CN111553157A - Entity replacement-based dialog intention identification method - Google Patents
Entity replacement-based dialog intention identification method Download PDFInfo
- Publication number
- CN111553157A CN111553157A CN202010271707.1A CN202010271707A CN111553157A CN 111553157 A CN111553157 A CN 111553157A CN 202010271707 A CN202010271707 A CN 202010271707A CN 111553157 A CN111553157 A CN 111553157A
- Authority
- CN
- China
- Prior art keywords
- entity
- text
- named entity
- dialog
- recognition
- 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 37
- 230000011218 segmentation Effects 0.000 claims abstract description 24
- 238000000605 extraction Methods 0.000 claims abstract description 10
- 238000001914 filtration Methods 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 25
- 238000002372 labelling Methods 0.000 claims description 14
- 230000002457 bidirectional effect Effects 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 9
- 238000006467 substitution reaction Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 2
- 238000013136 deep learning model Methods 0.000 claims 1
- 238000005215 recombination Methods 0.000 claims 1
- 230000006798 recombination Effects 0.000 claims 1
- 230000008569 process Effects 0.000 abstract description 5
- 238000013507 mapping Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 235000012054 meals Nutrition 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
Images
Classifications
-
- 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
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- 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
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- 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
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/242—Dictionaries
-
- 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
- G10L15/26—Speech to text systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a dialogue intention recognition method based on entity replacement, which comprises the following steps: step one, text word segmentation; step two, text filtering; step three, text named entity recognition: step four, text named entity replacement; step five, text feature extraction: step six, text intention recognition; the method replaces the entity name in the text information with the entity type by using the named entity recognition result, reduces the magnitude and the unbalance degree of the corpus data of the dialogue system, and comprehensively improves the accuracy of the intent recognition in the dialogue process.
Description
Technical Field
The invention relates to a dialog intention recognition method based on entity replacement, in particular to a dialog intention recognition method based on entity replacement.
Background
In recent years, under the influence of rapid development of artificial intelligence and semiconductor chip technology and increasing voice interaction requirements, various application products based on a dialog system, such as intelligent sound boxes, intelligent furniture, intelligent voice customer service and the like, are gradually in the market.
Such dialog systems generally consist of five modules, speech recognition (ASR), Natural Language Understanding (NLU), Dialog Management (DM), Natural Language Generation (NLG) and speech synthesis (TTS). At present, a voice recognition module has a better solution scheme by utilizing a deep learning technology, a natural language generation and voice synthesis module is relatively easy to control, and the difficulty of the design of a dialogue system mainly lies in a natural language understanding and dialogue management module. The natural language understanding module aims to convert text information obtained by the voice recognition module into semantic representation, so that the robot has the same language understanding capability as a human. Therefore, the accuracy of the language understanding module is a precondition and guarantee for maintaining the normal operation of the dialogue system.
With continuous optimization and upgrading of deep learning algorithms, machine computing power and big data technologies, the intention recognition accuracy of simple dialog systems such as voice meal ordering systems and voice song ordering systems has basically reached the commercialization level. However, since the magnitude of corpus data and the complexity of intent in a complex dialog system are significantly increased compared to the former, the difficulty of dialog intent recognition is aggravated by the problems of unbalanced corpus data, various intentions, and the like. For example, in the invention patent, "natural language intention understanding method and apparatus in human-computer interaction" (CN201710219326), a word vector of text information is used as an input, and an intention type of the text information is obtained by using an intention recognition model. Once the training sample class distribution is unbalanced, the intention recognition model is prone to serious over-fitting and under-fitting phenomena, and certain limitations exist. For example, in the invention patent "an intention recognition method and apparatus" (CN201811368503), text information is input into at least one intention recognition model, a prediction result corresponding to each intention recognition model is generated, and finally a text intention is determined. With the increase of the types of intentions, the cost and difficulty of model training of the method can be greatly increased, and the method is not suitable for intention recognition of a complex dialog system.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to make up the deficiency and the deficiency of the prior art means and provides a dialogue intention identification method based on entity replacement; the method replaces the entity name in the text information with the entity type by using the named entity recognition result, reduces the magnitude and the unbalance degree of the corpus data of the dialogue system, and therefore comprehensively improves the accuracy of intention recognition in the dialogue process.
The technical scheme is as follows: in order to achieve the purpose of the invention, the invention adopts the technical scheme that:
a dialogue intention recognition method based on entity replacement comprises the following steps:
step one, text word segmentation:
utilizing a word segmentation tool to segment words of the text information obtained by the voice recognition module to obtain a word segmentation result set Token; wherein the word segmentation result is represented as a set { W }, and W represents the segmented word;
step two, text filtering:
establishing a required stop word lexicon according to a conversation system, and filtering text information of the segmentation result set Token obtained in the step one by using the stop word lexicon to obtain a result Token after text cleaning*；
Step three, text named entity recognition:
and obtaining a named entity identification result as { E: t, wherein E represents an entity name and T represents an entity type;
step four, text named entity replacement:
and (3) carrying out one-to-one mapping on the named entity types involved in the dialog system by using specific characters, and recording the mapping as { T: c, recombining to obtain a new corpus, wherein T represents an entity type, and C represents a specific character; the specific character selected is guaranteed not to be present in the corpus of the dialog system;
step five, text feature extraction:
based on different types of pre-training models, fine tuning the pre-training models by using the new corpus obtained in the fourth step to obtain fine-tuned feature extraction models; obtaining a word vector Vec of the linguistic data of the dialogue system by using the fine-tuned feature extraction model;
step six, text intention recognition:
and realizing text intention recognition by adopting a network structure of bidirectional long-short term memory Bi-LSTM + Attention mechanism Attention.
Further, step three, the specific work step of text named entity recognition, which specifically works as follows:
1) based on the matching of the rules, the system,
designing a corresponding regular expression according to the requirement of the dialog system, extracting the named entity based on the regular expression, and matching fields meeting the requirement;
2) based on entity dictionary
Constructing a corresponding named entity dictionary according to the dialog system, and matching the word segmentation result obtained in the step one on the basis of the named entity dictionary;
3) based on a model
Acquiring an original corpus Sennce by collecting historical corpuses of a dialog system or a corpus generation mode, and manually or automatically labeling each position in the Sence to complete a sequence labeling task; obtaining the annotated Sentence Sennce after the annotation is finished*The named entity recognition model comprises B-T, I-T, O, E-T and S-T, and further realizes the named entity recognition based on the model by training the named entity recognition model.
Furthermore, in the model-based method in step three, the sequence labeling may adopt a BIO labeling mode or a biees labeling mode; in the BIOES labeling mode, B is Begin, which represents the beginning of an entity, I is Intermediate, which represents the middle of the entity, O is Other, which represents an unrelated character of a non-entity, E is End, which represents the End of the entity, and S is Single, which represents that the entity consists of Single characters.
Further, the step four, the concrete work step of text named entity replacement: and (3) identifying the named entity obtained in the third step by using a result { E: and (4) replacing the entity name T in the T by using a specific character C to obtain a result set { E: c, substituting the word segmentation result Token obtained in the step two*After the word W contained in the entity name E is replaced by the specific character C, recombining to obtain a new corpus Sennce';
further, the network structure in the sixth text intention recognition step mainly comprises 4 parts, specifically:
1) an input layer: taking the word vector Vec of the linguistic data of the dialog system obtained in the step five as an input V;
2) bidirectional LSTM layer: forward computing the word vector of the input layer by using a bidirectional long-short term memory network to obtain a vector VLAnd calculating backward to obtain vector VR(ii) a Splicing the front and rear vectors to obtain a spliced LSTM layer output vector VCIn which V isC＝[VL，VR]；
3) An Attention layer: output vector V to LSTM layerCPerforming Attention weighting to further obtain an output result VAThe calculation method is as follows:
Vm＝tanh(Vc)
α＝softmax(wTVm)
VA＝VcαT
where w is the weight matrix of the Attention layer.
4) An output layer: output result V of the Attention layerAPredicting the sentence meaning diagram by utilizing a Softmax classifier to obtain the meaning prediction result
Wherein WS，bSRespectively, the weight matrix and the offset value of the output layer.
Has the advantages that: compared with the prior art, the method has the advantages that the entity name in the text information is replaced by the entity type by utilizing the named entity recognition result, the magnitude and the unbalance degree of the corpus data of the dialogue system are reduced, and therefore the accuracy of intention recognition in the dialogue process is comprehensively improved.
