man-machine conversation can provide quick, efficient and uninterrupted service for the user, and can effectively reduce the labor cost while improving the service quality and the user experience.
When the user completes the task by self through the man-machine conversation, for example: when a user completes a food ordering task or a ticket ordering task by self through man-machine conversation, multiple rounds of man-machine conversation are generally required. The prior art can realize the management of the multiple rounds of man-machine conversation based on the maintenance of the man-machine conversation state. When the man-machine conversation state is maintained, the man-machine conversation state can be maintained according to the input of a user, so that the currently collected information and the information which needs to be collected for completing the task can be determined through the maintenance of the man-machine conversation state. However, in the prior art, the conversation state is usually maintained manually, which results in low efficiency of conversation management, high error rate and high cost investment.
Disclosure of Invention
The invention provides a conversation management method and a conversation management device, which are used for overcoming the defects of low conversation management efficiency, high error rate and high cost investment in the prior art and realizing more efficient, more accurate and lower cost conversation management.
The invention provides a dialogue management method, which comprises the following steps:
acquiring target information;
inputting the target information into a dialogue management model, and acquiring response information corresponding to the target information output by the dialogue management model;
the target information is information input by the user in the current conversation in the process of carrying out the current conversation with the user; the dialogue management model is obtained by training based on sample information and sample response information corresponding to the sample information;
the dialogue management model is used for updating the dialogue state representation corresponding to the current dialogue based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; the dialog states are represented as one-dimensional vectors.
According to a dialog management method provided by the invention, the dialog management model comprises: a natural language understanding submodel, a dialogue state updating submodel, a dialogue decision submodel and a natural language generating submodel;
Correspondingly, the inputting the target information into a dialog management model and acquiring response information corresponding to the target information output by the dialog management model includes:
inputting the target information into the natural language understanding submodel, and acquiring semantic information corresponding to the target information output by the natural language understanding submodel;
inputting the semantic information and the target information into the dialogue state updating sub-model, and updating the dialogue state representation corresponding to the dialogue by the dialogue state updating sub-model based on the semantic information and the target information so as to obtain the updated dialogue state representation output by the dialogue state updating sub-model;
inputting the updated conversation state representation into the conversation decision submodel, and acquiring a response strategy corresponding to the target information output by the conversation decision submodel;
and inputting the response strategy and the target information into the natural language generation submodel, and acquiring the response information output by the natural language generation submodel.
According to a dialog management method provided by the present invention, the dialog state update submodel includes: the device comprises a data query unit and a dialogue state updating unit;
Correspondingly, the inputting the semantic information and the target information into the dialog state update submodel, and the dialog state update submodel updating the dialog state representation corresponding to the current dialog based on the semantic information and the target information to further obtain the updated dialog state representation output by the dialog state update submodel includes:
acquiring historical interactive information in the conversation, inputting the semantic information into the data query unit, and performing data query by the data query unit based on the semantic information so as to acquire a query result output by the data query unit;
inputting the historical interaction information, the query result and the target information into the dialogue state updating unit, and updating the dialogue state representation corresponding to the dialogue by the dialogue state updating unit based on the historical interaction information, the query result and the target information so as to obtain the updated dialogue state representation output by the dialogue state updating unit.
According to a dialog management method provided by the present invention, the step of inputting the updated dialog state representation into the dialog decision sub-model and obtaining a response policy corresponding to the target information output by the dialog decision sub-model comprises:
And the query result and the updated conversation state representation are input into the conversation decision submodel, the conversation decision submodel generates the response strategy based on the query result and the updated conversation state representation, and the response strategy output by the conversation decision submodel is further obtained.
According to a dialog management method provided by the present invention, the inputting the response policy and the target information into the natural language generation submodel and obtaining the response information output by the natural language generation submodel includes:
and inputting the query result, the response strategy and the target information into the natural language generation submodel, generating the response information by the natural language generation submodel based on the query result, the response strategy and the target information, and further acquiring the response information output by the natural language generation submodel.
According to a dialog management method provided by the present invention, after inputting the target information into a dialog management model and acquiring response information corresponding to the target information output by the dialog management model, the method further includes:
Acquiring feedback information of the user to the conversation;
updating the dialogue management model based on the feedback information.
The present invention also provides a session management apparatus, comprising:
the data acquisition module is used for acquiring target information;
the dialogue management module is used for inputting the target information into a dialogue management model and acquiring response information corresponding to the target information output by the dialogue management model;
the target information is information input by the user in the current conversation process; the dialogue management model is obtained by training based on sample information and sample response information corresponding to the sample information;
the dialogue management model is used for updating the dialogue state representation corresponding to the current dialogue based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; the dialog states are represented as one-dimensional vectors.
The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the dialog management method as described in any of the above methods when executing the program.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a session management method as described in any one of the above.
The invention also provides a computer program product comprising a computer program which, when executed by a processor, implements a dialog management method as claimed in any one of the above.
According to the dialogue management method and device, the information input by the user in the current round of dialogue in the process of the user conducting the current dialogue is obtained and is used as the target information, the target information is input into the dialogue management model, the dialogue management model updates the dialogue state representation corresponding to the current dialogue based on the target information, the response information corresponding to the target information is generated and output based on the updated dialogue state representation, the dialogue state representation corresponding to the current dialogue is a one-dimensional vector, automatic maintenance of the man-machine dialogue state can be achieved without manual definition, the dialogue state corresponding to the current dialogue is represented as a one-dimensional vector, the calculation speed of the dialogue management model can be improved, occupation of calculation resources is reduced, and more efficient, more accurate and dialogue management with lower cost investment can be achieved.
