CN108153780B - Man-machine conversation device and method for realizing man-machine conversation - Google Patents
Man-machine conversation device and method for realizing man-machine conversation Download PDFInfo
- Publication number
- CN108153780B CN108153780B CN201611106354.XA CN201611106354A CN108153780B CN 108153780 B CN108153780 B CN 108153780B CN 201611106354 A CN201611106354 A CN 201611106354A CN 108153780 B CN108153780 B CN 108153780B
- Authority
- CN
- China
- Prior art keywords
- question
- user
- intention
- questions
- answer
- 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
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000008569 process Effects 0.000 claims abstract description 30
- 238000012512 characterization method Methods 0.000 claims description 56
- 230000004044 response Effects 0.000 claims description 21
- 230000003993 interaction Effects 0.000 claims description 18
- 238000013507 mapping Methods 0.000 claims description 10
- 230000006399 behavior Effects 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims 2
- 238000010586 diagram Methods 0.000 description 7
- 230000002452 interceptive effect Effects 0.000 description 5
- 239000003795 chemical substances by application Substances 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 241000238558 Eucarida Species 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
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
- 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
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- 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
- G06F16/35—Clustering; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The application discloses a man-machine conversation device and a method for realizing man-machine conversation thereof, comprising the following steps: in the conversation process of providing service for the user, acquiring the inquiry of the user; questions are presented to the user in a multi-turn question-answering mode based on the user questions and questions in the question database, and the user intention is determined according to the user answers. According to the technical scheme, when the user intention is uncertain, the user intention is clarified by actively asking questions to the user, the man-machine conversation process is intelligently achieved, and the accuracy of user intention identification is improved.
Description
Technical Field
The present application relates to a human-computer conversation technology, and more particularly, to a human-computer conversation apparatus and a method for implementing human-computer conversation.
Background
The intelligent man-machine conversation system adopts natural language as an interaction medium to provide services for users. The intelligent man-machine conversation system is a brand-new man-machine interaction mode, and the interaction process is more natural and efficient. For example, the well-known Eliza system was available in the sixties and allowed simple conversations with users, resulting in many users feeling as if they were communicating with real persons. After apple Siri, there are many dialog products that appear, including google Now, microsoft shanna (Cortana). A plurality of chat robots such as Microsoft ice are published in China and are widely noticed.
Most of the existing intelligent man-machine conversation systems only carry out single-round conversation, and for the inquiry of a user, namely a service object of a service, the intelligent man-machine conversation system generally carries out processing in two stages: the first stage, identifying a user intent, which means an intention that the user wishes to achieve some purpose, such as: a desire to solve a certain problem, a desire to complete a certain task, a desire to achieve a certain goal, etc.; in the second stage, AGENT, that is, a machine simulating customer service to provide customer service, performs corresponding solution or operation according to the user's intention. Here, accurately recognizing the user's intention is a key to deciding the effect of the entire intelligent human machine dialog system service. In the user intention recognition stage, the existing intelligent man-machine conversation system generally recognizes the intention of the user directly according to the expression of the user. The entire user intention identification process is passive, that is, the AGENT does not pose any question to the user's query, but rather passively determines the user's intention based on one query of the user. Thus, if the expression of the user is not clear, the user intention identified by the intelligent human-machine system may be inaccurate, and the answer or behavior given by AGENT accordingly may not reach the user intention, even resulting in errors and risks.
From the realization of the existing intelligent man-machine conversation system, the accuracy of identifying the intention of the user cannot be really ensured, and the realization of the existing intelligent man-machine conversation system depends on manual participation in a large quantity, thereby invisibly improving the cost and having longer realization period.
Disclosure of Invention
In order to solve the technical problem, the application provides a human-computer conversation device and a method for realizing human-computer conversation thereof, which can ensure the accuracy of identifying the intention of a user, reduce the realization cost and shorten the realization period.
To achieve the above object, the present application provides a human-machine interaction device, comprising: the system comprises an acquisition unit, a question answering unit and a question database for storing questions; wherein,
the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring the inquiry of a user in the conversation process of providing service for the user;
and the question-answering unit is used for proposing questions to the user in a multi-turn question-answering mode based on the user inquiry and the questions in the question database, and determining the user intention according to the user response.
Optionally, the method further comprises:
and the learning unit is used for determining the problem of the conversation based on the conversation corpus sample and storing the problem in the problem database.
Optionally, the learning unit is specifically configured to:
and extracting the questions submitted to the user by the customer service from the dialogue corpus sample, classifying the questions submitted to the user through text clustering, and storing the questions in the question database.
Optionally, the question and answer unit includes: an intention prediction module and an intention decision module; wherein,
the intention prediction module is used for predicting the probability that each question corresponding to the question in the question database is answered in the affirmative based on the question of the user and forming characterization information representing the intention of the user;
the intention decision module is used for updating the probability of positive user answer predicted in the characterization information by using the user response to the question proposed to the user based on the characterization information in the multi-turn question-answer mode; and the user intention is determined according to the characterization information obtained after multiple rounds of question answering.
Optionally, the intent decision module comprises: a question decision sub-module, and an intent decision sub-module, wherein,
the question decision submodule is used for predicting the information gain of each question in the representation information in each question-answer of the multiple rounds of question-answers, taking the question with the largest information gain as a question to be provided for a user, and providing the question for the user;
the intention decision submodule is used for updating the predicted probability that the user answers positively in the characterization information according to the response of the user to the question put forward to the user; until the user intention is determined according to the updated characterization information.
Optionally, the intent prediction module is specifically configured to:
learning from the corpus of dialogs, based on the user's query, associations between individual questions and system goals representing a solution or behavior to the user's intent; and predicting the probability of positive answer to each question corresponding to the question in the question database based on the question-answer distribution situation in the association, and forming the characterization information representing the user intention.
Optionally, a preset mapping relationship between the representation information and the user intention is stored in the intention decision sub-module;
the intent decision sub-module is specifically configured to: and finding out the user intention corresponding to the currently formed characterization information according to the mapping relation.
The application also provides a method for realizing man-machine conversation, which comprises the following steps:
in the conversation process of providing service for the user, acquiring the inquiry of the user;
questions are presented to the user in a multi-turn question-answering mode based on the user questions and questions in the question database, and the user intention is determined according to the user answers.
Optionally, the method further comprises: and determining the question of the conversation based on the conversation corpus sample and storing the question in the question database.
Optionally, the determining the question of the dialog includes:
and extracting the questions submitted to the user by the customer service from the dialogue corpus sample, classifying the questions submitted to the user through text clustering, and storing the questions in the question database.
Optionally, the presenting the question to the user in a multi-turn question-and-answer manner based on the user query and the question in the question database includes:
predicting the probability that each question corresponding to the question in the question database is answered in the affirmative based on the question of the user, and forming characterization information representing the intention of the user;
updating the probability of positive user answer predicted in the characterization information by using the user response to the question put forward to the user in the multi-turn question-answer mode based on the characterization information; and determining the user intention according to the characterization information obtained after multiple rounds of question answering.
Optionally, the updating, by using the response of the user to the question posed to the user, the probability of a predicted positive user answer in the characterization information, and the determining the user intention according to the characterization information obtained after multiple rounds of questions and answers includes:
predicting the information gain of each question in the characterization information in each question and answer of the multiple rounds of question and answers, taking the question with the largest information gain as the question to be proposed to the user in the next round of question and answer, and proposing the question to the user;
updating the predicted probability of positive user answer in the characterization information with the user response to the question made to the user; until the user intention is determined according to the updated characterization information.
Optionally, the predicting, based on the user's question, a probability that each question in the question database corresponding to the question is answered in the affirmative, the forming characterization information representing the user's intent comprising:
learning from the corpus of dialogs, based on the user's query, associations between individual questions and system goals representing a solution or behavior to the user's intent; and predicting the probability of positive answer to each question corresponding to the question in the question database based on the question-answer distribution situation in the association, and forming the characterization information representing the user intention.
Optionally, determining the user intention according to the formed representation information representing the user intention includes:
and finding out the user intention corresponding to the currently formed representation information according to the preset mapping relation between the representation information and the user intention.
The present application further provides an apparatus for implementing a human-machine dialog, comprising at least a memory and a processor, wherein,
the memory has stored therein the following executable instructions: in the conversation process of providing service for the user, acquiring the inquiry of the user; questions are presented to the user in a multi-turn question-answering mode based on the user questions and questions in the question database, and the user intention is determined according to the user answers.
The scheme provided by the application comprises the following steps: in the conversation process of providing service for the user, acquiring the inquiry of the user; questions are presented to the user in a multi-turn question-answering mode based on the user questions and questions in the question database, and the user intention is determined according to the user answers. According to the technical scheme for realizing the man-machine conversation, on the working mode of the system, when the intention of the user is uncertain, the intention of the user is clarified by actively asking questions to the user, the man-machine conversation process is intelligently realized, and the accuracy of the intention identification of the user is improved.
Additional features and advantages of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings are included to provide a further understanding of the claimed subject matter and are incorporated in and constitute a part of this specification, illustrate embodiments of the subject matter and together with the description serve to explain the principles of the subject matter and not to limit the subject matter.
FIG. 1 is a schematic diagram of the structure of a human-machine interaction device according to the present application;
FIG. 2(a) is a schematic diagram of a first embodiment of the distribution of responses to a question in the present application;
FIG. 2(b) is a schematic diagram of a second embodiment of the distribution of responses to a question in the present application;
FIG. 2(c) is a schematic diagram of a third embodiment of the distribution of responses to a question in the present application;
FIG. 2(d) is a schematic diagram of a fourth embodiment of the distribution of the responses of the present application to a question;
FIG. 3 is a flowchart of a method for implementing a human-machine conversation by the human-machine conversation device of the present application;
fig. 4 is a flowchart illustrating an embodiment of a method for implementing a human-machine conversation according to the present application.
Detailed Description
To make the objects, technical solutions and advantages of the present application more apparent, embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be noted that the embodiments and features of the embodiments in the present application may be arbitrarily combined with each other without conflict.
In one exemplary configuration of the present application, a computing device includes one or more processors (CPUs), input/output interfaces, a network interface, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory computer readable media (transient media), such as modulated data signals and carrier waves.
The steps illustrated in the flow charts of the figures may be performed in a computer system such as a set of computer-executable instructions. Also, while a logical order is shown in the flow diagrams, in some cases, the steps shown or described may be performed in an order different than here.
FIG. 1 is a schematic diagram of a structure of a human-machine interaction device according to the present application, as shown in FIG. 1, including an obtaining unit, a question answering unit, and a question database for storing questions; wherein,
the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring the inquiry of a user in the conversation process of providing service for the user;
and the question-answering unit is used for proposing questions to the user in a multi-turn question-answering mode based on the user inquiry and the questions in the question database, and determining the user intention according to the user response.
Wherein, the mode of multi-turn question answering is a process of one question answering or more than one question answering. Specifically, in the present embodiment, a round of question answering may specifically refer to a man-machine interaction device that presents a question to a user, and the user answers based on the presented question.
The questions stored in the question database are: and the questions are learned by artificial learning, cluster analysis or machine learning methods based on the dialogue corpus samples and are provided for the inquiry of the user. The corpus sample may be, for example: the dialog corpus of the real user and the customer about the natural language which achieves the user's intention, or the historical dialog data of the man-machine dialog, and so on. The questions in the question database can be updated in real time, and specifically, the questions to be asked for the user can be determined through machine learning or other methods according to the dialogue data collected over a period of time. In this way, questions in the question database can be updated in real time, and automatic feedback is formed by the question-answer corpus accumulated during the process of providing services to the user, thereby continuously improving the performance of the man-machine interaction device. Questions in the questions database include, but are not limited to: non-problem, specific problem, selection problem, etc.
Optionally, the human-machine conversation device of the application may further include a learning unit, configured to determine a question of the conversation based on the sample of the conversation corpus, for example, to learn the question posed to the query of the user, and to store the question in the question database. Through the learning unit, the questions in the question database can be obtained based on the linguistic data of the history.
For a process through artificial learning, learning questions posed against a user's query may include: suppose the user asks: "how useless did my xx ask about? "through artificial learning, customer service may present problems: "do you turn on xx service? "," do not turn on or do not turn on? "do you pay a relevant fee? "how to open xx service? "etc., which may be grouped into the same category, forming a series of proposed questions for the query xx service in a question database. There is a corresponding order of questions to be asked for these questions, for example, when customer service asks "do you turn on xx service? ", the user answers" none ", then the customer service will then ask" is it not open? "the user answers" no ", then the client asks" how to open xx service? ", the user answers" yes ", whereupon the customer service question is ended. For another example, when a customer service asks "do you open xx service? ", the user answered" open ", then the customer service will then ask" do you pay for the relevant fee? ", the user answers" none ", so far, the customer service question is ended.
Optionally, for a process through machine learning, learning raised questions for a user's query may include:
and extracting the questions submitted to the user by the customer service from the dialogue corpus sample, classifying the questions submitted to the user through text clustering, and storing the questions in the question database. Such as: the synonymous questions are classified into the same category, and each category forms a question set of the questions of the certain service of the user in the question database. The specific implementation of how to extract the question asked by the customer service to the user and how to process and classify the question in a text clustering manner can be implemented in various existing manners, and the specific implementation is also easily imaginable by those skilled in the art based on the technical scheme provided by the present application, and is not used for limiting the protection scope of the present application, and the emphasis here is that: a query for a user, such as a query for a certain service or a certain usage, is obtained by using the corpus sample of dialog, and a series of questions corresponding to the query are obtained.
The questions in the question database may be a class of questions for different services, a class of questions for different aspects of the same class of services, etc., as long as they can be referred to as a class.
Alternatively, the question-answering unit may include: an intention prediction module and an intention decision module; wherein,
an intention prediction module for predicting the probability of positive or negative answer of each question corresponding to the question in the question database based on the question of the user and forming characterization information representing the intention of the user; wherein each piece of information of the characterization information intended by the user corresponds to a question and a positive or negative probability for the question.
The intention decision module is used for updating the probability of positive (or negative) answer of the user predicted in the characterization information by using the answer of the user to the question put forward to the user in the multi-turn question-answer mode based on the characterization information; and the user intention is determined according to the characterization information obtained after multiple rounds of question answering. After the intention prediction module obtains the probability of positive answer of each question in the question database based on the query of the user and obtains the representation information of the intention of the user, the answer of the user to the proposed question is obtained based on each turn of question and answer in the multi-turn question and answer process, and the probability of positive answer (or negative answer) of the corresponding question in the obtained representation information is updated in real time, so that the intention of the user can be determined based on the updated representation information.
Optionally, the intent decision module may include: a question decision sub-module, and an intent decision sub-module, wherein,
the question decision submodule is used for predicting the information gain of each question in the representation information in each question-answer of the multiple rounds of question-answers, taking the question with the largest information gain as a question to be provided for a user, and providing the question for the user;
an intention decision sub-module for updating the predicted probability of a positive (or negative) user answer in the characterising information with a user answer to the question posed to the user; until the user intention is determined according to the updated characterization information. In a specific implementation, the number of times of multiple rounds of question answering may also be limited, that is, when the number of times of questioning reaches a preset number of times of questioning, even if the user intention cannot be determined, questioning may be stopped, and the closest user intention may be determined based on the characterization information, or the query is stopped and the user is told that the user intention cannot be determined, and the user is asked to re-submit the query, and so on.
The intention decision module can specifically propose corresponding questions (Q1, Q2, …, Qn) from a question database according to the information of the user query and answer during the dialog process of providing the service for the user, and predict the probability that the answer of each question is positive or negative to form a vector (p1, p2, …, pn) as the characterization information of the user intention.
If the user's answer to each question in the question database is known for a query from a user (e.g., for non-questions, whether the answer is positive or negative), then it is clear what question the user needs to solve or achieve what goal. That is, a series of questions in the extracted question database and their corresponding answers constitute one piece of characterizing information of the user's intention.
When a question is presented in response to the user's question, assuming that the user gives an affirmative answer, the corresponding question-answer prediction is updated to be affirmative based on the answer, as shown in fig. 2 (c); then, a question is presented again according to the question policy in the question database, assuming that the user gives a negative answer, the corresponding question-answer prediction is updated to be negative according to the question-answer, as shown in fig. 2 (d). In this way, the question is finished until the user intention is clear enough, such as a preset probability threshold is reached or the question reaches a preset question number threshold. And finally form token information representing the user's intent, i.e., a distributed token vector of user's intent (p1, p2, …, pn).
More specifically, at the beginning of the dialog, without any information of the user's intention, there is a priori a distribution of the answers to all possible questions for the user-initiated question, as shown in fig. 2 (a). After the user has set forth his question or goal (i.e., given a response) to the question posed, the answers to the individual questions may be predicted by a question-answer prediction model, forming a distributed representation of the user's intent, as shown in fig. 2 (b). Using a distributed expression of user intent, the association pt between individual questions (Q1, Q2, …, Qn) and system goals T (which may be a solution or action to achieve user intent) may be learned from the dialog corpus as F (p1, p2, …, pn). Based on the question-answer prediction distribution (p1, p2, …, pn), it can be predicted that each question Qi (i ═ 1,2, … n) may bring about an information gain infogain (Qi) for the target after obtaining a positive answer (i.e., pi ═ 1) or a negative answer (i.e., pi ═ 0), as shown in formula (1), i.e., after the question Qi is determined, the prediction for the target T is from uncertain to certain.
InfoGain(Qi)＝Entropy(pt)–pi×Entropy(pt|pi＝1)–(1-pi)×Entropy(pt|pi＝0)) (1)
And when asking questions, asking the questions with the maximum predicted information gain to the users. When the user gives an answer to a question, the corresponding question-answer prediction distribution (p1, p2, …, pn) is updated according to the question-answer; according to the updated distribution, the information gain of each question is predicted again, namely the InfoGain (Qi) is calculated again, and the question with the largest information gain is selected again to continuously ask the user; until the user intention is clear enough, for example, the predicted pt value > a preset probability threshold such as 0.9, that an accurate solution or behavior can be made, and the question is ended, or when the question initiated for the user's question reaches a preset question number threshold, the user intention is considered clear enough, and the question is ended.
The question-answer prediction model may be specifically obtained by training through machine learning according to a dialog expectation sample, and based on the question-answer prediction model, the question of the user is used as an input, so that the probability that each question in the question database is answered positively or negatively can be obtained.
Optionally, the intent prediction module is specifically configured to:
learning from the corpus of dialogs, based on the user's query, associations between individual questions and system goals representing a solution or behavior to the user's intent; and predicting the probability of affirming (or negating) the answer of each question corresponding to the question in the question database based on the question-answer distribution situation in the association to form the characterization information representing the user intention.
Optionally, the intention decision sub-module stores a mapping relationship between the characterization information and the user intention, and the intention decision module only needs to find the user intention corresponding to the currently formed characterization information according to the mapping relationship.
The distributed characterization vector of the user intention expresses all information related to the user intention in the conversation process, and finally, a solution aiming at the user intention is selected and pushed to the user according to the corresponding user intention of the characterization vector.
The problem database may be a database formed based on a set of problems of one type of service, or may be a database formed based on a set of problems of a plurality of different types of services, and the problem corresponding to each type of service corresponds to a service type. In this way, if the database is formed based on a set of questions of a plurality of different types of services, when a query of a user is obtained, a service category may be determined based on the query of the user, all questions corresponding to the service category may be found, and the representation information representing the user's intention may be composed according to responses to the questions.
Fig. 2 is a flowchart of a method for implementing a human-machine conversation by a human-machine conversation device according to the present application, as shown in fig. 2, including:
step 200: during a session for providing a service to a user, a query of the user is obtained.
Step 201: questions are presented to the user in a multi-turn question-answering mode based on the user questions and questions in the question database, and the user intention is determined according to the user answers.
In this step, the method of proposing questions to the user in a multi-turn question-answering mode based on the user inquiry and the questions in the question database includes:
predicting the probability of positive or negative answer of each question corresponding to the question in the question database based on the question of the user, and forming characterization information representing the intention of the user;
updating the probability of positive (or negative) answer of the user predicted in the characterization information by using the answer of the user to the question put forward to the user in the multi-turn question-answer mode based on the characterization information; and determining the user intention according to the characterization information obtained after multiple rounds of question answering.
Wherein,
predicting a probability of being answered in the positive (or negative) for each question in the question database corresponding to the question based on the user's question, forming characterization information indicative of the user's intent comprising:
learning from the corpus of dialogs, based on the user's query, associations between individual questions and system goals representing a solution or behavior to the user's intent; and predicting the probability of affirming (or negating) the answer of each question corresponding to the question in the question database based on the question-answer distribution situation in the association to form the characterization information representing the user intention.
If the user's answer to each question in the question database is known for a query from a user (e.g., for non-questions, whether the answer is positive or negative), then it is clear what question the user needs to solve or achieve what goal. That is, a series of questions in the extracted question database and their corresponding answers constitute one piece of characterizing information of the user's intention.
When a question is presented in response to the user's question, assuming that the user gives an affirmative answer, the corresponding question-answer prediction is updated to be affirmative based on the answer, as shown in fig. 2 (c); then, a question is presented again according to the question policy in the question database, assuming that the user gives a negative answer, the corresponding question-answer prediction is updated to be negative according to the question-answer, as shown in fig. 2 (d). And thus, according to the question-asking strategy, until the user intention is clear enough, or the question reaches a preset question-asking frequency threshold, and finishing the question asking. And finally form token information representing the user's intent, i.e., a distributed token vector of user's intent (p1, p2, …, pn).
More specifically, the present invention is to provide a novel,
wherein,
updating the predicted probability of positive (or negative) user answers in the characterization information by using responses of the user to the questions put forward to the user, and determining the user intention according to the characterization information obtained after multiple rounds of questions and answers comprises the following steps:
predicting the information gain of each question in the characterization information in each question and answer of the multiple rounds of question and answers, taking the question with the largest information gain as the question to be proposed to the user in the next round of question and answer, and proposing the question to the user;
updating the probability of positive (or negative) user answer predicted in the characterization information with user responses to the questions posed to the user; until the user intention is determined according to the updated characterization information.
More specifically, at the beginning of the dialog, without any information of the user's intention, there is a priori a distribution of the answers to all possible questions for the user-initiated question, as shown in fig. 2 (a). After the user has set forth his question or goal (i.e., given a response) to the question posed, the answers to the individual questions may be predicted by a question-answer prediction model, forming a distributed representation of the user's intent, as shown in fig. 2 (b). Using a distributed expression of user intent, the association pt between individual questions (Q1, Q2, …, Qn) and system goals T (which may be a solution or action to achieve user intent) may be learned from the dialog corpus as F (p1, p2, …, pn). Based on the question-answer prediction distribution (p1, p2, …, pn), it can be predicted that each question Qi (i ═ 1,2, … n) may bring about an information gain infogain (Qi) for the target after obtaining a positive answer (i.e., pi ═ 1) or a negative answer (i.e., pi ═ 0), as shown in formula (1), i.e., after the question Qi is determined, the prediction for the target T is from uncertain to certain.
And when asking questions, asking the questions with the maximum predicted information gain to the users. When the user gives an answer to a question, the corresponding question-answer prediction distribution (p1, p2, …, pn) is updated according to the question-answer; according to the updated distribution, the information gain of each question is predicted again, namely the InfoGain (Qi) is calculated again, and the question with the largest information gain is selected again to continuously ask the user; until the user intention is clear enough, for example, the predicted pt value > a preset probability threshold such as 0.9, that an accurate solution or behavior can be made, and the question is ended, or when the question initiated for the user's question reaches a preset question number threshold, the user intention is considered clear enough, and the question is ended.
More specifically, the present invention is to provide a novel,
wherein determining the user intent according to the formed characterization information representing the user intent comprises:
and finding out the user intention corresponding to the currently formed representation information according to the preset mapping relation between the representation information and the user intention. The distributed characterization vector of the user intention expresses all information related to the user intention in the conversation process, and finally, a solution aiming at the user intention is selected and pushed to the user according to the corresponding user intention of the characterization vector.
According to the technical scheme for realizing the man-machine conversation, in the working mode, when the intention of the user is uncertain, the intention of the user is clarified by actively asking questions to the user, the man-machine conversation process is intelligently realized, and the accuracy rate of identifying the intention of the user is improved.
The method of the present application further comprises:
questions of the conversation, such as the questions posed to the user's query, are determined based on the corpus sample of the conversation and stored in the question database.
Questions in the question database include, but are not limited to: besides the non-question, other types of questions such as special questions, choice questions, etc. are also asked.
Optionally, for the process by artificial learning, learning the posed questions for the user's query comprises: suppose the user asks: "how useless did my xx ask about? "through artificial learning, customer service may present problems: "do you turn on xx service? "," do not turn on or do not turn on? "do you pay a relevant fee? "how to open xx service? "etc., which may be grouped into the same category, forming a series of proposed questions for the query xx service in a question database. There is a corresponding order of questions to be asked for these questions, for example, when customer service asks "do you turn on xx service? ", the user answers" none ", then the customer service will then ask" is it not open? "the user answers" no ", then the client asks" how to open xx service? ", the user answers" yes ", whereupon the customer service question is ended. For another example, when a customer service asks "do you open xx service? ", the user answered" open ", then the customer service will then ask" do you pay for the relevant fee? ", the user answers" none ", so far, the customer service question is ended.
Optionally, for the process of learning by machine, learning raised questions for the user's query includes:
and extracting the questions submitted to the user by the customer service from the dialogue corpus sample, classifying the questions submitted to the user through text clustering, and storing the questions in the question database. Such as: the synonymous questions are classified into the same category, and each category forms a question set of the questions of the certain service of the user in the question database. The specific implementation of how to extract the question asked by the customer service to the user and how to process and classify the question in a text clustering manner can be implemented in various existing manners, and the specific implementation is also easily imaginable by those skilled in the art based on the technical scheme provided by the present application, and is not used for limiting the protection scope of the present application, and the emphasis here is that: a query for a user, such as a query for a certain service or a certain usage, is obtained by using the corpus sample of dialog, and a series of questions corresponding to the query are obtained.
Furthermore, on the evolution mode of the man-machine conversation device, conversation records in the man-machine conversation process are put into the conversation corpus samples so as to correct or supplement the question questions of learning, and the performance of the man-machine conversation device is continuously improved, so that the updating cost of the man-machine conversation device is reduced, and the updating speed is increased.
The method for implementing man-machine interaction in the present application is described in detail below with reference to a specific embodiment. Fig. 4 is a schematic flowchart of an embodiment of a method for implementing a human-machine conversation according to the present application, as shown in fig. 4, including the following parts:
firstly, a single-round problem recognition is carried out on the inquiry of the user, the input of the single-round interactive problem recognition model is a description text of a sentence of the inquiry of the user, such as 'my account number is stolen', and the output is a business classification corresponding to the description, such as 'how to limit'. For example, taking the query of the user, that is, the initial problem description as "my pay account number is not logged in" as an input, and obtaining that the classification target is 1000 service problems after the classification task is executed, then, it is considered that the monograph problem identification is successful, and the corresponding result of solving the problem is output; if the classification result cannot be obtained, the single-theory question recognition is considered to be failed, and a multi-round interaction process is entered, namely, the user is further asked questions and answers are obtained to help determine the intention of the user to send the questions;
then, the questions to be further proposed are predicted, and further identification is made according to the user's response to the predicted questions. The goal of problem prediction is to pick a question that will help most if a positive answer pair is obtained for a certain service. Suppose the number of questions in the question database is N, the total number of service classes is K, and suppose PiProbability that the ith (i 1 to N) question is an affirmative answer, TjP (traffic classification j | P)1P2…PN) For a conditional probability with a traffic type j (j ═ 1 to K), the information gain when the question i becomes an affirmative answer is as follows equation (1): infogain (i) for Entrophy (T) -Pi×Entropy(T|Pi＝1)–(1-Pi)×Entropy(T|Pi0)), the problem to be selected is i-argmax (infogain (i)). Wherein, a multilayer neural network FFNN can be used to model the mapping from the distribution of the answers of N questions to the distribution of the business types, the training data of the model is also derived from the sample data, for example, each call is a training data, similar to the previously established question database, the questions of the customer service questions and the answers of the users can be extracted from the telephone recording text, and the obtained questions and the answers are converted into (P)1P2…PN) The vector and the service type label of the telephone form the supervised training corpus of the data-label.
The interactive process of the multi-round flow is model prediction and question proposing, and a user gives an answer; based on the question and answer, the corresponding question and answer distribution (P) is updated1P2…PN). And recalculating the information gain of each question after the answer is changed according to the updated input, selecting the question with the largest information gain, and continuously asking the user until the intention of the user is clear enough, namely the multi-turn question identification is successful. A LSTM network with multiple rounds of interaction can give classification goals above a threshold. The multi-round problem identification can be realized by a trained multi-round interactive problem identification model, the multi-round interactive problem identification model takes all dialogue data with a user and takes the example that the user proposes a problem such as that i can not open a, AGENT further asks: "do you are seller", user replies: "yes" and the like are used for inputting and executing classification tasks, and classification targets are the same as the single-round interaction problem identification model and are 1000 business problems.
The input of the question prediction model is all the dialogue contents of the user, and the output is one of the questions in a predefined question bank. When the confidence score of the question recognition result is not higher than the set threshold, the user description information amount is considered to be insufficient, the question prediction model selects the questions which are most helpful for classification from the question library to inquire the users, and the answers of the users are collected to be distinguished again by the multi-round interactive question recognition model.
The above-mentioned single-round interaction problem recognition model, multi-round interaction problem recognition model and problem prediction model may be pre-trained models such as Deep Neural Network (DNN).
The application also provides a device for implementing man-machine conversation, which at least comprises a memory and a processor, wherein the memory stores the following executable instructions: in the conversation process of providing service for the user, acquiring the inquiry of the user; questions are presented to the user in a multi-turn question-answering mode based on the user questions and questions in the question database, and the user intention is determined according to the user answers.
Although the embodiments disclosed in the present application are described above, the descriptions are only for the convenience of understanding the present application, and are not intended to limit the present application. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the disclosure as defined by the appended claims.
Claims (11)
1. A human-computer interaction device, comprising: the system comprises an acquisition unit, a question answering unit and a question database for storing questions; wherein,
the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring the inquiry of a user in the conversation process of providing service for the user;
a question-answering unit comprising: an intention prediction module and an intention decision module;
the intention prediction module is used for predicting the probability that each question corresponding to the question in the question database is answered in the affirmative based on the question of the user and forming characterization information representing the intention of the user;
an intent decision module comprising a question decision sub-module, and an intent decision sub-module, wherein,
the question decision submodule is used for predicting the information gain of each question in the representation information in each question-answer of a plurality of rounds of question-answers, taking the question with the largest information gain as a question to be put forward to a user and putting forward the question to the user;
the intention decision submodule is used for updating the predicted probability that the user answers positively in the characterization information according to the response of the user to the question put forward to the user; until the user intention is determined according to the updated representation information;
the information gain is expressed as:
infogain (Qi) ═ entry (pt) — pi × entry (pt | pi ═ 1) - (1-pi) × entry (pt | pi ═ 0)), where pt ═ F (p1, p2, …, pn), Qi (i ═ 1,2, … n) is a question, pi ═ 1 indicates a positive answer, and pi ═ 0 indicates a negative answer.
2. The human-computer interaction device according to claim 1, further comprising:
and the learning unit is used for determining the problem of the conversation based on the conversation corpus sample and storing the problem in the problem database.
3. A human-computer interaction device according to claim 2, wherein the learning unit is specifically configured to:
and extracting the questions submitted to the user by the customer service from the dialogue corpus sample, classifying the questions submitted to the user through text clustering, and storing the questions in the question database.
4. A human-computer dialog device according to claim 1, characterized in that the intention prediction module is specifically configured to:
learning from the corpus of dialogs, based on the user's query, associations between individual questions and system goals representing a solution or behavior to the user's intent; and predicting the probability of positive answer to each question corresponding to the question in the question database based on the question-answer distribution situation in the association, and forming the characterization information representing the user intention.
5. The human-computer interaction device according to claim 1, wherein the intention decision submodule stores a mapping relationship between preset representation information and user intention;
the intent decision sub-module is specifically configured to: and finding out the user intention corresponding to the currently formed characterization information according to the mapping relation.
6. A method for implementing a human-machine conversation, comprising:
in the conversation process of providing service for the user, acquiring the inquiry of the user;
predicting the probability of each question corresponding to the question in a question database being answered in the affirmative based on the question of the user, and forming characterization information representing the user's intention;
predicting the information gain of each question in the characterization information in each question and answer of multiple rounds of question and answers, taking the question with the largest information gain as the question to be proposed to the user in the next round of question and answer, and proposing the question to the user;
updating the predicted probability of positive user answer in the characterization information with the user response to the question made to the user; until the user intention is determined according to the updated representation information;
the information gain is expressed as:
infogain (Qi) ═ entry (pt) — pi × entry (pt | pi ═ 1) - (1-pi) × entry (pt | pi ═ 0)), where pt ═ F (p1, p2, …, pn), Qi (i ═ 1,2, … n) is a question, pi ═ 1 indicates a positive answer, and pi ═ 0 indicates a negative answer.
7. The method of claim 6, further comprising: and determining the question of the conversation based on the conversation corpus sample and storing the question in the question database.
8. The method of claim 7, wherein determining the question of the conversation comprises:
and extracting the questions submitted to the user by the customer service from the dialogue corpus sample, classifying the questions submitted to the user through text clustering, and storing the questions in the question database.
9. The method of claim 6, wherein predicting a probability that each question in the question database corresponding to the question is answered in the affirmative based on the question of the user, and wherein forming characterization information indicative of the user's intent comprises:
learning from the corpus of dialogs, based on the user's query, associations between individual questions and system goals representing a solution or behavior to the user's intent; and predicting the probability of positive answer to each question corresponding to the question in the question database based on the question-answer distribution situation in the association, and forming the characterization information representing the user intention.
10. The method of claim 6, wherein determining the user intent comprises, from the formed characterization information representing the user intent:
and finding out the user intention corresponding to the currently formed representation information according to the preset mapping relation between the representation information and the user intention.
11. An apparatus for implementing a human-machine dialog, comprising at least a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the process of the method according to any one of claims 6 to 10 when executing the computer program.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611106354.XA CN108153780B (en) | 2016-12-05 | 2016-12-05 | Man-machine conversation device and method for realizing man-machine conversation |
| CN202111296453.XA CN113961690A (en) | 2016-12-05 | 2016-12-05 | Man-machine conversation device and method for realizing man-machine conversation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611106354.XA CN108153780B (en) | 2016-12-05 | 2016-12-05 | Man-machine conversation device and method for realizing man-machine conversation |
Related Child Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111296453.XA Division CN113961690A (en) | 2016-12-05 | 2016-12-05 | Man-machine conversation device and method for realizing man-machine conversation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108153780A CN108153780A (en) | 2018-06-12 |
| CN108153780B true CN108153780B (en) | 2021-11-23 |
Family
ID=62470879
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611106354.XA Active CN108153780B (en) | 2016-12-05 | 2016-12-05 | Man-machine conversation device and method for realizing man-machine conversation |
| CN202111296453.XA Pending CN113961690A (en) | 2016-12-05 | 2016-12-05 | Man-machine conversation device and method for realizing man-machine conversation |
Family Applications After (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111296453.XA Pending CN113961690A (en) | 2016-12-05 | 2016-12-05 | Man-machine conversation device and method for realizing man-machine conversation |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN108153780B (en) |
Families Citing this family (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109116743A (en) * | 2018-08-01 | 2019-01-01 | 珠海格力电器股份有限公司 | Intelligent household appliance interaction method and system |
| CN109241256B (en) * | 2018-08-20 | 2022-09-27 | 百度在线网络技术（北京）有限公司 | Dialogue processing method and device, computer equipment and readable storage medium |
| CN109299231B (en) * | 2018-09-14 | 2020-10-30 | 苏州思必驰信息科技有限公司 | Dialog state tracking method, system, electronic device and storage medium |
| CN109525480B (en) * | 2018-09-14 | 2021-09-10 | 阿里巴巴（中国）有限公司 | User question collection system and method |
| CN109460451A (en) * | 2018-09-29 | 2019-03-12 | 厦门快商通信息技术有限公司 | Actively obtain method, the intelligent customer service method and system of user information |
| JP6489670B1 (en) * | 2018-10-30 | 2019-03-27 | Ｊｅインターナショナル株式会社 | Chat system, chat bot server device, chat bot ID management device, chat agent server device, program, chat method, and chat agent method |
| CN109614464B (en) * | 2018-10-31 | 2023-10-27 | 创新先进技术有限公司 | Method and device for identifying business problems |
| CN109460459B (en) * | 2018-10-31 | 2020-09-22 | 神思电子技术股份有限公司 | Log learning-based dialogue system automatic optimization method |
| CN112673367A (en) * | 2018-10-31 | 2021-04-16 | 华为技术有限公司 | Electronic device and method for predicting user intention |
| CN109710941A (en) * | 2018-12-29 | 2019-05-03 | 上海点融信息科技有限责任公司 | User's intension recognizing method and device based on artificial intelligence |
| CN110019688A (en) * | 2019-01-23 | 2019-07-16 | 艾肯特公司 | The method that robot is trained |
| CN111552784A (en) * | 2019-02-12 | 2020-08-18 | 厦门邑通软件科技有限公司 | Man-machine conversation method based on ABC communication rule |
| CN109918492A (en) * | 2019-03-18 | 2019-06-21 | 百度在线网络技术（北京）有限公司 | System is arranged in a kind of human-computer dialogue setting method and human-computer dialogue |
| CN109977209A (en) * | 2019-03-22 | 2019-07-05 | 深圳狗尾草智能科技有限公司 | More wheel man-machine interaction methods, system, computer and medium |
| CN110059174B (en) * | 2019-04-28 | 2023-05-30 | 科大讯飞股份有限公司 | Query guiding method and device |
| CN110110048B (en) * | 2019-05-10 | 2023-06-02 | 科大讯飞股份有限公司 | Query guiding method and device |
| CN112115239A (en) * | 2019-06-20 | 2020-12-22 | 北京京东尚科信息技术有限公司 | Man-machine conversation method, device and storage medium |
| CN110442690B (en) * | 2019-06-26 | 2021-08-17 | 重庆兆光科技股份有限公司 | Query optimization method, system and medium based on probabilistic reasoning |
| CN110688459A (en) * | 2019-09-29 | 2020-01-14 | 联想(北京)有限公司 | Intelligent interaction method and intelligent equipment |
| CN111091011B (en) * | 2019-12-20 | 2023-07-28 | 科大讯飞股份有限公司 | Domain prediction method, domain prediction device and electronic equipment |
| CN113051375B (en) * | 2019-12-27 | 2024-06-14 | 阿里巴巴集团控股有限公司 | Question and answer data processing method and device based on question and answer equipment |
| CN111241237B (en) * | 2019-12-31 | 2023-05-23 | 中国建设银行股份有限公司 | Intelligent question-answer data processing method and device based on operation and maintenance service |
| CN111985248B (en) * | 2020-06-30 | 2024-08-27 | 联想(北京)有限公司 | Information interaction method and device |
| CN112232080A (en) * | 2020-10-20 | 2021-01-15 | 大唐融合通信股份有限公司 | Named entity identification method and device and electronic equipment |
| CN112395390B (en) * | 2020-11-17 | 2023-07-25 | 平安科技（深圳）有限公司 | Training corpus generation method of intention recognition model and related equipment thereof |
| CN112988991B (en) * | 2021-02-04 | 2023-04-18 | 支付宝(杭州)信息技术有限公司 | Method and system for performing anti-fraud intervention through man-machine conversation |
| CN113139045B (en) * | 2021-05-13 | 2023-05-05 | 八维(杭州)科技有限公司 | Selective question-answering method based on task-driven man-machine dialogue |
| CN113837638B (en) * | 2021-09-29 | 2024-04-26 | 支付宝(杭州)信息技术有限公司 | Method, device and equipment for determining speaking skill |
| CN117951280B (en) * | 2024-03-25 | 2024-06-28 | 可之(宁波)人工智能科技有限公司 | Intelligent learning-by-questioning-and-answering method based on generation type large model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20020072092A (en) * | 2001-03-08 | 2002-09-14 | 서정연 | Real-time Natural Language Question-Answering System Using Unit Paragraph Indexing Method |
| CN102456018A (en) * | 2010-10-18 | 2012-05-16 | 腾讯科技（深圳）有限公司 | Interactive search method and device |
| CN102722558A (en) * | 2012-05-29 | 2012-10-10 | 百度在线网络技术（北京）有限公司 | User question recommending method and device |
| JP2013143066A (en) * | 2012-01-12 | 2013-07-22 | Kddi Corp | Question and answer program, server and method which use large amount of comment texts |
| CN105468648A (en) * | 2014-09-11 | 2016-04-06 | 北大方正集团有限公司 | Method and apparatus for establishing specialized knowledge question/answer system |
-
2016
- 2016-12-05 CN CN201611106354.XA patent/CN108153780B/en active Active
- 2016-12-05 CN CN202111296453.XA patent/CN113961690A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20020072092A (en) * | 2001-03-08 | 2002-09-14 | 서정연 | Real-time Natural Language Question-Answering System Using Unit Paragraph Indexing Method |
| CN102456018A (en) * | 2010-10-18 | 2012-05-16 | 腾讯科技（深圳）有限公司 | Interactive search method and device |
| JP2013143066A (en) * | 2012-01-12 | 2013-07-22 | Kddi Corp | Question and answer program, server and method which use large amount of comment texts |
| CN102722558A (en) * | 2012-05-29 | 2012-10-10 | 百度在线网络技术（北京）有限公司 | User question recommending method and device |
| CN105468648A (en) * | 2014-09-11 | 2016-04-06 | 北大方正集团有限公司 | Method and apparatus for establishing specialized knowledge question/answer system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113961690A (en) | 2022-01-21 |
| CN108153780A (en) | 2018-06-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108153780B (en) | Man-machine conversation device and method for realizing man-machine conversation | |
| CN110334201B (en) | Intention identification method, device and system | |
| JP6448723B2 (en) | Dialog system, dialog method, and method of adapting dialog system | |
| CN110083693B (en) | Robot dialogue reply method and device | |
| CN108073600A (en) | A kind of intelligent answer exchange method, device and electronic equipment | |
| CN111078856B (en) | Group chat conversation processing method and device and electronic equipment | |
| CN116521850B (en) | Interaction method and device based on reinforcement learning | |
| CN111191450A (en) | Corpus cleaning method, corpus entry device and computer-readable storage medium | |
| CN112766990B (en) | Intelligent customer service auxiliary system and method based on multi-round dialogue improvement | |
| CN111199149A (en) | Intelligent statement clarifying method and system for dialog system | |
| CN109062944A (en) | New word consolidation method based on voice search and electronic equipment | |
| CN117688155A (en) | Service problem replying method and device, storage medium and electronic equipment | |
| CN110689359A (en) | Method and device for dynamically updating model | |
| CN113297365A (en) | User intention determination method, device, equipment and storage medium | |
| KR102039293B1 (en) | Method, apparatus and program for obtaining answer based on artificial intelligence by using clustering | |
| KR102039292B1 (en) | Method, apparatus and program for obtaining answer based on keyword | |
| CN117312521A (en) | Processing method for intelligent customer service dialogue and related products | |
| CN117332054A (en) | Form question-answering processing method, device and equipment | |
| CN117273750A (en) | Intelligent customer service system and interaction method thereof with customers | |
| CN112328774B (en) | Method for realizing task type man-machine conversation task based on multiple documents | |
| CN115544230A (en) | Question answer retrieval processing method and device | |
| KR102039294B1 (en) | Method, apparatus and program for obtaining place of the invention trips through answer based on artificial intelligence | |
| CN112989001A (en) | Question and answer processing method, device, medium and electronic equipment | |
| CN118444789B (en) | AI intelligent robot interaction method and system based on large language model | |
| CN113033160A (en) | Intention classification method and equipment for conversation and method for generating intention classification model |
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 | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 1256832 Country of ref document: HK |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |