CN110211573A - A kind of task-driven type dialogue decision-making technique based on neural network model - Google Patents
A kind of task-driven type dialogue decision-making technique based on neural network model Download PDFInfo
- Publication number
- CN110211573A CN110211573A CN201910450074.8A CN201910450074A CN110211573A CN 110211573 A CN110211573 A CN 110211573A CN 201910450074 A CN201910450074 A CN 201910450074A CN 110211573 A CN110211573 A CN 110211573A
- Authority
- CN
- China
- Prior art keywords
- neural network
- model
- dialog
- task
- confidence
- 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 18
- 238000003062 neural network model Methods 0.000 title claims abstract description 15
- 239000013598 vector Substances 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 20
- 230000004044 response Effects 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims description 14
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000000306 recurrent effect Effects 0.000 claims description 8
- 239000004576 sand Substances 0.000 claims description 6
- 230000009471 action Effects 0.000 claims description 5
- 125000004122 cyclic group Chemical group 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 2
- 238000012546 transfer Methods 0.000 abstract description 2
- 230000003993 interaction Effects 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 3
- 241000282414 Homo sapiens Species 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 206010025482 malaise Diseases 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/16—Speech classification or search using artificial neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- Acoustics & Sound (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Machine Translation (AREA)
Abstract
The present invention provides a kind of, and the task-driven type based on neural network model talks with decision-making technique, pass through the confidence state tracker and other partial parameters in training pattern, the coded vector generated by intention assessment model is as the confidence probability distribution over states for being intended to distribution and the generation of confidence state tracker, database is transferred to be inquired using the corpus being collected, use database search result, it is intended to distribution and probability distribution transfers to tactful combination of network to form system acting, it passes to and generates the response of combination of network output system, complete dialogue function, possess the features such as robustness is preferable, it is poor to solve model tormulation ability existing for task-driven type conversational system instantly, training difficulty is big, the limitation of model learnability is more, model training data volume is huge, model training reward mechanism is imperfect, practicability is poor etc. in specific field Problem.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a task-driven type conversation decision-making method based on a neural network model.
Background
In recent years, artificial intelligence technology is gradually applied widely, human-computer interaction is applied to various fields such as smart home, smart medical treatment, public service, intelligent networked automobiles and the like, and human beings are pursuing more convenient life, so that robots or systems which can interpret human languages and autonomously complete responses are gradually an active development direction. The task-driven multi-turn dialogue is one of the main forms of the current development of human-computer interaction, and the research on the task-driven multi-turn dialogue system is widely developed at home and abroad at present, mainly aims at the human-computer interaction in the field of customer service, and is mainly applied to the service industry. However, in vehicles, human-computer interaction is not rapidly developed due to small space, high cost control requirements and the like, and emerging technologies are not applicable to services for drivers and passengers. Human-vehicle dialogue interaction involves a variety of proprietary knowledge in the field of driving, which makes current service-class dialogue systems unable to meet the needs of human-vehicle interaction.
In the prior art, a man-vehicle interaction system mainly adopts a keyword and key sentence matching mode to search the answer of a client problem in an existing database. The mechanized answering mode severely limits the expression mode of the driver and passengers, and causes the conversation interaction to be too programmed. In a single-round conversation system, when the user has situations of unclear semantic expression, language sickness, spoken language expression and the like, the time for obtaining correct response is greatly increased, and great influence is caused on the user experience.
For various problems existing in the prior art and the market, an application of a multi-turn dialogue interaction technology aiming at a specific field and a specific group is urgently needed, and in recent years, the remarkable progress of big data collection and deep learning in the natural language processing direction provides a new idea for realizing the design of a vehicle-mounted multi-turn dialogue system facing the requirements of drivers and passengers.
Disclosure of Invention
The invention aims to provide a task-driven conversation decision method based on a neural network model, so as to improve the experience of drivers and passengers on a vehicle-mounted system and the rapidness and convenience for obtaining answers to questions.
In order to achieve the above object, the present invention provides a task-driven dialog decision method based on a neural network model, comprising:
collecting a multi-round dialogue training text in a set field, constructing an intention recognition model based on an LSTM network, and training and cross-verifying the intention recognition model;
constructing a confidence state tracker for each information slot, and constructing an updating rule of a convolutional neural network and a cyclic neural network so as to train the confidence state tracker;
designing a database and a query mode of the database;
generating an output single vector according to the graph recognition model, the confidence state tracker and the output result of the database by three-phase matrix transformation, and adjusting the output single vector based on a language model;
training the wheel dialog data is used, and all possible parallel dialog cases are distributed according to the global confidence probability of the dialog state, so that the dialog action proposed at the next moment is generated.