Drawings
In order to more clearly illustrate the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow diagram illustrating a session management method provided by the present invention;
FIG. 2 is a schematic diagram of a dialog management model in the dialog management method provided in the present invention;
FIG. 3 is a schematic structural diagram of a session management apparatus provided in the present invention;
fig. 4 is a schematic structural diagram of an electronic device provided in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the invention, it is to be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
It should be noted that task-oriented human-computer dialog mainly includes three parts, i.e., natural language understanding, dialog management, and natural language generation. The dialog management can control the process of the man-machine dialog, by means of which the response of the machine to the user can be determined.
When the user completes the task by self-help through the man-machine conversation, for example: when a user finishes a meal ordering task or a ticket ordering task by self-help through a man-machine conversation, the man-machine conversation can be called as one man-machine conversation from the establishment of the man-machine conversation to the end of the man-machine conversation. In case the user demand is more complex and/or comprises more restrictions, the user may state in multiple rounds in one man-machine-conversation, i.e. one man-machine-conversation may comprise multiple rounds of man-machine-conversation. A round of man-machine interaction may refer to the process of a user inputting information and the machine responding to the information. In one man-machine conversation, on one hand, a user can continuously modify and improve own requirements through multiple rounds of man-machine conversations, and on the other hand, under the condition that the user requirements cannot be specifically and definitely judged based on the input of the user, the machine can confirm the intention of the user through modes of inquiry, clarification, confirmation and the like, so that the task is completed.
The dialogue management can be generally divided into two parts, namely dialogue state maintenance and dialogue decision. When the conversation state is maintained, the state of the man-machine conversation can be maintained according to the information input by the user, so that the information collected in the man-machine conversation and the information which needs to be collected if the task needs to be completed can be determined. The dialog decision may determine the response of the machine to the information according to the man-machine dialog state after maintenance.
In a conventional dialog management method, a state of a human-computer dialog is usually maintained in a manually defined manner, that is, after an intention or an entity is extracted from information input by a user, the state of the human-computer dialog needs to be manually defined, so that the maintenance of the state of the human-computer dialog is completed. When the state of the man-machine conversation is maintained in a manually defined mode, the requirement on maintenance personnel is high, the workload of the maintenance personnel is high, the maintenance efficiency is low, the error rate is high, and therefore the efficiency of session management is low, the error rate is high, and the cost investment is high.
Moreover, the flow of the man-machine conversation in the conventional conversation management method usually needs to be configured in advance by a human. However, the preset flow content is fixed, and the corresponding response information content is rigid, so that redundancy exists in conversation management based on the traditional conversation management method, and user experience and conversation completion rate are influenced.
In view of this, the present invention provides a dialog management method. Based on the dialog management method provided by the invention, the one-dimensional dialog state expression corresponding to the dialog can be automatically updated based on the information input by the user in the current round of dialog in the process of the dialog without manual definition, so that more efficient and accurate dialog management with lower cost investment can be realized.
Fig. 1 is a flow chart illustrating a dialog management method according to the present invention. The session management method of the present invention is described below with reference to fig. 1. As shown in fig. 1, the method includes: and step 101, acquiring target information.
The target information is information input by the user in the current conversation process.
The execution subject of the embodiment of the present invention is a session management apparatus.
It should be noted that the dialog in the embodiment of the present invention is a human-machine dialog. The form of the man-machine conversation can be a voice conversation or a text conversation. The form of the dialog is not particularly limited in the embodiments of the present invention.
The session may refer to a session currently being performed by the user. The conversation starts from the time when the current ongoing man-machine conversation of the user is established to the time when the current ongoing man-machine conversation of the user is finished.
The conversation may include multiple rounds of conversations.
The current round of the dialog may refer to a process of inputting information by the user at the current moment and responding to the information input by the user at the next moment.
Specifically, the information input by the user in the current session may be acquired in a plurality of ways, for example: the method can receive the information input by the user in the current conversation sent by the user terminal; alternatively, after receiving the information input by the user in his or her turn of the dialog transmitted by the user terminal, the other electronic device may acquire the information input by the user in his or her turn of the dialog from the other electronic device.
After the information input by the current round of dialog in the process of the current dialog by the user is acquired, the information input by the current round of dialog by the user can be used as target information.
The target information may be voice information or text information. The type of the target information corresponds to the form of the user performing the session.
102, inputting target information into a conversation management model, and acquiring response information corresponding to the target information output by the conversation management model; the dialogue management model is obtained by training based on sample information and sample response information corresponding to the sample information; the dialogue management model is used for updating the dialogue state representation corresponding to the dialogue at this time based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; the dialog states are represented as one-dimensional vectors.
Before the acquired target information is input into the dialog management model, the dialog management model may be trained by using the sample information as a sample and using the sample response information corresponding to the sample information as a sample label, so as to obtain a trained dialog management model.
In particular, for any one sample session by a sample user, the sample session may include multiple sample sessions. Wherein, the sample conversation is human-machine conversation. The sample dialogue may be in the form of a voice dialogue or a text dialogue.
The sample information may include information input by the sample user in multiple sample dialogs. And the sample response information corresponding to the sample information comprises response information of the information input by the sample user in the sample conversations. Any two different sample sessions, the sample users performing the two different sample sessions may be the same or different.
The sample information and the sample response information corresponding to the sample information may be predetermined based on a priori knowledge. For example: sample information and sample response information corresponding to the sample information can be obtained from a historical dialogue record through a big data technology; alternatively, the sample information and the sample response information corresponding to the sample information may be generated based on a preconfigured flow.
