CN116595141A

CN116595141A - Multi-round dialogue method, device, computer equipment and storage medium

Info

Publication number: CN116595141A
Application number: CN202310530043.XA
Authority: CN
Inventors: 郑文俊
Original assignee: Ping An Property and Casualty Insurance Company of China Ltd
Current assignee: Ping An Property and Casualty Insurance Company of China Ltd
Priority date: 2023-05-11
Filing date: 2023-05-11
Publication date: 2023-08-15

Abstract

The embodiment of the application belongs to the field of artificial intelligence and big data, is applied to the field of digital systems, and relates to a multi-round dialogue method which comprises the steps of acquiring an inquiry text of a user at the current moment; processing the query text through a trained natural language understanding model, and determining a user query state of the user at the current moment; determining the system dialogue state of the user at the current moment according to the user query state and the trained deep learning model; determining the dialogue skills at the current moment according to the system dialogue state at the current moment and a preset dialogue strategy; and generating an answer text of the inquiry text based on the dialogue skill at the current moment. The application also provides a multi-round dialogue device, a computer device and a storage medium. The application can determine different dialogue skills according to different dialogue states of the system, and improves the expandability of the multi-round dialogue system through the expansion of the dialogue skills.

Description

Multi-round dialogue method, device, computer equipment and storage medium

Technical Field

The present application relates to the field of artificial intelligence and big data technology, and in particular, to a multi-round dialogue method, apparatus, computer device and storage medium

Background

The multi-round dialogue system is an important field in the field of artificial intelligence, and is mainly applied to dialogue products such as personal assistants, intelligent customer service and the like. In the conversation process, the human often has topic switching, emotion expression and the like, and a topic master usually decides whether to continue the existing topic or start a new topic according to the interests and emotion expression of the opposite side. For the dialogue system, how to correctly describe the dialogue behaviors of the user and track and update the dialogue states of the dialogue system determine the effect and experience of the dialogue system, and the part is the part with the greatest technical difficulty in the multi-round dialogue system. The multi-round dialogue system is realized by adopting a finite state machine-based multi-round dialogue system, which needs to manually comb a large number of dialogue scripts, has low expandability and is difficult to cover the continuously-changing business problems.

Disclosure of Invention

The embodiment of the application aims to provide a multi-round dialogue method, a device, computer equipment and a storage medium, which are used for solving the problem that a large number of dialogue scripts are required to be manually carded in a multi-round dialogue system and the expandability is low.

In order to solve the above technical problems, the embodiment of the present application provides a multi-round dialogue method, which adopts the following technical scheme:

Acquiring an inquiry text of a user at the current moment;

processing the query text through a trained natural language understanding model, and determining a user query state of the user at the current moment;

determining the system dialogue state of the user at the current moment according to the user query state and the trained deep learning model;

and determining the dialogue skills at the current moment according to the system dialogue state at the current moment and a preset dialogue strategy.

Further, the step of obtaining the query text of the user at the current moment specifically includes:

acquiring input information of a user at the current moment;

and carrying out text processing on the input information to obtain an inquiry text of the user at the current moment.

Further, the step of processing the query text through the trained natural language understanding model to determine the user query state of the user at the current moment specifically includes:

performing natural language processing on the query text through a trained natural language understanding model, and determining semantic information corresponding to the query text;

and determining the user inquiry state of the user at the current moment according to the semantic information corresponding to the inquiry text.

Further, the step of determining the system dialogue state of the user at the current moment according to the user query state and the trained deep learning model specifically includes:

acquiring a historical system dialogue state corresponding to a historical query text of the user in a preset time period;

and inputting the historical system dialogue state and the user query state into the trained deep learning model for prediction processing, and outputting to obtain the system dialogue state of the user at the current moment.

Further, the step of determining the dialogue skill at the current moment according to the system dialogue state at the current moment and a preset dialogue strategy specifically includes:

when the system dialogue state at the current moment accords with a preset state, determining dialogue skills at the current moment through a matching strategy corresponding to the preset state;

and when the system dialogue state at the current moment does not accord with the preset condition, determining the dialogue skill at the current moment through a preset reinforcement learning strategy.

Further, the step of generating the answer text of the query text based on the dialogue skill at the current moment specifically includes:

Natural language generation is carried out on the query text, the system dialogue state and the user query state through the dialogue skills at the current moment, and a natural language generation result corresponding to the query text is obtained;

and carrying out post-processing on the natural language generation result to obtain an answer text of the query text.

Further, the step of post-processing the natural language generation result to obtain the answer text of the query text specifically includes:

sorting the natural language generation results to obtain sorting results;

carrying out fusion treatment on the sequencing results to obtain fusion results;

and carrying out style migration processing on the fusion result to obtain a final answer text of the query text.