Optionally, the step of constructing the intention recognition model based on the LSTM network includes:
coding each turn of dialog in a plurality of turns of dialogs in an intention recognition model to obtain a coding vector t of each turn of dialog;
the LSTM network is constructed according to the following formula:
wherein z istFor input encoded by a sequence as tIs represented by the distribution oftDenotes an input gate, ftIndicating forgetting to leave door otRepresents an output gate, ci-1,ciRepresenting a short-term memory state. Wxc,WhcIs a trainable parameter, hi-1Representing a hidden layer.
Optionally, the confidence state tracker comprises a feature extractor with a convolutional neural network and a Jordan type recurrent neural network.
Optionally, the step of building an update rule of the convolutional neural network and the cyclic neural network includes:
building the convolutional neural network, extracting word intermediate characteristics and sentence representation characteristics, and designing a characteristic vectorFor the concatenation of two convolutional neural network derived features, the input u of round t-1 is processed according to the following formulatProcessing the response s of the t-1 roundt-1：
Both sides of the sentence are filled with sentences before each convolution operation according to the following formula:
wherein the vector wsThe matrix WsBias term bsAnd b'sAnd a scalar gφ,sIs a parameter that is a function of,is the probability that the value is not mentioned up to the t round, the recurrent neural network weights of each value v are combined together, and each activation function is updatedTime varying feature
Optionally, the query content q of the databasetIs realized by the following formula:
where S' is the information entered, SIIs a set of information slots, and the information slots,is the output of the confidence state tracker.
Optionally, identifying the output z of the model from said graphtOutput of confidence state trackerAnd true value vector x obtained from output result of databasetGenerating an output single vector O by performing three-phase matrix transformation according to the following formulat：
Wherein, Wzo、Wpo、WxoIs a parameter that is a function of,is a concatenation of all the aggregated confidence vectors.
The task-driven dialog decision method based on the neural network model provided by the invention takes the coding vector generated by the intention recognition model as the intention distribution and the confidence state probability distribution generated by the confidence state tracker through the confidence state tracker and other partial parameters in the training model, sends the results to the database to be inquired by using the collected corpus, uses the database to search the results, the intention distribution and the probability distribution are combined by a strategy network to form system actions, the system actions are transmitted to a generation network combination output system to respond, a conversation function is completed, the characteristics of better robustness and the like are achieved, and the problems that the current task driving type conversation system is poor in model expression capability, large in training difficulty, more in model learnability limitation, huge in model training data volume, incomplete in model training reward mechanism, poor in practicability in a specific field and the like are solved.
Drawings
FIG. 1 is a flow chart of a task-driven conversational decision-making method based on a neural network model according to the present invention;
FIG. 2 is a general framework diagram of the vehicle-mounted multi-turn dialogue oriented by the present invention;
FIG. 3 is a Recurrent Neural Network (RNN) confidence tracking model of the bound Convolutional Neural Network (CNN) of the present invention.
Detailed Description
The following describes in more detail embodiments of the present invention with reference to the schematic drawings. Advantages and features of the present invention will become apparent from the following description and claims. It is to be noted that the drawings are in a very simplified form and are not to precise scale, which is merely for the purpose of facilitating and distinctly claiming the embodiments of the present invention.
As shown in fig. 1 to fig. 3, the present embodiment provides a task-driven dialog decision method based on a neural network model, which includes the following steps:
step 1: intent recognition, comprising the following 3 sub-steps:
(1) collecting dialogue training texts in the field of driver and passenger requirements for training a subsequent intention recognition model and a confidence state tracker;
(2) coding each dialog in multiple dialogs in the intention recognition model to obtain a coding vector t, z of the dialogtInput coded by sequence as tThe distribution of (2) is expressed, an LSTM network is constructed, and the output layer of the last step isThe method is used as probability representation of different intentions in the field of driver and passenger requirements, and the intention recognition model design is completed; wherein,
ztfor input encoded by a sequence as tIs represented by the distribution oftDenotes an input gate, ftIndicating forgetting to leave door otRepresents an output gate, ci-1,ciTo representShort term memory states. Wxc,WhcIs a trainable parameter, hi-1Representing a hidden layer.
(3) And for a given intention recognition model training data set, training and cross-verifying the intention recognition model, and finally generating an analysis result of the user intention.
Step 2: confidence state tracking, comprising the following 4 sub-steps:
(1) constructing a special confidence state tracker for each information slot, wherein each confidence state tracker is composed of a convolutional neural network feature extractor and a Jordan type cyclic neural network;
(2) modeling each turn of utterance context background, feature vectorProcessing t rounds of user input u for concatenation of derived features of two convolutional neural networkstProcessing the system response s of the t-1 roundt-1The calculation formula is as follows:
convolution neural network operation special for slot valueNot only sentence representation is extracted, but also the position of the lexical mark is extracted, and the embedding of the intermediate class n-gram model is determined. In each dialog, if multiple matches are observed, the corresponding embeddings are summed. On the other hand, if the particular slots or values do not match, null n-gram embedding is padded with zeros. To track the location of the de-lexical tokens, the two edges of the sentence are filled with sentences before each convolution operation, the number of vectors being determined by the filter size of each layer.
(3) Designing an updating rule of a recurrent neural network, and iteratively calculating the confidence state probability of the content expressed by each round of users;
wherein the vector wsThe matrix WsBias term bsAnd b'sAnd a scalar gφ,sIs a parameter that is a function of,is the probability that the value is not mentioned up to the t round, the recurrent neural network weights of each value v are combined together, and each activation function is updatedTime varying feature
The confidence state tracker maintains a polynomial distribution p for each information slot (a slot that can be used to constrain a search, such as parking) and a binary distribution for each request slot (a slot of interrogatable values, such as an address).
(4) Training confidence state tracker, and inquiring content q of database through output of confidence state trackertThe method is realized by the following algorithm:
s' is the information entered, SIIs a set of information slots, and the information slots,is the output of the confidence state tracker. Applying the content to a database, creating a binary truth vector x on the databasetWhere 1 indicates that the corresponding entity is consistent with the query, it can be concluded that it is consistent with the most likely confidence state. If x is not completely empty, the associated entity pointer keeps identifying a matching entity that was randomly selected, and the entity referenced by the entity pointer is used to form the final system response.
And step 3: dialog generation, comprising the following 4 sub-steps:
(1) z of output of intention recognition modeltOutput of confidence state trackerAnd true value vector x obtained from database search resulttInput and output of an output unit vector O representing the system operationtGenerating the appropriate sentence form, the individual probabilities of the classification values in the informative confidence states are irrelevant and are added together to form a summarized confidence vector for each information slot. Represented by three parts: the sum value probabilities, the probability that the user indicates that they "don't care" this slot and the probability that the slot is not mentioned. Finally, the output is generated by a three-way matrix transform:
wherein, Wzo、Wpo、WxoIs a parameter that is a function of,is the concatenation of all summary confidence vectors;
(2) generating sentence tokens similar to the templates based on the language model, and adjusting according to the output one-way quantity to generate system response; the token generation process is enhanced by a set of pointer networks to transfer entity specific information into a response by randomly sampling from a surface form list to replace the non-linearized tags, e.g., < s.place > to a place or region, replacing the de-lexical values with the actual attribute values of the entity currently selected by the database pointer.
(3) Training using multiple rounds of session data;
(4) and generating the dialog action proposed at the next moment according to all possible parallel dialog cases of the global confidence probability distribution of the dialog state.
The above description is only a preferred embodiment of the present invention, and does not limit the present invention in any way. It will be understood by those skilled in the art that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (6)
1. A task-driven conversation decision method based on a neural network model is characterized by comprising the following steps:
collecting a multi-round dialogue training text in a set field, constructing an intention recognition model based on an LSTM network, and training and cross-verifying the intention recognition model;
constructing a confidence state tracker for each information slot, and constructing an updating rule of a convolutional neural network and a cyclic neural network so as to train the confidence state tracker;
designing a database and a query mode of the database;
generating an output single vector according to the graph recognition model, the confidence state tracker and the output result of the database by three-phase matrix transformation, and adjusting the output single vector based on a language model;
training the wheel dialog data is used, and all possible parallel dialog cases are distributed according to the global confidence probability of the dialog state, so that the dialog action proposed at the next moment is generated.
2. The neural network model-based task-driven dialog decision method of claim 1, wherein the step of building an intent recognition model based on an LSTM network comprises:
coding each turn of dialog in a plurality of turns of dialogs in an intention recognition model to obtain a coding vector t of each turn of dialog;
the LSTM network is constructed according to the following formula:
wherein z istFor input encoded by a sequence as tIs represented by the distribution oftDenotes an input gate, ftIndicating forgetting to leave door otRepresents an output gate, ci-1,ciRepresenting a short-term memory state. Wxc,WhcIs a trainable parameter, hi-1Representing a hidden layer.