After the sample information and the sample response information corresponding to the sample information are obtained, the sample information is input to a training session management model, and the training session management model may obtain a predicted session state representation based on the sample information and obtain predicted response information corresponding to the sample information based on the predicted session state representation. By comparing the predicted response information corresponding to the sample information with the sample response information corresponding to the sample information, the model parameters of the trained dialog management model can be adjusted, and the trained dialog management model can be obtained.
After the trained dialog management model is obtained, target information may be entered into the trained dialog management model.
The trained dialog management model may update the dialog state expression corresponding to the current dialog based on the target information, and obtain an updated dialog state expression corresponding to the current dialog. After obtaining the updated dialog state expression corresponding to the current dialog, the trained dialog management model may generate and output response information corresponding to the target information based on the updated dialog state expression corresponding to the current dialog. And further response information corresponding to the target information output by the trained dialogue management model can be obtained.
It should be noted that, in the embodiment of the present invention, the dialog state representation corresponding to the dialog is updated based on the target information, so that automatic maintenance of the human-machine dialog state can be implemented without manual definition, efficiency and accuracy of dialog management can be improved, and labor cost of dialog management can be reduced. Further, in the embodiment of the present invention, the dialog state corresponding to the dialog is represented as a one-dimensional vector. The information collected and the information still needing to be collected in the conversation are determined through one-dimensional vector, the calculation speed of the conversation management model is higher, the occupied calculation resources are less, and the conversation management efficiency can be further improved.
Optionally, after obtaining the response information corresponding to the target information output by the trained dialog management model, in a case that the current dialog is a voice dialog, the response information corresponding to the target information may be sent to a voice playing device, so that the voice playing device may be used to play the response information corresponding to the target information, thereby completing the current dialog; or, when the current dialog is a text dialog, the response information corresponding to the target information may be sent to the display device, so that the response information corresponding to the target information may be displayed by using the display device, thereby completing the current dialog.
According to the embodiment of the invention, the information input by the user in the current round of conversation in the process of the current conversation by the user is acquired and is used as the target information, then the target information is input into the conversation management model, the conversation management model updates the conversation state representation corresponding to the current conversation based on the target information, and generates and outputs the response information corresponding to the target information based on the updated conversation state representation, the conversation state corresponding to the current conversation is represented as a one-dimensional vector, so that the automatic maintenance of the man-machine conversation state can be realized without manual definition, the conversation state corresponding to the current conversation is represented as a one-dimensional vector, the calculation speed of the conversation management model can be improved, the occupation of calculation resources can be reduced, and the more efficient, more accurate and lower-cost conversation management can be realized.
Fig. 2 is a schematic structural diagram of a dialog management model in the dialog management method provided by the present invention, and as shown in fig. 2, the dialog management model includes: a natural language understanding submodel, a dialogue state updating submodel, a dialogue decision submodel and a natural language generating submodel.
Correspondingly, inputting the target information into the dialogue management model, and acquiring the response information corresponding to the target information output by the dialogue management model, wherein the response information comprises: and inputting the target information into the natural language understanding submodel, and acquiring semantic information corresponding to the target information output by the natural language understanding submodel.
It should be noted that the natural language understanding submodel may be trained by using the sample information as a sample and using the sample semantic information corresponding to the sample information as a sample label. The semantic information of the sample corresponding to the sample information can be obtained based on the prior knowledge.
Specifically, after the sample information is input into the natural language understanding submodel in the training, the natural language understanding submodel in the training can identify the intention and/or entity of the sample user for sample conversation by performing semantic understanding on the sample information, so as to obtain and output the predicted semantic information corresponding to the sample information. The semantic understanding may include recognizing the intention of the sample user to perform sample dialogue through semantic similarity calculation and/or classification algorithm, and implementing entity recognition through a dictionary and/or sequence labeling algorithm (e.g., algorithm such as BiLSTM + crf).
After the predicted semantic information corresponding to the sample information output by the natural language understanding submodel in training is obtained, the model parameters of the natural language understanding submodel in training can be adjusted by comparing the predicted semantic information with the sample semantic information, so that the trained natural language generating submodel is obtained.
After the trained natural language generation submodel is obtained, the target information may be input into the trained natural language understanding submodel.
The natural language understanding submodel can carry out semantic understanding on the target information, and the intention and/or the entity of the user for the conversation are/is identified by carrying out semantic understanding on the target information, so that the semantic information corresponding to the target information is obtained and output.
And inputting the semantic information and the target information into the dialogue state updating submodel, updating the dialogue state representation corresponding to the dialogue by the dialogue state updating submodel based on the semantic information and the target information, and further acquiring the updated dialogue state representation output by the dialogue state updating submodel.
The entity may refer to information collected from the dialog to the user, for example, information such as "departure place", "destination", or "departure time" collected by semantically understanding the target information, that is, the entity.
It should be noted that the dialog state update submodel may be trained by using the sample information and the sample semantic information corresponding to the sample information as samples and using the sample dialog state representation corresponding to the sample information as a sample label, so as to obtain a trained dialog state update submodel. And the sample dialog state corresponding to the sample information is represented as a one-dimensional vector. The sample dialog state representation corresponding to the sample information may be obtained based on a priori knowledge.
It should be noted that the sample session state representation corresponding to the sample information may be used to describe the collected information and the information that needs to be collected in the sample session in which the sample information is located.
Specifically, the sample information and the sample semantic information corresponding to the sample information may be input into a dialog state update sub-model in training.
The dialog state updating sub-model in training can obtain and output a predicted dialog state representation corresponding to the sample information based on the sample information and the sample semantic information corresponding to the sample information. The predicted dialog state representation corresponding to the sample information is also a one-dimensional vector.