In order to solve the above technical problems, the embodiment of the present application further provides a multi-round dialogue device, which adopts the following technical scheme:

the acquisition module is used for acquiring an inquiry text of a user at the current moment;

the first determining module is used for processing the query text through a trained natural language understanding model and determining the user query state of the user at the current moment;

The second determining module is used for determining the system dialogue state of the user at the current moment according to the user query state and the trained deep learning model;

the third determining module is used for determining the dialogue skill at the current moment according to the system dialogue state at the current moment and a preset dialogue strategy;

and the generating module is used for generating an answer text of the query text based on the dialogue skill at the current moment.

In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical schemes:

the computer device includes a memory having stored therein computer readable instructions that when executed implement the steps of the multi-round dialog method of any of the embodiments of the present application.

In order to solve the above technical problems, an embodiment of the present application further provides a computer readable storage medium, which adopts the following technical schemes:

the computer readable storage medium has stored thereon computer readable instructions which when executed by a processor implement the steps of the multi-round dialog method of any of the embodiments of the present application

Compared with the prior art, the embodiment of the application has the following main beneficial effects: after acquiring the inquiry text of the user at the current moment, determining the user inquiry state of the user at the current moment according to the inquiry text, determining the system dialogue state by utilizing the user inquiry state, and further determining the dialogue skill at the current moment according to the system dialogue state of the user at the current moment, thereby generating the answer text of the inquiry text by utilizing the dialogue skill at the current moment, determining different dialogue skills according to different system dialogue states, and improving the expandability of the multi-round dialogue system by expanding the dialogue skills.

Drawings

In order to more clearly illustrate the solution of the present application, a brief description will be given below of the drawings required for the description of the embodiments of the present application, it being apparent that the drawings in the following description are some embodiments of the present application, and that other drawings may be obtained from these drawings without the exercise of inventive effort for a person of ordinary skill in the art.

FIG. 1 is an exemplary system architecture diagram in which the present application may be applied;

FIG. 2 is a flow chart of one embodiment of a multi-round dialog method in accordance with the present application;

FIG. 3 is a flow chart of one embodiment of step S201 in FIG. 2;

FIG. 4 is a flow chart of one embodiment of step S202 of FIG. 2;

FIG. 5 is a flow chart of one embodiment of step S203 of FIG. 4;

FIG. 6 is a flow chart of one embodiment of step S204 of FIG. 2;

FIG. 7 is a flow chart of one embodiment of step S205 in FIG. 2;

FIG. 8 is a flow chart of one embodiment of step S2052 of FIG. 7;

FIG. 9 is a schematic diagram of one embodiment of a multi-round dialog device in accordance with the present application;

FIG. 10 is a schematic diagram illustrating a configuration of an embodiment of the acquisition module 901 of FIG. 9;

FIG. 11 is a schematic diagram illustrating the structure of one embodiment of the first determination module 902 of FIG. 9;

FIG. 12 is a schematic diagram of an embodiment of the second determination sub-module 903 of FIG. 11;

FIG. 13 is a schematic diagram illustrating the structure of an embodiment of the third determination module 904 of FIG. 9;

FIG. 14 is a schematic diagram of an embodiment of the generating module 905 of FIG. 9;

FIG. 15 is a schematic diagram of one embodiment of the aftertreatment sub-module 9052 of FIG. 14;

FIG. 16 is a schematic structural view of one embodiment of a computer device according to the present application.

Detailed Description

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description of the applications herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having" and any variations thereof in the description of the application and the claims and the description of the drawings above are intended to cover a non-exclusive inclusion. The terms first, second and the like in the description and in the claims or in the above-described figures, are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.

In order to make the person skilled in the art better understand the solution of the present application, the technical solution of the embodiment of the present application will be clearly and completely described below with reference to the accompanying drawings.

As shown in fig. 1, a system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. Various communication client applications, such as a web browser application, a shopping class application, a search class application, an instant messaging tool, a mailbox client, social platform software, etc., may be installed on the terminal devices 101, 102, 103.

The terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablet computers, electronic book readers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic video expert compression standard audio plane 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic video expert compression standard audio plane 4) players, laptop and desktop computers, and the like.

The server 105 may be a server providing various services, such as a background server providing support for pages displayed on the terminal devices 101, 102, 103.

It should be noted that, the multi-round dialogue method provided by the embodiment of the present application is generally executed by the server/terminal device, and accordingly, the multi-round dialogue device is generally disposed in the server/terminal device.

It should be understood that the number of terminal devices, networks and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

With continued reference to FIG. 2, a flow chart of one embodiment of a multi-round dialog method in accordance with the present application is shown. The multi-round dialogue method comprises the following steps:

step S201, acquiring the query text of the user at the current moment.

In this embodiment, the electronic device (for example, the server shown in fig. 1) on which the multi-round dialogue method operates may receive the inquiry request of the terminal device through a wired connection manner or a wireless connection manner. It should be noted that the wireless connection may include, but is not limited to, 3G/4G/5G connection, wiFi connection, bluetooth connection, wiMAX connection, zigbee connection, UWB (ultra wideband) connection, and other now known or later developed wireless connection.

