CN113139044A - Question-answering multi-turn dialogue method supporting multi-intention switching for question-answering system - Google Patents
Question-answering multi-turn dialogue method supporting multi-intention switching for question-answering system Download PDFInfo
- Publication number
- CN113139044A CN113139044A CN202110504405.9A CN202110504405A CN113139044A CN 113139044 A CN113139044 A CN 113139044A CN 202110504405 A CN202110504405 A CN 202110504405A CN 113139044 A CN113139044 A CN 113139044A
- Authority
- CN
- China
- Prior art keywords
- intention
- slot
- state
- information
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000000605 extraction Methods 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 238000012790 confirmation Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 3
- 238000011084 recovery Methods 0.000 abstract description 2
- 239000013589 supplement Substances 0.000 abstract description 2
- 238000012545 processing Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 3
- 239000000047 product Substances 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000005352 clarification Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000011160 research Methods 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
- G06F16/3343—Query execution using phonetics
-
- 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
- G06F16/3344—Query execution using natural language analysis
-
- 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
- G06F16/338—Presentation of query results
Abstract
The invention provides a question-answering multi-turn dialogue method supporting multi-intention switching for a question-answering system, which realizes task-driven multi-intention multi-turn multi-talk management. The method divides multiple rounds of conversations into an intention completing state and an intention inheriting state, realizes interruption and recovery of the intention in a stack mode, enables a question answering system to process multiple user intentions, switches among the intentions, introduces historical slot position information, and can supplement default slot position information of the user through context information.
Description
Technical Field
The invention relates to the technical field of natural language intelligent question answering, in particular to a question answering multi-turn dialogue method supporting multi-intention switching and used for a question answering system.
Background
With the continuous development of natural language technology, more and more intelligent question-answering system products such as chat robots, voice assistants, automatic customer service and the like have come into the lives of people, and key technologies of the intelligent question-answering system products also become one of the hot spots of research. The multi-round conversation technology is divided into two types, one is multi-round conversation of an open domain, and the other is multi-round conversation of a closed domain. The open-domain multi-turn dialog is mainly used for the chat robot, an end-to-end mode is used, such as a probability generation model based on maximum likelihood estimation maximization reply and dialog strategy learning based on a deep reinforcement learning method, and the generated reply can be used for guiding continuous dialog according to the information by considering recent dialog history information through a large amount of multi-turn dialog training data. While the multi-round conversation of the closed domain is mainly used for task-driven conversation tasks, such as voice assistants, most of related work is carried out based on a slot filling mode, in natural conversation of human-computer interaction, short sentences and ellipses are used in human conversation, conversation contents are difficult to be handed over clearly in a single-round conversation, and conversation experience of most products is poor. More semantic information can be collected through multiple rounds of conversation, and the intention of the questioner can be understood more accurately.
Thus, a closed multi-turn dialog is a way to get the necessary information to get the explicit user instructions finally, after the initial user intent is made explicit in the man-machine dialog. Generally, a plurality of rounds of dialogue processing correspond to the processing of one thing, and by means of a finite state machine, clarification intentions are gradually asked backwards, and word slots are filled until the slot positions are complete. However, in practical applications, a user may process multiple transactions simultaneously, and common information between the transactions is shared, and the common multi-turn dialog method has the following defects in processing the applications: 1) when the processing of a certain intention is not finished, other intents cannot be processed, and the processing of the previous intention is required to be finished; 2) when processing two association intents, all slot position information still needs to be acquired step by step and reversely, and the default word slot cannot be supplemented from the context environment.
Disclosure of Invention
The purpose of the invention is as follows: the technical problem to be solved by the present invention is to provide a question-answering multi-turn dialogue method for a question-answering system supporting multi-intention switching, which is used for realizing task-driven multi-turn multi-dialogue management, so that the question-answering system can help a user collect complete task execution information and respond to a task execution request of the user, and specifically comprises the following steps:
step 1, when receiving question input, firstly reading the stored intention state information, judging the current intention state, if the intention state is an intention completion state, turning to step 2, and if the intention inheritance state is an intention inheritance state, turning to step 6;
step 2, performing intention identification on the question, and if the step 3 is successfully executed, returning to an invalid state unsuccessfully;
step 3, extracting the slot position of the question sentence, judging whether the slot position is complete or not, if so, returning to success, storing slot position information, executing step 8, and if not, executing step 4;
step 4, loading historical slot position information, judging whether the slot position is complete, if so, confirming to a user, if so, returning success, storing the slot position information, executing step 9, and if not, executing step 5;
step 5, storing the existing slot position information, converting the intention state into an intention inheritance state, inquiring the slot position information in a reverse manner, and turning to the step 1;
step 6, judging whether the input is slot position information to be supplemented, if not, turning to step 7, if so, judging whether the slot position is complete, if so, confirming to a user, if so, returning to success, storing the slot position information, executing step 8, and if not, returning to step 5;
step 7, identifying the intention, judging whether the intention is a new intention, if so, storing the previously processed intention into a stack, executing the step 3, and returning to an invalid state unsuccessfully;
and 8, judging whether the stack has an incomplete intention, if so, outputting an incomplete intention by the pop (the stack is a general data structure, and the pop and the push are data fetching and storing operations thereof), converting the state into an intention inheriting state, and if not, returning to execute the step 1.
In step 1, the intention state information is an identifier for saving an intention state and is used for judging the current intention state; the intention state comprises an intention completion state and an intention inheritance state, wherein the intention completion state refers to a state that the current instruction is processed and no unprocessed intention exists in the stack, or is an initial state of the question-answering system, and the intention inheritance state refers to a state that slot information of the current intention is incomplete and the slot information needs to be input by a user. The intention refers to a certain service which needs consultation or completion in the user conversation; the slot refers to parameter information which needs to be provided by the user in the session after completing the service.
In step 1, the slot information refers to a slot name and a corresponding value.
In step 4, the confirmation to the user means that the obtained complete slot position information generates a standard statement sentence pattern of the current intention, asks the user back and confirms the validity of the slot position.
In step 6, the determining whether the input is slot information to be supplemented includes: the slot matching method is divided into three types, namely enumeration type slot matching, numerical value type slot matching and entity type slot matching: wherein:
the enumeration type slot matching judges whether slot information to be supplemented is in a dictionary or not by judging whether input is in the dictionary or not in a dictionary mode;
the numerical slot matching is matched through a rule, and whether the input meets the format requirement and the boundary requirement of the slot is judged;
and the actual slot matching judges whether the extracted entity meets the parameter type required by the slot or not in the ways of entity extraction and type judgment.
In step 6, the determining whether the input is slot information to be supplemented means that the current and all unfinished intentions in the stack are matched, the slot of the current intention is matched first, and when the input can be matched with more than two slots, all intentions in the stack are sorted in a LambdaMART sorting learning mode, and the adopted characteristics are as follows: location information intended in the stack, degree of matching of the input to the slot name.
The position information refers to pop operation times executed by acquiring an intention from a stack, and the current intention is set to be 0;
the matching degree of the input Q and the slot position name is obtained through neural network learning of a BERT (the BERT is the name of a language model and a special name) and an MLP (multi-layer perceptron, namely a fully-connected neural network), the input is recorded as Q, the slot position name is recorded as C model training input and formed by splicing Q and C, and the input is marked as 1(Q and C are matched) and 0(Q and C are not matched).
In step 6, the confirmation to the user means that the obtained complete slot position information generates a standard statement sentence pattern of the current intention, asks the user back, and confirms the validity of the slot position.
In step 7, the storing of the previously processed intentions into the stack means that currently processed intention classification information and slot information are stored and stored in a stack form, the intention classification information refers to an intention category obtained by performing intention recognition on a question, and the intention classification information is a processing flow of a question-answering system during intention recognition.
Has the advantages that: compared with the traditional multi-turn dialogue method with single intention processing, the method has the following advantages: 1) an intention interruption recovery mechanism is introduced, so that a question answering system can process a plurality of user intentions and switch among the intentions; 2) historical slot information is introduced, and the default slot information of the user can be supplemented through context information.
Drawings
The foregoing and/or other advantages of the invention will become further apparent from the following detailed description of the invention when taken in conjunction with the accompanying drawings.
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
As shown in fig. 1, the present invention provides a question-answering multi-turn dialogue method supporting multi-intent switching for a question-answering system, comprising the following steps:
step 1, when receiving question input, firstly judging the current intention state, namely the intention completion state or the intention inheritance state, and turning to step 2 when the intention completion state is in the current intention state, or turning to step 6 when the intention completion state is not in the current intention inheritance state;
step 2, if the intention is in an intention completion state, which indicates that no uncompleted intention exists, performing intention recognition on a question, wherein the intention recognition is generally recognized by a sentence pattern matching or short text classification method, sometimes some situation information is introduced to judge, and if the step 3 is successfully executed, an invalid state is returned unsuccessfully;
step 3, extracting the slot position of the question sentence, judging whether the slot position is complete or not, if so, returning to success, storing slot position information, executing step 8, and if not, executing step 4; the slot extraction generally adopts a regular matching or sequence labeling mode, or a parameter type and a parameter value predefined by the intention;
step 4, loading historical slot position information, judging whether the slot position is complete, if so, confirming to a user, if so, returning success, storing the slot position information, executing the step 9, and if not, executing the step 5;
step 5, storing the existing slot position information, converting the intention state into an intention inheritance state, inquiring the slot position information in a reverse manner, and turning to the step 1;
step 6, if the intention inherits the state, judging whether the input is the slot position information to be supplemented, if not, turning to step 7, if so, judging whether the slot position is complete, if so, confirming to a user, returning to success, storing the slot position information, executing step 8, and if not, executing step 5;
step 7, identifying the intention, judging whether the intention is a new intention, if so, storing the previously processed intention into a stack, executing the step 3, and returning to an invalid state unsuccessfully;
and 8, judging whether the stack has an incomplete intention, if so, the pop outputs an incomplete intention, converting the state into an intention inheritance state, and if not, finally executing the step 1.
The intention completing state in the step 1 is a state that the current instruction is processed and no unprocessed intention exists in the stack, or is an initial state of the system, and the intention inheriting state is a state that the slot information of the current intention is incomplete and the user needs to input the slot information. The intention refers to a certain service which needs to be consulted or completed in a user conversation, such as a ticketing robot for ticket buying service, and possible intentions of the user include ticket checking, ticket buying, order checking and the like; the slot position refers to parameter information required to be provided by a user in a conversation after the service is finished, for example, three items of information including a departure place, a destination and time are required to be provided for ticket checking, and three slot positions are required to be provided for the intention.
The step 7 of storing the processing intent into the stack means that currently processed intent classification information and slot information are stored and stored in a stack form.
In step 6, judging whether the input is slot position information to be supplemented, the slot position matching method is divided into three types, namely an enumeration type (such as place name and service name), a numerical type (such as time, pure number, multiplying power, distance, longitude and latitude) and a real type (such as name of a person), wherein:
the enumeration type slot position matching judges whether input is in a dictionary to judge whether slot position information to be supplemented is in a dictionary construction mode;
such as: when a ticketing robot providing a ticket-buying inquiry service recognizes an origin or a destination, a dictionary is as follows:
beijing: [ Beijing, capital ]
Shanghai: [ Shanghai, magic capital ]
Shenzhen (Shenzhen)
The numerical slot matching is performed through rules, if the regular expression is matching, whether the input meets the slot format requirement and the boundary requirement is judged;
such as: the number of the ticketing robots is matched, the regular expression is "[ 1-9] \ d ″, if the number is required, such as no more than 5, the regular expression is" [1-9] ", and the boundary condition is" < ═ 5 ".
The actual slot matching judges whether the extracted entity meets the parameter type required by the slot in a manner of entity extraction and type judgment;
such as: the passenger ticket inquiry service of the ticketing robot needs the parameters including the name of the passenger, and for the question of inquiring all tickets of three sheets which are not going out, the entity of three sheets is extracted through an entity extraction model, such as BilSTM-CRF, and the entity type of the name of the passenger conforms to the parameter requirements of the name of the passenger.
The judgment of whether the input is slot information to be supplemented in step 6 refers to a slot matching the current and all the incomplete intentions in the stack, and when the input can match a plurality of slots, all intentions in the stack are sorted in a mode of sorting learning based on LambdaMART, and the adopted characteristics are as follows: location information intended in the stack, degree of matching of the input to the slot name.
The position information refers to pop operation times executed by acquiring the intention from a stack, and the current intention is set to be 0;
the matching degree of the input (marked as Q) and the slot position name (marked as C) is obtained through neural network learning of BERT + MLP, and the model training input is formed by splicing Q and C and is marked as 1(Q and C matching) and 0(Q and C mismatching).
Step 4 and step 6 confirm to the user, mean that the complete slot position information obtained will be, produce the standard statement sentence pattern of the intention, ask users back, confirm the validity of the slot position, generally not obtain the complete parameter information from users' operation at one time and need confirm, supplement the information that may not be expressed of the user through the historical slot position, ask the slot position information obtained back may match the parameter of the wrong intention when there are multiple incomplete intentions, therefore need to confirm.
The present invention provides a question-answering multi-turn dialogue method supporting multi-intent switching for a question-answering system, and a method and a way for implementing the technical scheme are numerous, the above description is only a preferred embodiment of the present invention, and it should be noted that, for a person skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention. All the components not specified in the present embodiment can be realized by the prior art.
Claims (7)
1. A question-answering multi-turn dialogue method supporting multi-intention switching for a question-answering system is characterized by comprising the following steps:
step 1, when receiving question input, firstly reading the stored intention state information, judging the current intention state, if the intention state is an intention completion state, turning to step 2, and if the intention inheritance state is an intention inheritance state, turning to step 6;
step 2, performing intention identification on the question, and if the step 3 is successfully executed, returning to an invalid state unsuccessfully;
step 3, extracting the slot position of the question sentence, judging whether the slot position is complete or not, if so, returning to success, storing slot position information, executing step 8, and if not, executing step 4;
step 4, loading historical slot position information, judging whether the slot position is complete, if so, confirming to a user, if so, returning success, storing the slot position information, executing step 9, and if not, executing step 5;
step 5, storing the existing slot position information, converting the intention state into an intention inheritance state, inquiring the slot position information in a reverse manner, and turning to the step 1;
step 6, judging whether the input is slot position information to be supplemented, if not, turning to step 7, if so, judging whether the slot position is complete, if so, confirming to a user, if so, returning to success, storing the slot position information, executing step 8, and if not, returning to step 5;
step 7, identifying the intention, judging whether the intention is a new intention, if so, storing the previously processed intention into a stack, executing the step 3, and returning to an invalid state unsuccessfully;
and 8, judging whether the stack has an incomplete intention, if so, the pop outputs an incomplete intention, converting the state into an intention inheritance state, and if not, returning to execute the step 1.
2. The method of claim 1, wherein in step 1, the intention state information is an identifier of a saved intention state for determining a current intention state; the intention state comprises an intention completion state and an intention inheritance state, wherein the intention completion state refers to a state that the current instruction is processed and no unprocessed intention exists in the stack, or is an initial state of the question-answering system, and the intention inheritance state refers to a state that slot information of the current intention is incomplete and the slot information needs to be input by a user.
3. The method of claim 2, wherein the slot information in step 1 refers to a slot name and a corresponding value.
4. The method of claim 3, wherein in step 4, the confirmation to the user means that the complete slot position information to be obtained generates a standard statement sentence pattern of the current intention, asks the user back, and confirms the validity of the slot position.
5. The method of claim 4, wherein the determining whether the input is slot information to be supplemented in step 6 comprises: the slot matching method is divided into three types, namely enumeration type slot matching, numerical value type slot matching and entity type slot matching: wherein:
the enumeration type slot matching judges whether slot information to be supplemented is in a dictionary or not by judging whether input is in the dictionary or not in a dictionary mode;
the numerical slot matching is matched through a rule, and whether the input meets the format requirement and the boundary requirement of the slot is judged;
and the actual slot matching judges whether the extracted entity meets the parameter type required by the slot or not in the ways of entity extraction and type judgment.
6. The method as claimed in claim 5, wherein in step 6, the determining whether the input is slot information to be supplemented refers to a slot matching the current and all unfinished intentions in the stack, and when the input can match more than two slots, all intentions in the stack are sorted by a way of sorting learning based on LambdaMART, and the characteristics adopted are as follows: the matching degree of the position information, input and slot position name of the intention in the stack;
the position information refers to pop operation times executed by acquiring an intention from a stack, and the current intention is set to be 0;
the matching degree of the input Q and the slot position name is obtained through neural network learning of BERT and MLP, the input is recorded as Q, the slot position name is recorded as C model training input, the Q and C are formed by splicing, and the input is marked as 1 and 0.
7. The method of claim 6, wherein the step 7 of storing the previously processed intentions into the stack means that the currently processed intentions classification information and slot information are stored in the form of a stack.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110504405.9A CN113139044A (en) | 2021-05-10 | 2021-05-10 | Question-answering multi-turn dialogue method supporting multi-intention switching for question-answering system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110504405.9A CN113139044A (en) | 2021-05-10 | 2021-05-10 | Question-answering multi-turn dialogue method supporting multi-intention switching for question-answering system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113139044A true CN113139044A (en) | 2021-07-20 |
Family
ID=76818000
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110504405.9A Pending CN113139044A (en) | 2021-05-10 | 2021-05-10 | Question-answering multi-turn dialogue method supporting multi-intention switching for question-answering system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113139044A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114328882A (en) * | 2022-03-04 | 2022-04-12 | 中航信移动科技有限公司 | Information processing method, information processing device, computer equipment and storage medium |
| CN115617975A (en) * | 2022-12-20 | 2023-01-17 | 国家电网有限公司客户服务中心 | Intention identification method and device for few-sample and multi-turn conversations |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018149326A1 (en) * | 2017-02-16 | 2018-08-23 | 阿里巴巴集团控股有限公司 | Natural language question answering method and apparatus, and server |
| CN109446306A (en) * | 2018-10-16 | 2019-03-08 | 浪潮软件股份有限公司 | A kind of intelligent answer method of more wheels dialogue of task based access control driving |
| CN109446307A (en) * | 2018-10-16 | 2019-03-08 | 浪潮软件股份有限公司 | A kind of method for realizing dialogue management in Intelligent dialogue |
-
2021
- 2021-05-10 CN CN202110504405.9A patent/CN113139044A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018149326A1 (en) * | 2017-02-16 | 2018-08-23 | 阿里巴巴集团控股有限公司 | Natural language question answering method and apparatus, and server |
| CN109446306A (en) * | 2018-10-16 | 2019-03-08 | 浪潮软件股份有限公司 | A kind of intelligent answer method of more wheels dialogue of task based access control driving |
| CN109446307A (en) * | 2018-10-16 | 2019-03-08 | 浪潮软件股份有限公司 | A kind of method for realizing dialogue management in Intelligent dialogue |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114328882A (en) * | 2022-03-04 | 2022-04-12 | 中航信移动科技有限公司 | Information processing method, information processing device, computer equipment and storage medium |
| CN115617975A (en) * | 2022-12-20 | 2023-01-17 | 国家电网有限公司客户服务中心 | Intention identification method and device for few-sample and multi-turn conversations |
| CN115617975B (en) * | 2022-12-20 | 2023-04-18 | 国家电网有限公司客户服务中心 | Intention recognition method and device for few-sample multi-turn conversation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110704641B (en) | Ten-thousand-level intention classification method and device, storage medium and electronic equipment | |
| CN109918680B (en) | Entity identification method and device and computer equipment | |
| CN111212190B (en) | Conversation management method, device and system based on conversation strategy management | |
| CN108073600B (en) | Intelligent question-answer interaction method and device and electronic equipment | |
| CN108228764A (en) | A kind of single-wheel dialogue and the fusion method of more wheel dialogues | |
| CN111783439B (en) | Man-machine interaction dialogue processing method and device, computer equipment and storage medium | |
| CN110222162A (en) | A kind of intelligent answer method based on natural language processing and knowledge mapping | |
| CN111639168A (en) | Multi-turn conversation processing method and device, electronic equipment and storage medium | |
| CN113139044A (en) | Question-answering multi-turn dialogue method supporting multi-intention switching for question-answering system | |
| CN111540353B (en) | Semantic understanding method, device, equipment and storage medium | |
| CN109829036A (en) | A kind of dialogue management method and relevant apparatus | |
| CN109933654A (en) | A kind of dialogue management method based on State Tree | |
| CN111696558A (en) | Intelligent outbound method, device, computer equipment and storage medium | |
| CN112541068A (en) | Method, system, device and storage medium for recognizing intention of voice conversation | |
| CN111858888A (en) | Multi-round dialogue system of check-in scene | |
| CN113591463B (en) | Intention recognition method, device, electronic equipment and storage medium | |
| CN111738018A (en) | Intention understanding method, device, equipment and storage medium | |
| CN114003700A (en) | Method and system for processing session information, electronic device and storage medium | |
| CN112214589B (en) | Method for multi-round session framework based on cold start | |
| CN111858874A (en) | Conversation service processing method, device, equipment and computer readable storage medium | |
| CN111104504A (en) | Natural language processing and knowledge graph based dialogue method | |
| CN114860910A (en) | Intelligent dialogue method and system | |
| CN114154517A (en) | Deep learning-based dialogue quality assessment method and system | |
| CN110222161B (en) | Intelligent response method and device for conversation robot | |
| CN111091011B (en) | Domain prediction method, domain prediction device and electronic equipment |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210720 |