And obtaining a prediction dialog state representation corresponding to the sample information output by the dialog state update sub-model in training, and adjusting model parameters of the dialog state update sub-model in training by comparing the prediction dialog state representation corresponding to the sample information with the sample dialog state representation corresponding to the sample information so as to obtain the trained dialog state update sub-model.
After obtaining the semantic information corresponding to the target information output by the natural language understanding submodel, the semantic information and the target information may be input into the trained dialog state updating submodel.
The dialog state updating submodel may update the dialog state representation corresponding to the dialog based on the semantic information and the target information, and acquire and output the dialog state representation after the dialog is updated.
And representing the updated conversation state into the conversation decision submodel, and acquiring a response strategy corresponding to the target information output by the conversation decision submodel.
It should be noted that the sample dialog state corresponding to the sample information may be represented as a sample, and the sample response strategy corresponding to the sample information may be used as a sample label to train the dialog decision sub-model, so as to obtain the trained dialog decision sub-model. The sample response strategy corresponding to the sample information may be obtained based on a priori knowledge.
Specifically, after the sample dialog state corresponding to the sample information is input into the dialog decision sub-model in training, the dialog decision sub-model in training may obtain and output a predicted response strategy corresponding to the sample information based on the sample dialog state.
After the predicted response strategy corresponding to the sample information output by the dialog decision sub-model in training is obtained, model parameters of the dialog decision sub-model in training can be adjusted by comparing the predicted response strategy corresponding to the sample information with the sample response strategy corresponding to the sample information, and therefore the trained dialog decision sub-model is obtained.
After obtaining the updated dialog state representation of the current dialog output by the dialog state update submodel, the updated dialog state representation can be input into the trained dialog decision submodel.
The trained dialog decision submodel can obtain and output a response strategy corresponding to the target information based on the updated dialog state representation.
And inputting the response strategy and the target information into the natural language generation submodel, and acquiring response information output by the natural language generation submodel.
It should be noted that the sample response strategy and the sample information corresponding to the sample information may be used as a sample, and the sample response information corresponding to the sample information may be used as a sample label to train the natural language generation submodel, so as to obtain a trained natural language generation submodel. The sample response information corresponding to the sample information may be obtained based on a priori knowledge.
Specifically, after the sample response strategy and the sample information corresponding to the sample information are input into the natural language generation submodel in training, the natural language generation submodel in training may generate and output the predicted response information corresponding to the sample information based on the sample response strategy and the sample information corresponding to the sample information.
After the predicted response information corresponding to the sample information output by the natural language generation submodel in training is obtained, model parameters of the natural language generation submodel in training can be adjusted by comparing the predicted response information corresponding to the sample information with the sample response information corresponding to the sample information, and therefore the trained natural language generation submodel can be obtained.
After the response strategy corresponding to the target information output by the dialog decision submodel is obtained, the response strategy corresponding to the target information and the target information can be input into the trained natural language generation submodel.
The trained natural language generation submodel may acquire and output response information corresponding to the target information based on the response strategy and the target information corresponding to the target information.
The embodiment of the invention inputs the target information into a natural language understanding submodel in a dialogue management model, obtains the semantic information corresponding to the target information output by the natural language understanding submodel, inputs the semantic information and the target information into a dialogue state updating submodel in the dialogue management model, obtains the updated dialogue state representation corresponding to the dialogue output by the dialogue state updating submodel, inputs the updated dialogue state representation into a dialogue decision submodel in the dialogue management model, obtains the response strategy corresponding to the target information output by the dialogue decision submodel, inputs the response strategy into a natural language generation submodel in the dialogue management model, obtains the response information corresponding to the target information output by the natural language generation submodel, and can realize the automatic maintenance of the man-machine dialogue state more efficiently and more accurately without manual definition, the method and the device can avoid responding to the user according to the pre-configured flow, respond to the user more flexibly, reduce the redundancy problem of response and improve the user experience and the completion rate of conversation.
Based on the content of the foregoing embodiments, the dialog state update submodel includes: the device comprises a data query unit and a dialogue state updating unit.
Correspondingly, the semantic information and the target information are input into the dialogue state updating submodel, the dialogue state updating submodel updates the dialogue state representation corresponding to the dialogue at this time based on the semantic information and the target information, and then the updated dialogue state representation output by the dialogue state updating submodel is obtained, and the method comprises the following steps: historical interaction information in the conversation is obtained, semantic information is input into the data query unit, the data query unit conducts data query based on the semantic information, and then a query result corresponding to the semantic information output by the data query unit is obtained.
Specifically, the historical interaction information in the current dialogue may include interaction information in each round of historical dialogue from the establishment of the current dialogue to the establishment of the current dialogue; or, the historical interaction information in the current conversation may further include interaction information in a preset number of rounds of historical conversations before the current round of conversation in the current conversation. The interactive information may include information input by a user and corresponding response information.
It should be noted that the preset number may be determined according to actual situations. The specific preset number is not limited in the embodiment of the present invention.
In the embodiment of the present invention, historical interaction information in the current conversation may be obtained in multiple ways, for example: after the current dialogue is established, the interactive information in each round of dialogue in the current dialogue can be cached, and after the target information is obtained, the historical interactive information in the current dialogue can be obtained from the cache.
After obtaining the semantic information corresponding to the target information output by the natural language understanding submodel, the semantic information may be input to a data query unit in the dialogue state updating submodel.
The data query unit can perform data query based on the semantic information, and obtain and output a query result corresponding to the semantic information through the data query, wherein the query result can satisfy the intention and/or entity of the user for performing the session. For example: the semantic information comprises airline tickets reserved from the A place to the B place, and the data query unit can perform data query in the internet or a predetermined database based on the semantic information, and acquire all airline ticket information from the A place to the B place as a query result to be output.