The terminal device can be a terminal device with a plurality of dialogue functions, such as personal assistant, intelligent customer service and the like, and the multi-round dialogue method can be applied to services such as online claim settlement, online sales, online consultation service, online mall, online data query and the like.

Specifically, the user may input a corresponding query request through the terminal device, where the query request includes specific query data, where the query data may be different types of data such as text type data, audio type data, and picture type data. After receiving the inquiry request, the server analyzes the inquiry request to obtain corresponding inquiry data, and converts the inquiry data which is not text type data into text type data, so as to obtain the inquiry text of the user at the current moment.

It should be understood that the query text of the user before the current time may be a historical query text; the user may also continue to query after the current time, and the query text at the current time may then be multiplexed as historical query text.

Step S202, the query text is processed through the trained natural language understanding model, and the user query state of the user at the current moment is determined.

In this embodiment, the query text may be identified through a trained natural language understanding model, so as to identify semantic information of the query text, so that the query text is understood at the semantic level of the natural language, and a user query state of the user at the current moment is output according to the semantic information of the query text.

The user query status may include information such as intent category, topic category, entity category, dialogue action, emotion, etc.

Specifically, the trained natural language understanding model performs preprocessing such as error correction, reference resolution, sentence completion and the like on the query text, performs intention recognition, topic recognition, entity recognition, dialogue action recognition, emotion recognition and emotion recognition on the preprocessed query text to obtain information such as intention category, topic category, entity category, dialogue action, emotion and emotion of the user at the current moment, and obtains the system dialogue state of the user at the current moment according to the information such as the intention category, topic category, entity category, dialogue action, emotion and emotion of the user at the current moment.

In one possible embodiment, when the query text is subject to be identified, three parts of subject identification, subject discovery and subject recommendation can be included, and when the user does not provide clear subject information, the subject identification part can not obtain an accurate subject, and the subject identification part can chat with the user through the subject discovery or recommend the subject possibly interested by the user through the subject recommendation.

Step S203, determining the system dialogue state of the user at the current moment according to the user query state and the trained deep learning model.

In this embodiment, the system session state may include a theme category, a slot filling condition, a session action, a emotion, and an emotion. The dialogue state of the system can be determined according to the service requirement, and the topic category, the slot filling condition, the dialogue action, the emotion and the mode schema corresponding to the emotion can be formulated according to the service requirement.

The subject categories may be claims, sales, services, malls, car insurance, non-car insurance, and the like. The slots with different attributes can be formulated in advance according to different theme categories, and are used for filling corresponding attribute data, and taking car insurance as an example, the slots are used for filling insurance policy numbers, license plate numbers, insurance passing, insurance-giving places, contacts, contact ways and the like. The dialog may include an inquiry, information provision, confirmation, welcome, etc. The emotion may include happiness, sadness, restlessness, anger, peace, and the like. Such emotions include affirmative, negative, neutral, etc.

Step S204, determining the dialogue skills at the current moment according to the system dialogue state at the current moment and a preset dialogue strategy.

In this embodiment, different dialogue states of the system may correspond to different dialogue skills, where the dialogue skills refer to a module that generates an answer according to the context of the query text, such as a policy checking skill, a case reporting skill, a FAQ skill, and the like.

After obtaining the system dialogue state at the current moment according to the query text at the current moment, the server can match one or more corresponding dialogue skills in the dialogue skill library according to the system dialogue state at the current moment as the dialogue skills at the current moment.

It is to be appreciated that the corresponding system dialog states at different times may be different, and thus, the corresponding dialog skills at different times may also be different.

Step S205, generating answer text of the inquiry text based on the dialogue skill at the current moment.

In this embodiment, after determining the dialogue skill at the current time, the server may input the corresponding query text into the module corresponding to the dialogue skill, and the module corresponding to the dialogue skill may generate the corresponding answer according to the context of the query text.

Further, after obtaining the answer text, the answer text may be returned to the user.

After the query text of the user at the current moment is obtained, the user query state of the user at the current moment is determined according to the query text, the system dialogue state is determined by utilizing the user query state, and then the dialogue skill at the current moment is determined according to the system dialogue state of the user at the current moment, so that the answer text of the query text is generated by utilizing the dialogue skill at the current moment, different dialogue skills can be determined according to different system dialogue states, and the expandability of the multi-round dialogue system is improved by expanding the dialogue skills.

With continued reference to FIG. 3, a flowchart of one embodiment of step S201 of FIG. 2 is shown. In step S201, the method specifically includes the following steps:

s2011, input information of a user at the current moment is acquired.

In this embodiment, the input information may be an inquiry request input by the user through the terminal device. Specifically, the user may input a corresponding query request through the terminal device, where the query request includes specific query data, where the query data may be different types of data such as text type data, audio type data, and picture type data.

The terminal device sends an inquiry request input by a user at the current moment to the server so that the server obtains the input information of the user at the current moment.

S2012, the input information is subjected to text processing, and a query text of the user at the current moment is obtained.

In this embodiment, after receiving the query request, the server parses the query request to obtain corresponding query data, and converts the query data that is not text type data into text type data, so as to obtain a query text of the user at the current moment.

Specifically, after receiving the input information of the user, if the data type of the input information of the user is audio, the server can convert the voice into a text by adopting a voice recognition algorithm; if the data type of the user input information is a picture, a picture identification algorithm can be adopted to identify picture semantics, and the identified picture semantics are converted into a structured text; if the type of data entered by the user is text, no additional processing may be done.

According to the application, the input information of the non-text data type is subjected to text processing, so that the input information of different data types can be unified into one data type, and compared with the audio data type and the image data type, the text data type is easier to process.

With continued reference to FIG. 4, a flowchart of one embodiment of step S202 of FIG. 2 is shown. In step S202, the method specifically includes the following steps:

in step S2021, natural language processing is performed on the query text by the trained natural language understanding model, and semantic information corresponding to the query text is determined.

In this embodiment, the above natural language understanding model may be used to perform natural language understanding on text information, where the above natural language understanding may be implemented by using a corresponding natural language understanding model, where the natural language understanding model may include a natural language understanding module such as error correction, reference resolution, sentence completion, intention recognition, topic recognition, entity recognition, dialogue action recognition, emotion recognition, and using the natural language understanding model to perform natural language processing on an inquiry text, and may perform comprehensive understanding and structuring on input of a user, so as to perform comprehensive understanding and structuring on input of the user, and output semantic information corresponding to the inquiry text at the current time of the user, and output a user inquiry state corresponding to the inquiry text at the current time of the user, where the user inquiry state corresponding to the inquiry text at the current time of the user may include information such as an intention category, a topic category, an entity category, dialogue action, emotion, and emotion of the user.

It will be appreciated that the query text at each instant has its corresponding semantic information. The natural language understanding model may be a module based on the BERT network model or the LSTM network model.

Step S2022, determining the user query status of the user at the current moment according to the semantic information corresponding to the query text.

In this embodiment, different semantic information may correspond to different user query states, where the user query states may include information in multiple dimensions, such as an intent category, a topic category, an entity category, a dialogue action, an emotion, and an emotion. The system dialogue state may include information of multiple dimensions such as system topic category, system slot filling, system dialogue action, system emotion, and system emotion.

Optionally, after obtaining the user query state, the system dialogue state of the user at the current moment can be determined based on the deep learning neural network model. The deep learning neural network model can be trained according to the service data, and specifically, a plurality of dialogue states in the system dialogue states can be jointly modeled, and each dialogue state in the system dialogue states can be independently modeled. The service data may be determined according to a specific service scenario, for example, the service data may be data generated by a service scenario such as a user checking a policy, reporting a case, settling a claim, selling, serving, and the like.

In the case where the service data is rich, joint modeling may be employed, for example, when the number of service data exceeds a preset number. The joint modeling uses a deep learning neural network model, the deep learning neural network model can be BERT+ATTENTION, LSTM+ATTENTION and the like, the input of the deep learning neural network model is the result of natural language understanding, the result of natural language understanding comprises information such as intention category, theme category, entity category, dialogue action, emotion and emotion, and the output of the deep learning neural network model is the system dialogue state.

When the service data does not meet the preset number, each dialogue state in the system dialogue states can be independently modeled, wherein the independent modeling uses a deep learning neural network model, the deep learning neural network model can be BERT+ATTENTION, LSTM+ATTENTION and the like, the input of the deep learning neural network model is information of an intention type, a theme type, an entity type, dialogue actions, emotion and the like, the output of the deep learning neural network model is the corresponding system dialogue state, and the output of each deep learning neural network model corresponds to one type of system dialogue state. For example, a BERT-based system topic category model, a BERT-based slot fill situation model, a BERT-based system dialogue action model, a BERT-based system emotion model. The input of the system topic class model is an intention class and a topic class, and the output system dialogue state is a system topic class; the input of the slot filling condition model is an entity type, and the output system dialogue state is a slot filling condition; the input of the system dialogue action model is dialogue action, and the output system dialogue state is system dialogue action; the input of the system emotion model is emotion, and the output system dialogue state is system emotion; the input of the system emotion model is emotion, and the output system dialogue state is system emotion.

The application can draw the user inquiry state from a plurality of dimensions by carrying out natural language processing on the inquiry text, and can obtain more accurate system dialogue state by determining the system dialogue state of the user at the current moment by utilizing the user inquiry of the plurality of dimensions corresponding to the inquiry text.

With continued reference to fig. 5, a flowchart of one embodiment of step S203 of fig. 4 is shown. In step S203, the method specifically includes the following steps:

step S2031, obtaining a historical system dialogue state corresponding to a historical query text of a user in a preset time period.

In this embodiment, the preset time period may be the first time to initiate a session to the current pair

The corresponding time period between each other. In particular, the historical query text may be the query text preceding the query text at the current time during the current round of dialogue. And carrying out natural language understanding on the historical query text through the natural language understanding model to obtain a natural language understanding result of the historical query text, and determining the system dialogue state of the user at the historical moment based on the deep learning neural network model.

Of course, since the historical system dialogue state of the historical query text can be recorded, the historical system dialogue states of the first K times of the current time can be read directly from the record.

The historical system dialogue state can comprise information of multiple dimensions such as system theme category, system slot filling condition, system dialogue action, system emotion and the like at corresponding moments.

Step S2032, the historical system dialogue state and the user query state are input into the trained deep learning model to perform prediction processing, and the system dialogue state of the user at the current moment is output.

In this embodiment, the deep learning model may also be referred to as a deep learning neural network model, where the user query state is semantic information corresponding to the query text, and the user query state is a natural language understanding result obtained by performing natural language processing on the query text. The user query state may include information for multiple dimensions of intent category, topic category, entity category, dialogue action, emotion, and the like. The system dialog state of the user at the current time may be determined based on the deep learning neural network model. The input of the deep learning neural network model is a user inquiry state and a historical system dialogue state, and the output of the deep learning neural network model is the system dialogue state at the current moment. In one possible embodiment, the input of the deep learning neural network model is a user query state and a history query text, and the output of the deep learning neural network model is a system dialogue state at the current time.

Furthermore, the deep learning neural network model also comprises an attention mechanism, and the time sequence dependence in the dialogue state of the historical system is captured through the attention mechanism, so that the accuracy of the dialogue state of the system is improved.

The application combines the user inquiry state of the inquiry text with the historical system dialogue state, adds the dependency of the historical system dialogue state on time sequence, and further determines the system dialogue state of the user at the current moment, thus obtaining more accurate system dialogue state.

With continued reference to FIG. 6, a flow chart of one embodiment of step S204 of FIG. 2 is shown. In step S204, the method specifically includes the following steps:

in step S2041, when the system dialogue state at the current time accords with the preset state, the dialogue skill at the current time is determined by the matching policy corresponding to the preset state.

In this embodiment, the dialogue skills refer to a module that generates an answer according to the context of the query text, such as a policy checking skill, a case reporting skill, a FAQ skill, and the like. The above-mentioned matching policy may be rule matching, and the server stores different rules in advance to match the dialogue skills of the response. The rules may be a table of rules for system dialog states and dialog skills, and the different rules may correspond to different dialog techniques.

The preset state may be a system dialogue state in a rule table, and after obtaining the system dialogue state at the current time, the server determines the dialogue skill at the current time in the dialogue skill library through the rule table when the system dialogue state at the current time accords with the preset state.

For example, when the system topic category in the system dialogue state at the current moment is "policy check", the corresponding "policy check skill" can be found through the rule table. When the system topic category of the system dialog state at the current moment is "report case", the corresponding "report case skill" can be found out through the upper rule table.

In step S2042, when the system dialogue state at the current time does not meet the preset condition, determining the dialogue skill at the current time through the preset reinforcement learning strategy.

In this embodiment, after obtaining the system dialogue state at the current time, the server cannot determine the dialogue skill at the current time in the dialogue skill library through the rule table when the system dialogue state at the current time does not conform to the preset state. At this time, the dialogue skills at the current time may be determined by the reinforcement learning strategy.

Specifically, based on historical dialogue data, dialogue skill selection list data corresponding to a series of system dialogue states can be marked to obtain training data; based on the reinforcement learning framework, constructing a learning target and a prediction model, and performing model training on the prediction model based on training data; after training is completed, a prediction service based on a prediction model is built. And under the condition that the rule table is not recalled, inputting the system dialogue state at the current moment into a prediction model for prediction, and obtaining the predicted dialogue skill as the dialogue skill at the current moment.

It should be noted that the session skills at the current time may be one or more.

When the system dialogue state at the current moment accords with the preset state, the dialogue skill at the current moment is determined through the rule table, and when the system dialogue state at the current moment does not accord with the preset condition, the dialogue skill at the current moment is predicted through the prediction model of the reinforcement learning strategy, so that the system expansibility during dialogue skill selection is improved, and system waiting or dialogue skill misselection caused by non-recall of the rule table is avoided.

With continued reference to fig. 7, a flow chart of one embodiment of step S205 in fig. 2 is shown. In step S205, the method specifically includes the following steps:

in step S2051, natural language generation is performed on the query text, the system dialogue state, and the user query state by the dialogue skill at the current time, and a natural language generation result corresponding to the query text is obtained.

In this embodiment, after obtaining the dialogue skills at the current moment, the server may select a natural language generating module corresponding to the dialogue skills, and the natural language generating module may generate a corresponding natural language generating result based on the query text, the system dialogue state and the user query state information. One dialog skill may include multiple natural language generation modules, with different types of dialog skills corresponding to different combinations of natural language generation modules.

For example, the dialog skills are task-type dialog skills, and may include a natural language generation module combination for generating task-type results; the dialogue skills are task question-answering skills, and can comprise a natural language generation module combination for generating question-answering results; the conversation skill is a boring conversation skill, and may include a natural language generation module combination for generating a boring result; the dialog skills are other dialog skills, and may include a natural language generation module combination for generating other types of results.

The natural language generating module may be an existing natural language generating module such as GPT, UNILM, T5, etc.

Step S2052, post-processing is performed on the natural language generation result to obtain the answer text of the query text.

In this embodiment, the post-processing may be sorting, fusing, stylizing, or the like. After the server obtains the natural language generation result, the answer text which is more in line with the habit of the user can be obtained through post-processing.

According to the application, the natural language generation module is used for carrying out natural language generation on the query text, the dialogue state of the system and the semantic information corresponding to the query text to obtain the natural language generation result corresponding to the query text, so that the natural language generation result which is more in line with the intention of the user can be obtained, the natural language generation result is subjected to post-processing, the answer text which is more in line with the habit of the user can be obtained, and the user experience is improved.

With continued reference to fig. 8, a flow chart of one embodiment of step S2052 of fig. 7 is shown. In step S2052, the following steps are specifically included:

and step S20521, sorting the natural language generation results to obtain sorting results.

In this embodiment, after the server obtains the natural language generating results corresponding to the n dialogue skills, it may be determined, based on the semantic ranking algorithm, that the m natural language generating results should be returned to the user. Wherein m is less than or equal to n. In the semantic ranking algorithm, the semantic similarity between the natural language generation result and the query text can be calculated, ranking is performed according to the semantic similarity from high to low, and m natural language generation results with the highest semantic similarity are selected to be returned to the user.

And step S20522, carrying out fusion processing on the sequencing result to obtain a fusion result.

In this embodiment, after m natural language generating results are obtained, semantic fusion may be performed on the m natural language generating results, so that the whole natural language generating result is closer to the expression habit of the user, and further, the answer text becomes anthropomorphic, rather than a segment-by-segment.

And step S20523, performing style migration processing on the fusion result to obtain a final answer text of the query text.

In this embodiment, the style migration is to perform style migration on the answer text, so that the returned answer text has a uniform style, such as fun, professional, lovely, etc. The target style of the style migration can be determined according to the user portrait, so that an answer text which is more in line with the habit of the user can be obtained, and the user experience is improved.

After the final answer text is obtained, the final answer text is returned to the user so that the user can obtain the answer of the queried content.

According to the application, the natural language generation results are subjected to sorting processing, fusion processing and style migration processing, so that the answer text which is more in line with the habit of the user can be obtained, and the user experience is improved.

The embodiment of the application can acquire and process the related data required by the knowledge base configuration based on the artificial intelligence technology. Among these, artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use knowledge to obtain optimal results.

Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions. The application can be applied to the fields of electronic commerce and intelligent office, thereby promoting the construction of the digital industry.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by computer readable instructions stored in a computer readable storage medium that, when executed, may comprise the steps of the embodiments of the methods described above. The storage medium may be a nonvolatile storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a random access Memory (Random Access Memory, RAM).

It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited in order and may be performed in other orders, unless explicitly stated herein. Moreover, at least some of the steps in the flowcharts of the figures may include a plurality of sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, the order of their execution not necessarily being sequential, but may be performed in turn or alternately with other steps or at least a portion of the other steps or stages.

With further reference to fig. 9, as an implementation of the method shown in fig. 2, the present application provides an embodiment of a multi-round dialog device, which corresponds to the embodiment of the method shown in fig. 2, and which is particularly applicable to various electronic devices.

As shown in fig. 9, the multi-turn dialogue device 900 according to the present embodiment includes: an acquisition module 901, a first determination module 902, a second determination module 903, a third determination module, and a generation module 904. Wherein:

the acquiring module 901 is configured to acquire an inquiry text of a user at a current moment;

a first determining module 902, configured to process the query text through a trained natural language understanding model, and determine a user query state of the user at the current moment;

a second determining module 903, configured to determine a system dialogue state of the user at the current moment according to the user query state and the trained deep learning model;

a third determining module 904, configured to determine a dialogue skill at the current time according to the system dialogue state at the current time and a preset dialogue policy;

the generating module 905 is configured to generate an answer text of the query text based on the dialogue skill at the current time.

In this embodiment, after the query text of the user at the current time is obtained, the user query state of the user at the current time is determined according to the query text, and the system dialogue state is determined by using the user query state, and then the dialogue skill at the current time is determined according to the system dialogue state of the user at the current time, so that the answer text of the query text is generated by using the dialogue skill at the current time, different dialogue skills can be determined according to different system dialogue states, and the expandability of the multi-round dialogue system is improved by expanding the dialogue skills.

Referring to fig. 10, which is a schematic structural diagram of one embodiment of the acquisition module 901 in fig. 9, in some alternative implementations of this embodiment, the acquisition module 901 includes an acquisition sub-module 9011 and a processing sub-module 9012. Wherein:

the obtaining sub-module 9011 is configured to obtain input information of a user at a current time;

the processing sub-module 9012 is configured to perform text processing on the input information, so as to obtain an inquiry text of the user at the current moment.

In this embodiment, by performing text processing on input information of a non-text data type, input information of different data types can be unified into one data type, and text data types are easier to process than audio data types and image data types.

Referring to fig. 11, which is a schematic structural diagram of an embodiment of the first determining module 902 in fig. 9, the first determining module 902 includes a first determining sub-module 9021 and a second determining sub-module 9022. Wherein:

the first determining submodule 9021 is used for performing natural language processing on the query text through a trained natural language understanding model to determine semantic information corresponding to the query text;

the second determining submodule 9022 is configured to determine, according to semantic information corresponding to the query text, a user query state of the user at a current time.

In this embodiment, through performing natural language processing on the query text, the user query state can be described from multiple dimensions, and the system dialogue state of the user at the current moment can be determined by using the user query of multiple dimensions corresponding to the query text, so that a more accurate system dialogue state can be obtained.

Referring to fig. 12, which is a schematic structural diagram of an embodiment of the second determining module 903 in fig. 11, the second determining module 903 includes an obtaining unit 9031 and a determining unit 9032. Wherein:

the acquiring unit 9031 acquires a historical system dialogue state corresponding to a historical query text of a user within a preset time period;

The determining unit 9032 is configured to input the historical system dialogue state and the user query state to the trained deep learning model for prediction processing, and output the predicted deep learning model to obtain the system dialogue state of the user at the current moment.

In this embodiment, by combining the user query state of the query text with the historical system dialogue state, the dependency of the historical system dialogue state is added in time sequence, so as to determine the system dialogue state of the user at the current moment, and thus, a more accurate system dialogue state can be obtained.

Referring to fig. 13, which is a schematic structural diagram of an embodiment of the second determining module 904 in fig. 9, the third determining module 904 includes a third determining sub-module 9041 and a fourth determining sub-module 9042. Wherein:

the third determining submodule 9041 is configured to determine, when the system dialogue state at the current time accords with a preset state, a dialogue skill at the current time through a matching policy corresponding to the preset state;

the fourth determining submodule 9042 is configured to determine, when the system dialogue state at the current time does not meet the preset condition, a dialogue skill at the current time through a preset reinforcement learning strategy.

In this embodiment, when the system dialogue state at the current time accords with the preset state, the dialogue skill at the current time is determined through the rule table, and when the system dialogue state at the current time does not accord with the preset condition, the dialogue skill at the current time is predicted through the prediction model of the reinforcement learning strategy, so that the system expansibility during dialogue skill selection is improved, and system waiting or dialogue skill misselection caused by non-recall of the rule table is avoided.

Referring to fig. 14, which is a schematic structural diagram of an embodiment of the generating module 905 in fig. 9, the generating module 905 includes a generating sub-module 9051 and a post-processing sub-module 9052.

Wherein:

the generating sub-module 9051 is configured to perform natural language generation on the query text, the system dialogue state, and the user query state according to the dialogue skill at the current time, so as to obtain a natural language generation result corresponding to the query text;

the post-processing sub-module 9052 is configured to post-process the natural language generation result to obtain an answer text of the query text.

In this embodiment, the natural language generating module generates natural language for the query text, the dialogue state of the system and the semantic information corresponding to the query text according to the dialogue skill at the current moment to obtain a natural language generating result corresponding to the query text, so as to obtain a natural language generating result more conforming to the user intention, and post-processes the natural language generating result to obtain an answer text more conforming to the user habit, thereby improving the user experience.

Referring to fig. 15, which is a schematic structural diagram of an embodiment of the post-processing submodule 9052 in fig. 14, the post-processing submodule 9052 includes a sorting unit 90521, a fusion unit 90522, and a style migration unit 90523. Wherein:

the ranking unit 90521 is configured to perform ranking processing on the natural language generating result to obtain a ranking result;

the fusion unit 90522 is used for carrying out fusion processing on the sequencing results to obtain fusion results;

and the style migration unit 90523 is used for performing style migration processing on the fusion result to obtain a final answer text of the query text.

In this embodiment, the ranking process, the fusion process and the style migration process are performed on the natural language generation result, so that an answer text more conforming to the habit of the user can be obtained, and the user experience is improved.

In order to solve the technical problems, the embodiment of the application also provides computer equipment. Referring specifically to fig. 16, fig. 16 is a basic structural block diagram of a computer device according to the present embodiment.

The computer device 16 includes a memory 161, a processor 162, and a network interface 163 communicatively coupled to each other via a system bus. It should be noted that only computer device 16 having components 161-163 is shown in the figures, but it should be understood that not all of the illustrated components need be implemented, and that more or fewer components may be implemented instead. It will be appreciated by those skilled in the art that the computer device herein is a device capable of automatically performing numerical calculations and/or information processing in accordance with predetermined or stored instructions, the hardware of which includes, but is not limited to, microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASICs), programmable gate arrays (fields-Programmable Gate Array, FPGAs), digital processors (Digital Signal Processor, DSPs), embedded devices, etc.

The computer equipment can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing equipment. The computer equipment can perform man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch pad or voice control equipment and the like.

The memory 161 includes at least one type of readable storage medium including flash memory, hard disk, multimedia card, card memory (e.g., SD or DX memory, etc.), random Access Memory (RAM), static Random Access Memory (SRAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), programmable Read Only Memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 161 may be an internal storage unit of the computer device 16, such as a hard disk or a memory of the computer device 16. In other embodiments, the memory 161 may also be an external storage device of the computer device 16, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Card (Flash Card) or the like, which are provided on the computer device 16. Of course, the memory 161 may also include both internal storage units of the computer device 16 and external storage devices. In this embodiment, the memory 161 is typically used to store an operating system and various application software installed on the computer device 16, such as computer readable instructions for a multi-turn dialog method. Further, the memory 161 may be used to temporarily store various types of data that have been output or are to be output.

The processor 162 may be a central processing unit (Central Processing Unit, CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments. The processor 162 is generally used to control the overall operation of the computer device 16. In this embodiment, the processor 162 is configured to execute computer readable instructions stored in the memory 161 or process data, such as computer readable instructions for executing the multi-round dialog method.

The network interface 163 may include a wireless network interface or a wired network interface, and the network interface 163 is typically used to establish communication connections between the computer device 16 and other electronic devices.

In this embodiment, the influence data of the system in the knowledge base may be used to generate multiple dialogues influenced by the target to-be-tested system to the user, so that the user may intuitively obtain the influence range of the target to-be-tested system, and may evaluate the influence of modification more accurately and more quickly according to the influence range of the target to-be-tested system, eliminate irrelevant interference points, simplify regression use cases, avoid that partial correlations are found to be not processed when the previous evaluation of the influence points are omitted, and save development time and supplement logic when the previous evaluation of the influence points are co-tuned, so that serious possible scheme design needs to override and redo the problem, thereby improving the efficiency of project development.

The present application also provides another embodiment, namely, a computer-readable storage medium storing computer-readable instructions executable by at least one processor to cause the at least one processor to perform the steps of the multi-round dialog method as described above.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present application.

It is apparent that the above-described embodiments are only some embodiments of the present application, but not all embodiments, and the preferred embodiments of the present application are shown in the drawings, which do not limit the scope of the patent claims. This application may be embodied in many different forms, but rather, embodiments are provided in order to provide a thorough and complete understanding of the present disclosure. Although the application has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described in the foregoing description, or equivalents may be substituted for elements thereof. All equivalent structures made by the content of the specification and the drawings of the application are directly or indirectly applied to other related technical fields, and are also within the scope of the application.

Claims

1. A multi-round dialog method, comprising the steps of:

acquiring an inquiry text of a user at the current moment;

determining the dialogue skills at the current moment according to the system dialogue state at the current moment and a preset dialogue strategy;

and generating an answer text of the inquiry text based on the dialogue skill at the current moment.

2. The multi-turn conversation method of claim 1 wherein the step of obtaining the query text of the user at the current time comprises:

acquiring input information of a user at the current moment;

3. The multi-turn dialogue method according to claim 2, wherein the step of determining the user query status of the user at the current moment by processing the query text through a trained natural language understanding model specifically comprises:

4. A multi-turn conversation method as claimed in claim 3 wherein said step of determining the system conversation state of said user at said current time based on said user query state and a trained deep learning model comprises:

5. The multi-turn conversation method of claim 4 wherein the step of determining the conversation skills at the current time according to the system conversation state at the current time and a preset conversation strategy specifically comprises:

6. The multi-turn conversation method of claim 5 wherein the step of generating answer text for the query text based on the conversation skills at the current time, comprises:

7. The multi-turn dialogue method according to claim 5 or 6, wherein the step of post-processing the natural language generation result to obtain the answer text of the query text specifically comprises:

sorting the natural language generation results to obtain sorting results;

8. A multi-round dialog device, comprising:

9. A computer device comprising a memory having stored therein computer readable instructions which when executed by a processor implement the steps of the multi-round dialog method of any of claims 1 to 7.

10. A computer readable storage medium having stored thereon computer readable instructions which when executed by a processor implement the steps of the multi-round dialog method of any of claims 1 to 7.