3. The neural network model-based task-driven dialog decision method of claim 2 in which the confidence state tracker consists of a recurrent neural network with convolutional neural network feature extractor and Jordan-type recurrent neural network.
4. The task-driven dialog decision method based on the neural network model as claimed in claim 3, wherein the step of building the update rules of the convolutional neural network and the cyclic neural network comprises:
building the convolutional neural network, extracting word intermediate characteristics and sentence representation characteristics, and designing a characteristic vectorFor the concatenation of two convolutional neural network derived features, the input u of round t-1 is processed according to the following formulatProcessing the response s of the t-1 roundt-1：
Both sides of the sentence are filled with sentences before each convolution operation according to the following formula:
wherein,vector wsThe matrix WsBias term bsAnd b'sAnd a scalar gφ,sIs a parameter that is a function of,is the probability that the value is not mentioned up to the t round, the recurrent neural network weights of each value v are combined together, and each activation function is updatedTime varying feature
5. The neural network model-based task-driven conversational decision making method of claim 4, wherein query content q of the databasetIs realized by the following formula:
where S' is the information entered, SIIs a set of information slots, and the information slots,is the output of the confidence state tracker.
6. The neural network model-based task-driven dialog decision method of claim 4, characterized in that the output z of the model is recognized from the graphtOutput of confidence state trackerAnd true value vector x obtained from output result of databasetGenerating a three-phase matrix transformation according to the following formulaOutput of a single vector Ot：
Wherein, Wzo、Wpo、WxoIs a parameter that is a function of,is a concatenation of all the aggregated confidence vectors.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910450074.8A CN110211573A (en) | 2019-05-28 | 2019-05-28 | A kind of task-driven type dialogue decision-making technique based on neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910450074.8A CN110211573A (en) | 2019-05-28 | 2019-05-28 | A kind of task-driven type dialogue decision-making technique based on neural network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110211573A true CN110211573A (en) | 2019-09-06 |
Family
ID=67788983
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910450074.8A Pending CN110211573A (en) | 2019-05-28 | 2019-05-28 | A kind of task-driven type dialogue decision-making technique based on neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110211573A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110599065A (en) * | 2019-09-23 | 2019-12-20 | 合肥工业大学 | Pointer neural network-based multi-satellite emergency task planning method and system |
| CN110619478A (en) * | 2019-09-23 | 2019-12-27 | 合肥工业大学 | Pointer neural network-based single-satellite emergency task planning method and system |
| CN112328776A (en) * | 2021-01-04 | 2021-02-05 | 北京百度网讯科技有限公司 | Dialog generation method and device, electronic equipment and storage medium |
| CN112579758A (en) * | 2020-12-25 | 2021-03-30 | 北京百度网讯科技有限公司 | Model training method, device, equipment, storage medium and program product |
| WO2021129262A1 (en) * | 2019-12-26 | 2021-07-01 | 思必驰科技股份有限公司 | Server-side processing method and server for actively initiating conversation, and voice interaction system capable of actively initiating conversation |
| CN113705652A (en) * | 2021-08-23 | 2021-11-26 | 西安交通大学 | Task type conversation state tracking system and method based on pointer generation network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170053646A1 (en) * | 2015-08-17 | 2017-02-23 | Mitsubishi Electric Research Laboratories, Inc. | Method for using a Multi-Scale Recurrent Neural Network with Pretraining for Spoken Language Understanding Tasks |
| CN108874782A (en) * | 2018-06-29 | 2018-11-23 | 北京寻领科技有限公司 | A kind of more wheel dialogue management methods of level attention LSTM and knowledge mapping |
| CN109933659A (en) * | 2019-03-22 | 2019-06-25 | 重庆邮电大学 | A kind of vehicle-mounted more wheel dialogue methods towards trip field |
-
2019
- 2019-05-28 CN CN201910450074.8A patent/CN110211573A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170053646A1 (en) * | 2015-08-17 | 2017-02-23 | Mitsubishi Electric Research Laboratories, Inc. | Method for using a Multi-Scale Recurrent Neural Network with Pretraining for Spoken Language Understanding Tasks |
| CN108874782A (en) * | 2018-06-29 | 2018-11-23 | 北京寻领科技有限公司 | A kind of more wheel dialogue management methods of level attention LSTM and knowledge mapping |
| CN109933659A (en) * | 2019-03-22 | 2019-06-25 | 重庆邮电大学 | A kind of vehicle-mounted more wheel dialogue methods towards trip field |
Non-Patent Citations (2)
| Title |
|---|
| TSUNG-HSIEN WEN ET AL.: "《A Network-based End-to-End Trainable Task-oriented Dialogue System》", 《PROCEEDINGS OF THE 15TH CONFERENCE OF THE EUROPEAN CHAPTER OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS》 * |
| 赵淑芳等: "《基于改进的LSTM深度神经网络语音识别研究》", 《郑州大学学报（工学版）》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110599065A (en) * | 2019-09-23 | 2019-12-20 | 合肥工业大学 | Pointer neural network-based multi-satellite emergency task planning method and system |
| CN110619478A (en) * | 2019-09-23 | 2019-12-27 | 合肥工业大学 | Pointer neural network-based single-satellite emergency task planning method and system |
| CN110619478B (en) * | 2019-09-23 | 2022-04-22 | 合肥工业大学 | Pointer neural network-based single-satellite emergency task planning method and system |
| CN110599065B (en) * | 2019-09-23 | 2022-04-22 | 合肥工业大学 | Pointer neural network-based multi-satellite emergency task planning method and system |
| WO2021129262A1 (en) * | 2019-12-26 | 2021-07-01 | 思必驰科技股份有限公司 | Server-side processing method and server for actively initiating conversation, and voice interaction system capable of actively initiating conversation |
| CN112579758A (en) * | 2020-12-25 | 2021-03-30 | 北京百度网讯科技有限公司 | Model training method, device, equipment, storage medium and program product |
| CN112579758B (en) * | 2020-12-25 | 2024-08-09 | 广东智城时代科技服务有限公司 | Model training method, device, apparatus, storage medium and program product |
| CN112328776A (en) * | 2021-01-04 | 2021-02-05 | 北京百度网讯科技有限公司 | Dialog generation method and device, electronic equipment and storage medium |
| US12086555B2 (en) | 2021-01-04 | 2024-09-10 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method for generating dialogue, electronic device, and storage medium |
| CN113705652A (en) * | 2021-08-23 | 2021-11-26 | 西安交通大学 | Task type conversation state tracking system and method based on pointer generation network |
| CN113705652B (en) * | 2021-08-23 | 2024-05-28 | 西安交通大学 | Task type dialogue state tracking system and method based on pointer generation network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110211573A (en) | A kind of task-driven type dialogue decision-making technique based on neural network model | |
| CN110781680B (en) | Semantic similarity matching method based on twin network and multi-head attention mechanism | |
| US20220309348A1 (en) | Method for generating personalized dialogue content | |
| CN111897941A (en) | Dialog generation method, network training method, device, storage medium and equipment | |
| CN110427461B (en) | Intelligent question and answer information processing method, electronic equipment and computer readable storage medium | |
| CN112417894B (en) | Conversation intention identification method and system based on multi-task learning | |
| Perez et al. | Dialog state tracking, a machine reading approach using memory network | |
| CN111274362B (en) | Dialogue generation method based on transformer architecture | |
| CN117521675A (en) | Information processing method, device, equipment and storage medium based on large language model | |
| CN114565104A (en) | Language model pre-training method, result recommendation method and related device | |
| CN114443827A (en) | Local information perception dialogue method and system based on pre-training language model | |
| CN114722839A (en) | Man-machine collaborative dialogue interaction system and method | |
| CN111984780A (en) | Multi-intention recognition model training method, multi-intention recognition method and related device | |
| CN112214591A (en) | Conversation prediction method and device | |
| CN109344242A (en) | A kind of dialogue answering method, device, equipment and storage medium | |
| CN117933271A (en) | Intelligent problem optimization dialogue method and system based on intention recognition of structural information | |
| CN113362852A (en) | User attribute identification method and device | |
| CN112328748A (en) | Method for identifying insurance configuration intention | |
| CN112559706B (en) | Training method of dialogue generating model, dialogue method, device and storage medium | |
| CN112364148B (en) | Deep learning method-based generative chat robot | |
| CN109308316B (en) | Adaptive dialog generation system based on topic clustering | |
| CN110297894B (en) | Intelligent dialogue generating method based on auxiliary network | |
| CN111914553A (en) | Financial information negative subject judgment method based on machine learning | |
| CN111522923A (en) | Multi-round task type conversation state tracking method | |
| CN113297374A (en) | Text classification method based on BERT and word feature fusion |
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: 20190906 |