It should be noted that, in the embodiment of the present invention, the order of obtaining the historical interaction information in the current conversation, inputting the semantic information corresponding to the target information into the data query unit, and obtaining the query result corresponding to the semantic information output by the data query unit is not limited.
When the current session is the first session in the current session, the history interactive information in the current session is empty.
And inputting the historical interaction information, the query result and the target information into a conversation state updating unit, updating the conversation state representation corresponding to the current conversation by the conversation state updating unit based on the historical interaction information, the query result and the target information, and further acquiring the updated conversation state representation output by the conversation state updating unit.
It should be noted that the trained dialog state update unit in the dialog state update sub-model may be trained by using the historical sample interaction information corresponding to the sample information, the sample query result corresponding to the sample information, and the sample information as samples, and using the sample dialog state representation corresponding to the sample information as a sample label, so as to obtain the trained dialog state update unit. The historical sample interaction information corresponding to the sample information may refer to interaction information in a preset number of sample conversations before a sample conversation wheel in which the sample information is located, or interaction information in each sample conversation established from the sample conversation in which the sample information is located. And obtaining a sample query result corresponding to the sample information based on the sample semantic information corresponding to the sample information. By inputting the sample semantic information corresponding to the sample information into the data query unit, the sample query result corresponding to the sample information output by the data query unit can be obtained.
And inputting the historical sample interaction information, the sample query result and the sample information into a dialogue state updating unit in training. The dialog state updating unit during training may obtain a predicted dialog state representation corresponding to the sample information based on the historical sample interaction information, the sample query result, and the sample information. By comparing the predicted dialog state representation with the sample dialog state representation corresponding to the sample information, the model parameters of the dialog state update unit in training can be adjusted, and the trained dialog state update unit can be obtained.
After obtaining the historical interaction information in the current session and the query result corresponding to the semantic information output by the data query unit, the trained session state update unit may output the historical interaction information, the query result, and the target information.
The trained dialog state updating unit may update the dialog state representation corresponding to the dialog based on the historical interaction information, the query result, and the target information, and may further acquire and output the updated dialog state representation corresponding to the dialog.
If the current session is the first session in the current session, and the historical interaction information in the current session is empty, the query result and the target information may be input into the trained session state updating unit, and then the session state updating unit may output the first session state representation of the current session, and the first session state representation may be a random value.
The embodiment of the invention inputs semantic information and target information corresponding to the target information into a data query unit in a dialogue state updating submodel, acquires a query result corresponding to the target information output by the data query unit, and inputs the query result, the historical interaction information and the target information into a dialogue state updating unit in the dialogue state updating submodel after acquiring the historical interaction information in the current dialogue, the dialogue state updating unit updates the dialogue state representation corresponding to the current dialogue based on the query result, the historical interaction information and the target information, and further acquires the updated dialogue state representation corresponding to the current dialogue output by the dialogue state updating unit, and the invention can update the dialogue state representation corresponding to the current dialogue more flexibly and accurately based on the historical interaction information in the current dialogue and the query result corresponding to the current dialogue, therefore, the method can respond to the user more flexibly, and can improve the user experience, the completion rate of the conversation and the fluency of the conversation.
Based on the content of each embodiment, the step of representing the updated dialog state into the dialog decision sub-model and obtaining a response strategy corresponding to the target information output by the dialog decision sub-model includes: and representing the query result and the updated conversation state to input a conversation decision submodel, generating a response strategy by the conversation decision submodel based on the query result and the updated conversation state, and further acquiring the response strategy output by the conversation decision submodel.
It should be noted that, when the sub-dialog decision model is trained, the sub-dialog decision model may also be trained by using the sample query result corresponding to the sample information and the sample dialog state corresponding to the sample information as samples, and using the sample response strategy corresponding to the sample information as a sample label.
After the query result corresponding to the target information output by the data query unit and the updated dialog state representation corresponding to the dialog output by the dialog state unit are obtained, the query result and the updated dialog state representation can be input into the trained dialog decision sub-model.
The trained dialog decision submodel may generate and output a response strategy corresponding to the target information based on the query result and the updated dialog state representation.
Optionally, the reply policy corresponding to the target information may include multiple types. The types include question, confirm, complete, and select.
According to the embodiment of the invention, the query result corresponding to the target information and the updated dialog state corresponding to the dialog are represented and input into the dialog decision sub-model, so that the response strategy corresponding to the target information output by the dialog decision sub-model is obtained, the response strategy corresponding to the target information can be more accurately obtained based on the query result and the updated dialog state representation, and the user experience, the dialog completion rate and the dialog fluency can be further improved.
Based on the content of each embodiment, inputting the response policy and the target information into the natural language generation submodel, and acquiring the response information output by the natural language generation submodel, including: and inputting the query result, the response strategy and the target information into a natural language generation submodel, generating response information by the natural language generation submodel based on the query result, the response strategy and the target information, and further acquiring the response information output by the natural language generation submodel.
In general, users may express the same intention in different expression manners. The expression modes of the information input by the user are different, and the response to the information with different expression modes can be different, for example: if the information input by the user includes "i want to order air tickets for tomorrow to W city", the response information corresponding to the information may include "there are a flight, B flight and C flight for inquiring the air tickets for tomorrow to W city", and if the information input by the user includes "there is no air ticket for tomorrow to W city", the response information corresponding to the information may include "there are air tickets for tomorrow to W city, there are a flight, B flight and C flight". According to different user expression modes, different responses are carried out, and man-machine conversation can be carried out more humanizedly.