Drawings
FIG. 1 is a schematic flow chart diagram of a dialog intention recognition method based on entity replacement according to the present invention;
FIG. 2 is an example of a text-naming entity replacement process of the present invention;
FIG. 3 is a corpus sequence tagging method according to the present invention;
FIG. 4 is a network architecture for implementing text intent recognition in accordance with the present invention.
Detailed Description
The present invention is further illustrated by the following specific examples, which are not intended to be limiting.
A dialog intention recognition method based on entity replacement, as shown in fig. 1, the method comprising the following steps:
the method comprises the following steps: text word segmentation
And performing word segmentation on the text information obtained by the voice recognition module by using a word segmentation tool to obtain a word segmentation result set Token, wherein the word segmentation result can be represented as a set { W }, and W represents a segmented word.
Step two: text filtering
The required word stock of stop words is established according to the dialogue system, and the stop words generally include but are not limited to auxiliary words, language words, conjunctions and the like. Filtering the text information of the word segmentation result set Token obtained in the step one by using a stop word lexicon to obtain a result Token after the text is cleaned*。
Step three: text named entity recognition
Named entity recognition includes, but is not limited to, the following three ways, and the ways can be mixed and used, and the result of named entity recognition is { E: t, wherein E represents an entity name and T represents an entity type.
1) Based on the matching of the rules, the system,
designing a corresponding regular expression according to the requirements of the conversation system, extracting named entities of types such as telephone numbers, mailbox addresses, identity card numbers and the like based on the regular expression, and matching fields meeting the requirements.
2) Based on entity dictionary
And constructing a corresponding named entity dictionary according to the dialog system, and matching the word segmentation result obtained in the step one based on the named entity dictionary, wherein the matching mode comprises but is not limited to character string multi-mode matching, word segmentation matching and the like.
3) Based on a model
The original corpus sequence is obtained by collecting the historical corpus of the dialog system or the corpus generation mode, and each position in the sequence is manually or automatically labeled to complete the sequence labeling task. In general, sequence annotation can adopt a BIO annotation mode or a BIOES annotation mode. Taking BIOES notation as an example, B is Begin, which represents the beginning of an entity, I is Intermediate, which represents the middle of an entity, O is Other, which represents an unrelated character Other than an entity, E is End, which represents the End of an entity, and S is Single, which represents that the entity consists of Single characters. Obtaining the annotated Sentence Sennce after the annotation is finished*The named entity recognition model comprises B-T, I-T, O, E-T and S-T, and further realizes the named entity recognition based on the model by training the named entity recognition model. Specifically, fig. 3 shows the result of labeling the corpus data sequence of a certain meal ordering system. The named entity recognition can generally adopt models such as HMM, CRF and the like, preferably, the named entity recognition can achieve better effect by adopting a two-way long-short term memory (BilSTM) + Conditional Random Field (CRF) model in the invention patent.
Step four: text named entity replacement
And (3) carrying out one-to-one mapping on the named entity types involved in the dialog system by using specific characters, and recording the mapping as { T: c, wherein T represents entity type and C represents specific character. The particular character selected is ensured to be absent from the corpus of the dialog system, including, but not limited to, english characters, roman numerals, greek letters, and the like.
And (3) identifying the named entity obtained in the third step by using a result { E: and (4) replacing the entity name T in the T by using a specific character C to obtain a result set { E: c, substituting the word segmentation result Token obtained in the step two*The word W contained in the entity name E is replaced with the specific character C and then replaced with the new wordAnd combining to obtain a new corpus Sennce'.
For example, the corpus includes 3 sentences S1, S2, S3, and the sentence is segmented by text information to obtain S1 ═ abc1d，S2＝abc2d，S3＝abc3d, wherein a, b, c1、c2、c3D represents different vocabulary in the corpus participle result Token, and c1、c2、c3Representing different entity names under the same named entity type. By special characters c0Replacement c1、c2、c3Then, the obtained 3 corpora after the named entity replacement are respectively S1 ', S2' and S3 ', wherein S1' is abcod，S2′＝abcod，S3′＝abcoAnd d, reducing the diversity of the linguistic data in the intention recognition model and reducing the unbalance degree of the text information. Specifically, fig. 2 shows an alternative example of the corpus data named entity of a weather query system.
Step five: text feature extraction
Based on pretrained models such as BERT, GPT, XLNET, XLM and the like, fine tuning is carried out on the pretrained models by utilizing the corpus Sennce' obtained in the fourth step, and fine-tuned feature extraction models are obtained. And obtaining a word vector Vec of the linguistic data of the dialogue system by using the fine-tuned feature extraction model.
Step six: text intent recognition
The text intention recognition method adopts a network structure of bidirectional long-short term memory (Bi-LSTM) + Attention mechanism (Attention) to realize text intention recognition. The network structure mainly comprises 4 parts, as shown in fig. 4, specifically:
1) an input layer: taking the word vector Vec of the linguistic data of the dialog system obtained in the step five as an input V;
2) bidirectional LSTM layer: forward computing the word vector of the input layer by using a bidirectional long-short term memory network to obtain a vector VLAnd calculating backward to obtain vector VR. Splicing the front and rear vectors to obtain a spliced LSTM layer output vector VCIn which V isC＝[VL，VR]；
3) An Attention layer: for transmission to LSTM layerOutput vector VCPerforming Attention weighting to further obtain an output result VAThe calculation method is as follows:
Vm＝tanh(Vc)
α＝softmax(wTVm)
VA＝VcαT
where w is the weight matrix of the Attention layer.
4) An output layer: output result V of the Attention layerAPredicting the sentence meaning diagram by utilizing a Softmax classifier to obtain the meaning prediction result
Wherein Ws，bsRespectively, the weight matrix and the offset value of the output layer.
The method replaces the entity name in the text information with the entity type by using the named entity recognition result, reduces the magnitude and the unbalance degree of the corpus data of the dialogue system, and therefore comprehensively improves the accuracy of intention recognition in the dialogue process.
Although the present invention has been described with reference to the preferred embodiments, it should be understood 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 (10)
1. A conversation intention recognition method based on entity replacement is characterized by comprising the following steps:
step one, text word segmentation:
utilizing a word segmentation tool to segment words of the text information obtained by the voice recognition module to obtain a word segmentation result set Token;
step two, text filtering:
establishing a required stop word lexicon according to a conversation system, and filtering text information of the segmentation result set Token obtained in the step one by using the stop word lexicon to obtain a result Token after text cleaning*；
Step three, text named entity recognition:
carrying out named entity recognition on the text cleaning result obtained in the step two through a deep learning model;
step four, text named entity replacement:
using specific characters to map named entity types related in a dialog system one by one, marking as { T: C }, recombining to obtain a new corpus, wherein T represents an entity type, and C represents specific characters; the specific character selected is guaranteed not to be present in the corpus of the dialog system;
step five, text feature extraction:
based on different types of pre-training models, fine tuning the pre-training models by using the new corpus obtained in the fourth step to obtain fine-tuned feature extraction models; obtaining a word vector Vec of the linguistic data of the dialogue system by using the fine-tuned feature extraction model;
step six, text intention recognition:
and realizing text intention recognition by adopting a network structure of bidirectional long-short term memory Bi-LSTM + Attention mechanism Attention.
2. The dialog intention recognition method based on entity substitution of claim 1, characterized in that: the word segmentation result of the first set of steps Token is represented as a set { W }, wherein W represents a segmented word;
3. the dialog intention recognition method based on entity substitution of claim 1, characterized in that: in the step 3, after entity identification is performed through naming, a named entity identification result is obtained as { E: t, wherein E represents an entity name and T represents an entity type.
4. The dialog intention recognition method based on entity substitution of claim 1, characterized in that: step three, identifying the text named entity, wherein the specific work comprises the following steps:
1) based on the matching of the rules, the system,
designing a corresponding regular expression according to the requirement of the dialog system, extracting the named entity based on the regular expression, and matching fields meeting the requirement;
2) based on entity dictionary
Constructing a corresponding named entity dictionary according to the dialog system, and matching the word segmentation result obtained in the step one on the basis of the named entity dictionary;
3) based on a model
Acquiring an original corpus Sennce by collecting historical corpuses of a dialog system or a corpus generation mode, and manually or automatically labeling each position in the Sence to complete a sequence labeling task; obtaining the annotated Sentence Sennce after the annotation is finished*And then, the named entity recognition based on the model is realized by training the named entity recognition model.
5. The dialog intention recognition method based on entity replacement of claim 4, characterized in that: the annotation statement Sennce*Is composed of B-T, I-T, O, E-T and S-T.
6. The dialog intention recognition method based on entity replacement of claim 4, characterized in that: based on the model, the sequence labeling can adopt a BIO labeling mode or a BIOES labeling mode.
7. The dialog intent recognition method based on entity replacement of claim 6, characterized in that: in the BIOES labeling mode, B is Begin, which represents the beginning of an entity, I is Intermediate, which represents the middle of an entity, O is Other, which represents an unrelated character Other than an entity, E is End, which represents the End of an entity, and S is Single, which represents that the entity consists of Single characters.
8. The dialog intention recognition method based on entity substitution of claim 1, characterized in that: step four, the concrete work steps of replacing the named entity of the Chinese text are as follows: replacing the entity name T in the named entity recognition result { E: T } obtained in the third step by a specific character C to obtain a result set { E: C } after the named entity is replaced, and substituting the result set into the word segmentation result Token obtained in the second step*After the word W included in the entity name E is replaced with the specific character C, the new corpus sequence' is obtained by recombination.
9. The dialog intention recognition method based on entity substitution of claim 1, characterized in that: in the sixth step, the network structure in the text intention recognition mainly comprises 4 parts, specifically:
1) an input layer: taking the word vector Vec of the linguistic data of the dialog system obtained in the step five as an input V;
2) bidirectional LSTM layer: forward computing the word vector of the input layer by using a bidirectional long-short term memory network to obtain a vector VLAnd calculating backward to obtain vector VR(ii) a Splicing the front and rear vectors to obtain a spliced LSTM layer output vector VCIn which V isC＝[VL，VR]；
3) An Attention layer: output vector V to LSTM layerCPerforming Attention weighting to further obtain an output result VAThe calculation method is as follows:
Vm＝tanh(Vc)
α＝softmax(wTVm)
VA＝VcαT
where w is the weight matrix of the Attention layer.
4) An output layer: output result V of the Attention layerAStatement semantics using Softmax classifierPredicting to obtain the intention prediction result
Wherein WS，bSRespectively, the weight matrix and the offset value of the output layer.
10. The dialog intention recognition method based on entity substitution of claim 1, characterized in that: and 3, the named entity recognition is realized by adopting a bidirectional long-short term memory (BilSTM + conditional random field CRF) model to obtain a better effect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010271707.1A CN111553157A (en) | 2020-04-08 | 2020-04-08 | Entity replacement-based dialog intention identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010271707.1A CN111553157A (en) | 2020-04-08 | 2020-04-08 | Entity replacement-based dialog intention identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111553157A true CN111553157A (en) | 2020-08-18 |
Family
ID=72002342
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010271707.1A Pending CN111553157A (en) | 2020-04-08 | 2020-04-08 | Entity replacement-based dialog intention identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111553157A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112905774A (en) * | 2021-02-22 | 2021-06-04 | 武汉市聚联科软件有限公司 | Human-computer conversation deep intention understanding method based on affair map |
| CN113779229A (en) * | 2021-08-31 | 2021-12-10 | 康键信息技术（深圳）有限公司 | User requirement identification matching method, device, equipment and readable storage medium |
| CN115064170A (en) * | 2022-08-17 | 2022-09-16 | 广州小鹏汽车科技有限公司 | Voice interaction method, server and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101598495A (en) * | 2009-06-29 | 2009-12-09 | 上海泽玛克敏达机械设备有限公司 | Condensate water recovery device |
| CN108874774A (en) * | 2018-06-05 | 2018-11-23 | 浪潮软件股份有限公司 | A kind of service calling method and system based on intention understanding |
| CN109359293A (en) * | 2018-09-13 | 2019-02-19 | 内蒙古大学 | Mongolian name entity recognition method neural network based and its identifying system |
| CN110287479A (en) * | 2019-05-20 | 2019-09-27 | 平安科技（深圳）有限公司 | Name entity recognition method, electronic device and storage medium |
| CN110298044A (en) * | 2019-07-09 | 2019-10-01 | 广东工业大学 | A kind of entity-relationship recognition method |
-
2020
- 2020-04-08 CN CN202010271707.1A patent/CN111553157A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101598495A (en) * | 2009-06-29 | 2009-12-09 | 上海泽玛克敏达机械设备有限公司 | Condensate water recovery device |
| CN108874774A (en) * | 2018-06-05 | 2018-11-23 | 浪潮软件股份有限公司 | A kind of service calling method and system based on intention understanding |
| CN109359293A (en) * | 2018-09-13 | 2019-02-19 | 内蒙古大学 | Mongolian name entity recognition method neural network based and its identifying system |
| CN110287479A (en) * | 2019-05-20 | 2019-09-27 | 平安科技（深圳）有限公司 | Name entity recognition method, electronic device and storage medium |
| CN110298044A (en) * | 2019-07-09 | 2019-10-01 | 广东工业大学 | A kind of entity-relationship recognition method |
Non-Patent Citations (2)
| Title |
|---|
| 徐啸;朱艳辉;冀相冰;: "基于自注意力深度学习的微博实体识别研究" * |
| 王子岳;邵曦;: "基于S-LSTM模型利用‘槽值门’机制的说话人意图识别" * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112905774A (en) * | 2021-02-22 | 2021-06-04 | 武汉市聚联科软件有限公司 | Human-computer conversation deep intention understanding method based on affair map |
| CN113779229A (en) * | 2021-08-31 | 2021-12-10 | 康键信息技术（深圳）有限公司 | User requirement identification matching method, device, equipment and readable storage medium |
| CN115064170A (en) * | 2022-08-17 | 2022-09-16 | 广州小鹏汽车科技有限公司 | Voice interaction method, server and storage medium |
| CN115064170B (en) * | 2022-08-17 | 2022-12-13 | 广州小鹏汽车科技有限公司 | Voice interaction method, server and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11488586B1 (en) | System for speech recognition text enhancement fusing multi-modal semantic invariance | |
| CN109887484B (en) | Dual learning-based voice recognition and voice synthesis method and device | |
| CN113205817A (en) | Speech semantic recognition method, system, device and medium | |
| CN110717341B (en) | Method and device for constructing old-Chinese bilingual corpus with Thai as pivot | |
| CN110619043A (en) | Automatic text abstract generation method based on dynamic word vector | |
| CN109086269B (en) | Semantic bilingual recognition method based on semantic resource word representation and collocation relationship | |
| CN112347796B (en) | Mongolian Chinese neural machine translation method based on combination of distillation BERT and improved Transformer | |
| CN111553157A (en) | Entity replacement-based dialog intention identification method | |
| CN110851601A (en) | Cross-domain emotion classification system and method based on layered attention mechanism | |
| CN113326367B (en) | Task type dialogue method and system based on end-to-end text generation | |
| CN114676255A (en) | Text processing method, device, equipment, storage medium and computer program product | |
| CN115510230A (en) | Mongolian emotion analysis method based on multi-dimensional feature fusion and comparative reinforcement learning mechanism | |
| CN115935959A (en) | Method for labeling low-resource glue word sequence | |
| Bai et al. | A universal bert-based front-end model for mandarin text-to-speech synthesis | |
| Sun | [Retracted] Analysis of Chinese Machine Translation Training Based on Deep Learning Technology | |
| Zeng | Implementation of Embedded Technology-Based English Speech Identification and Translation System. | |
| EP4281966A1 (en) | Generation of optimized spoken language understanding model through joint training with integrated acoustic knowledge-speech module | |
| CN114595700A (en) | Zero-pronoun and chapter information fused Hanyue neural machine translation method | |
| CN112257432A (en) | Self-adaptive intention identification method and device and electronic equipment | |
| CN117493548A (en) | Text classification method, training method and training device for model | |
| CN114238605B (en) | Automatic conversation method and device for intelligent voice customer service robot | |
| CN116029301A (en) | Cross-language text intention and slot joint recognition method and model | |
| WO2022159198A1 (en) | Generation of optimized knowledge-based language model through knowledge graph multi-alignment | |
| Qiang et al. | Back-translation-style data augmentation for mandarin chinese polyphone disambiguation | |
| WO2022159211A1 (en) | Generation of optimized spoken language understanding model through joint training with integrated knowledge-language module |
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: 20200818 |