Therefore, when the natural language generation submodel is trained, the query result corresponding to the sample information, the sample response strategy corresponding to the sample information and the sample information can be used as samples, the sample response information corresponding to the sample information is used as a sample label, and the natural language generation submodel is trained to obtain the trained natural language generation submodel.
And after a query result corresponding to the target information output by the data query unit and a response strategy corresponding to the target information output by the dialogue decision submodel are obtained, the query result, the response strategy and the target information are input into the trained natural language generation submodel.
The trained natural language generation submodel may generate and output response information corresponding to the target information based on the query result, the response policy, and the target information.
According to the embodiment of the invention, the query result corresponding to the target information, the response strategy corresponding to the target information and the target information are input into the natural language generation submodel, so that the response information corresponding to the target information output by the natural language generation submodel is obtained, the response information corresponding to the target information can be more humanized and more flexible, and the user experience, the conversation completion rate and the conversation fluency can be further improved.
Based on the content of the foregoing embodiments, after inputting the target information into the session management model and acquiring the response information corresponding to the target information output by the session management model, the method further includes: and acquiring feedback information of the user to the conversation.
Specifically, the feedback information of the user to the current session may include, but is not limited to: confirmation information indicating that the user confirms the order in the current conversation, first unfinished information indicating that the user transfers the order manually in the current conversation, second unfinished information indicating that the user exits in the middle of the current conversation, satisfaction information or dissatisfaction information indicating the degree of satisfaction of the user with the current conversation, and the like.
After the conversation is finished, feedback information of the user to the conversation can be acquired through various modes, for example: satisfaction information can be obtained based on the received satisfaction score of the user to the conversation; or, the confirmation information, the first unfinished information or the second unfinished information may be acquired according to the call state of the current conversation.
The dialogue management model is updated based on the feedback information.
Specifically, after obtaining feedback information of the user for the current session, the information input by the user in the current session may be used as sample information, the feedback information may be used as a sample label, and the session management model may be updated based on the sample and the sample label.
Optionally, in a case that the feedback information includes confirmation information or satisfaction information, the feedback information may be marked as 1; in case that the feedback information includes the first incomplete information, the second incomplete information, or the unsatisfied information, the above feedback information may be marked as 0.
According to the embodiment of the invention, the feedback information of the user to the conversation is obtained, and the conversation management model is updated based on the feedback information, so that the workload of manual marking input can be reduced, the flexibility, humanization and accuracy of the conversation management model can be rapidly improved, the conversation can be more efficiently completed based on the updated conversation management model, and the user experience can be improved.
Fig. 3 is a schematic structural diagram of a session management apparatus provided in the present invention. The session management device provided by the present invention is described below with reference to fig. 3, and the session management device described below and the session management method provided by the present invention described above may be referred to in correspondence with each other. As shown in fig. 3, the apparatus includes: a data acquisition module 301 and a session management module 302.
A data obtaining module 301, configured to obtain target information.
The dialog management module 302 is configured to input the target information into the dialog management model, and obtain response information corresponding to the target information output by the dialog management model.
The target information is information input by the user in the current conversation process; the dialogue management model is obtained by training based on the sample information and the sample response information corresponding to the sample information.
The dialogue management model is used for updating the dialogue state representation corresponding to the dialogue at this time based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; dialog states are represented as one-dimensional vectors.
Specifically, the data acquisition module 301 and the session management module 302 are electrically connected.
The data obtaining module 301 may be configured to obtain, in a process of performing the current session by the user, information input by the user in the current session in a plurality of ways, for example: the method can receive the information input by the user in the current conversation sent by the user terminal; alternatively, after receiving the information input by the user in his or her turn of the dialog transmitted by the user terminal, the other electronic device may acquire the information input by the user in his or her turn of the dialog from the other electronic device.
Optionally, the dialog management device may further comprise a model update module.
The model updating module can be used for acquiring feedback information of the user to the conversation; the dialogue management model is updated based on the feedback information.
According to the embodiment of the invention, the information input by the user in the current round of conversation in the process of the current conversation by the user is acquired and is used as the target information, then the target information is input into the conversation management model, the conversation management model updates the conversation state representation corresponding to the current conversation based on the target information, and generates and outputs the response information corresponding to the target information based on the updated conversation state representation, the conversation state corresponding to the current conversation is represented as a one-dimensional vector, so that the automatic maintenance of the man-machine conversation state can be realized without manual definition, the conversation state corresponding to the current conversation is represented as a one-dimensional vector, the calculation speed of the conversation management model can be improved, the occupation of calculation resources can be reduced, and the more efficient, more accurate and lower-cost conversation management can be realized.
Fig. 4 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 4: a processor (processor)410, a communication Interface 420, a memory (memory)430 and a communication bus 440, wherein the processor 410, the communication Interface 420 and the memory 430 are communicated with each other via the communication bus 440. Processor 410 may invoke logic instructions in memory 430 to perform a dialog management method comprising: acquiring target information; inputting the target information into a dialogue management model, and acquiring response information corresponding to the target information output by the dialogue management model; the target information is information input by the user in the current conversation process; the dialogue management model is obtained by training based on the sample information and the sample response information corresponding to the sample information; the dialogue management model is used for updating the dialogue state representation corresponding to the dialogue at this time based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; the dialog states are represented as one-dimensional vectors.
In addition, the logic instructions in the memory 430 may be implemented in the form of software functional units and stored in a computer readable storage medium when the software functional units are sold or used as independent products. Based on such understanding, the technical solution of the present invention or a part thereof which substantially contributes to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product, the computer program product including a computer program, the computer program being storable on a non-transitory computer readable storage medium, the computer program, when executed by a processor, being capable of executing a dialogue management method provided by the above methods, the method including: inputting the target information into a dialogue management model, and acquiring response information corresponding to the target information output by the dialogue management model; the target information is information input by the user in the current conversation process; the dialogue management model is obtained by training based on the sample information and the sample response information corresponding to the sample information; the dialogue management model is used for updating the dialogue state representation corresponding to the dialogue at this time based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; dialog states are represented as one-dimensional vectors.
In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform a dialog management method provided by the above-mentioned methods, the method comprising: inputting the target information into a dialogue management model, and acquiring response information corresponding to the target information output by the dialogue management model; the target information is information input by the user in the current conversation process; the dialogue management model is obtained by training based on the sample information and the sample response information corresponding to the sample information; the dialogue management model is used for updating the dialogue state representation corresponding to the dialogue at this time based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; dialog states are represented as one-dimensional vectors.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment may be implemented by software plus a necessary general hardware platform, and may also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (7)
1. A method for session management, comprising:
acquiring target information;
inputting the target information into a dialogue management model, and acquiring response information corresponding to the target information output by the dialogue management model;
the target information is information input by the user in the current conversation in the process of carrying out the current conversation with the user; the dialogue management model is obtained by training based on sample information and sample response information corresponding to the sample information;
the dialogue management model is used for updating the dialogue state representation corresponding to the current dialogue based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; the dialog state is represented as a one-dimensional vector; the collected information and the information needing to be collected in the conversation are determined based on the conversation state representation;
the dialog management model comprising: a natural language understanding submodel, a dialogue state updating submodel, a dialogue decision submodel and a natural language generating submodel;
correspondingly, the inputting the target information into a dialog management model and acquiring response information corresponding to the target information output by the dialog management model includes:
Inputting the target information into the natural language understanding submodel, and acquiring semantic information corresponding to the target information output by the natural language understanding submodel;
inputting the semantic information and the target information into the dialogue state updating submodel, and updating the dialogue state representation corresponding to the dialogue by the dialogue state updating submodel based on the semantic information and the target information so as to obtain the updated dialogue state representation output by the dialogue state updating submodel;
inputting the updated conversation state representation into the conversation decision submodel, and acquiring a response strategy corresponding to the target information output by the conversation decision submodel;
inputting the response strategy and the target information into the natural language generation submodel, and acquiring the response information output by the natural language generation submodel;
the dialogue state updating submodel comprises the following steps: the device comprises a data query unit and a dialogue state updating unit;
correspondingly, the inputting the semantic information and the target information into the dialog state update submodel, and updating, by the dialog state update submodel, the dialog state representation corresponding to the current dialog based on the semantic information and the target information, so as to obtain the updated dialog state representation output by the dialog state update submodel, includes:
Obtaining historical interaction information in the conversation, inputting the semantic information into the data query unit, and performing data query by the data query unit based on the semantic information so as to obtain a query result output by the data query unit;
and inputting the historical interaction information, the query result and the target information into the dialog state updating unit, and updating a dialog state representation corresponding to the dialog based on the historical interaction information, the query result and the target information by the dialog state updating unit so as to obtain the updated dialog state representation output by the dialog state updating unit.
2. The dialog management method according to claim 1, wherein the step of inputting the updated dialog state representation into the dialog decision submodel and obtaining a response policy corresponding to the target information output by the dialog decision submodel comprises:
and the query result and the updated conversation state representation are input into the conversation decision submodel, the conversation decision submodel generates the response strategy based on the query result and the updated conversation state representation, and the response strategy output by the conversation decision submodel is further obtained.
3. The dialog management method of claim 1 wherein said entering the response policy and the target information into the natural language generation submodel to obtain the response information output by the natural language generation submodel comprises:
and inputting the query result, the response strategy and the target information into the natural language generation submodel, generating the response information by the natural language generation submodel based on the query result, the response strategy and the target information, and further acquiring the response information output by the natural language generation submodel.
4. The dialog management method according to any one of claims 1 to 3, further comprising, after the step of inputting the target information into a dialog management model and obtaining response information corresponding to the target information output by the dialog management model:
acquiring feedback information of the user to the conversation;
updating the dialogue management model based on the feedback information.
5. A dialog management device, comprising:
the data acquisition module is used for acquiring target information;
the dialogue management module is used for inputting the target information into a dialogue management model and acquiring response information corresponding to the target information output by the dialogue management model;
The target information is information input by the user in the current conversation process; the dialogue management model is obtained by training based on sample information and sample response information corresponding to the sample information;
the dialogue management model is used for updating the dialogue state representation corresponding to the current dialogue based on the target information and generating response information corresponding to the target information based on the updated dialogue state representation; the dialog state is represented as a one-dimensional vector; the collected information and the information needing to be collected in the conversation are determined based on the conversation state representation;
the dialog management model comprising: a natural language understanding submodel, a dialogue state updating submodel, a dialogue decision submodel and a natural language generating submodel;
correspondingly, the inputting the target information into a dialog management model and acquiring response information corresponding to the target information output by the dialog management model includes:
inputting the target information into the natural language understanding submodel, and acquiring semantic information corresponding to the target information output by the natural language understanding submodel;
Inputting the semantic information and the target information into the dialogue state updating submodel, and updating the dialogue state representation corresponding to the dialogue by the dialogue state updating submodel based on the semantic information and the target information so as to obtain the updated dialogue state representation output by the dialogue state updating submodel;
inputting the updated conversation state representation into the conversation decision submodel, and acquiring a response strategy corresponding to the target information output by the conversation decision submodel;
inputting the response strategy and the target information into the natural language generation submodel, and acquiring the response information output by the natural language generation submodel;
the dialogue state update sub-model comprises: the device comprises a data query unit and a dialogue state updating unit;
correspondingly, the inputting the semantic information and the target information into the dialog state update submodel, and the dialog state update submodel updating the dialog state representation corresponding to the current dialog based on the semantic information and the target information to further obtain the updated dialog state representation output by the dialog state update submodel includes:
Obtaining historical interaction information in the conversation, inputting the semantic information into the data query unit, and performing data query by the data query unit based on the semantic information so as to obtain a query result output by the data query unit;
and inputting the historical interaction information, the query result and the target information into the dialog state updating unit, and updating a dialog state representation corresponding to the dialog based on the historical interaction information, the query result and the target information by the dialog state updating unit so as to obtain the updated dialog state representation output by the dialog state updating unit.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the dialog management method according to any of claims 1 to 4 when executing the program.
7. A non-transitory computer-readable storage medium on which a computer program is stored, the computer program, when being executed by a processor, implementing the dialog management method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210308618.9A CN114490994B (en) | 2022-03-28 | 2022-03-28 | Conversation management method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210308618.9A CN114490994B (en) | 2022-03-28 | 2022-03-28 | Conversation management method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114490994A CN114490994A (en) | 2022-05-13 |
| CN114490994B true CN114490994B (en) | 2022-06-28 |
Family
ID=81488981
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210308618.9A Active CN114490994B (en) | 2022-03-28 | 2022-03-28 | Conversation management method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114490994B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116610790B (en) * | 2023-07-18 | 2023-10-10 | 阿里健康科技(中国)有限公司 | Method, device, equipment and medium for acquiring response data |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543005A (en) * | 2018-10-12 | 2019-03-29 | 平安科技（深圳）有限公司 | The dialogue state recognition methods of customer service robot and device, equipment, storage medium |
| CN112100349A (en) * | 2020-09-03 | 2020-12-18 | 深圳数联天下智能科技有限公司 | Multi-turn dialogue method and device, electronic equipment and storage medium |
| CN112365892A (en) * | 2020-11-10 | 2021-02-12 | 杭州大搜车汽车服务有限公司 | Man-machine interaction method, device, electronic device and storage medium |
| CN113918698A (en) * | 2021-10-12 | 2022-01-11 | 北京仁科互动网络技术有限公司 | Customer service processing system and device |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6601069B2 (en) * | 2015-09-01 | 2019-11-06 | カシオ計算機株式会社 | Dialog control apparatus, dialog control method, and program |
| CN110096516B (en) * | 2019-03-25 | 2022-01-28 | 北京邮电大学 | User-defined database interaction dialog generation method and system |
| CN112836025A (en) * | 2019-11-22 | 2021-05-25 | 航天信息股份有限公司 | Intention identification method and device |
| CN111460115B (en) * | 2020-03-17 | 2023-05-26 | 深圳市优必选科技股份有限公司 | Intelligent man-machine conversation model training method, model training device and electronic equipment |
-
2022
- 2022-03-28 CN CN202210308618.9A patent/CN114490994B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543005A (en) * | 2018-10-12 | 2019-03-29 | 平安科技（深圳）有限公司 | The dialogue state recognition methods of customer service robot and device, equipment, storage medium |
| CN112100349A (en) * | 2020-09-03 | 2020-12-18 | 深圳数联天下智能科技有限公司 | Multi-turn dialogue method and device, electronic equipment and storage medium |
| CN112365892A (en) * | 2020-11-10 | 2021-02-12 | 杭州大搜车汽车服务有限公司 | Man-machine interaction method, device, electronic device and storage medium |
| CN113918698A (en) * | 2021-10-12 | 2022-01-11 | 北京仁科互动网络技术有限公司 | Customer service processing system and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114490994A (en) | 2022-05-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107886949B (en) | Content recommendation method and device | |
| CN110543552B (en) | Conversation interaction method and device and electronic equipment | |
| CN103458056B (en) | Speech intention judging system based on automatic classification technology for automatic outbound system | |
| CN110837548B (en) | Answer matching method and device, electronic equipment and storage medium | |
| CN110990543A (en) | Intelligent conversation generation method and device, computer equipment and computer storage medium | |
| JP2016212405A (en) | Device and method for voice dialogue system | |
| Wang et al. | Policy learning for domain selection in an extensible multi-domain spoken dialogue system | |
| CN111191450A (en) | Corpus cleaning method, corpus entry device and computer-readable storage medium | |
| CN111739519A (en) | Dialogue management processing method, device, equipment and medium based on voice recognition | |
| US10929781B1 (en) | Systems and methods for determining training parameters for dialog generation | |
| CN110347863A (en) | Talk about art recommended method and device and storage medium | |
| CN108628908B (en) | Method, device and electronic equipment for classifying user question-answer boundaries | |
| CN116737908A (en) | Knowledge question-answering method, device, equipment and storage medium | |
| CN114490994B (en) | Conversation management method and device | |
| CN111259124A (en) | Dialogue management method, device, system and storage medium | |
| CN113918698A (en) | Customer service processing system and device | |
| CN112199486A (en) | Task type multi-turn conversation method and system for office scene | |
| CN113821620B (en) | Multi-round dialogue task processing method and device and electronic equipment | |
| CN111858874A (en) | Conversation service processing method, device, equipment and computer readable storage medium | |
| EP3908941A1 (en) | Artificial intelligence system for business processes | |
| CN116303930A (en) | Session intelligent generation method based on semantic matching and generation model | |
| CN111091011B (en) | Domain prediction method, domain prediction device and electronic equipment | |
| CN110888971B (en) | Multi-round interaction method and device for robot customer service and user | |
| CN112084300A (en) | Response information output method and device, electronic equipment and readable storage medium | |
| CN112908296A (en) | Dialect identification method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |