CN109271632A - A kind of term vector learning method of supervision - Google Patents
A kind of term vector learning method of supervision Download PDFInfo
- Publication number
- CN109271632A CN109271632A CN201811075603.2A CN201811075603A CN109271632A CN 109271632 A CN109271632 A CN 109271632A CN 201811075603 A CN201811075603 A CN 201811075603A CN 109271632 A CN109271632 A CN 109271632A
- Authority
- CN
- China
- Prior art keywords
- term vector
- word
- vector
- model
- neural network
- 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.)
- Granted
Links
- 239000013598 vector Substances 0.000 title claims abstract description 135
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000003062 neural network model Methods 0.000 claims abstract description 26
- 238000013135 deep learning Methods 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims description 18
- 230000007246 mechanism Effects 0.000 claims description 11
- 238000003058 natural language processing Methods 0.000 abstract description 10
- 238000012549 training Methods 0.000 description 16
- 230000006870 function Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000005215 recombination Methods 0.000 description 2
- 230000006798 recombination Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 240000008790 Musa x paradisiaca Species 0.000 description 1
- 235000018290 Musa x paradisiaca Nutrition 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003780 insertion Methods 0.000 description 1
- 230000037431 insertion Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
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
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
The present patent application discloses a kind of term vector learning method of supervision, it is related to natural language processing method field, the following steps are included: step 1 builds deep learning network model by increasing word relationship disaggregated model on the basis of word2vec neural network model；Multiple a certain specified term vectors of adjacent input word vector sum are input in deep learning network model and carry out multi-task learning by step 2；Step 3 repeats step 4, is iterated calculating, word2vec neural network model and word relationship disaggregated model after being optimized.The present patent application can obtain the relationship between the term vector and specified term vector while term vector is calculated.
Description
Technical field
The present invention relates to natural language processing method fields, and in particular to a kind of term vector learning method of supervision.
Background technique
Term vector (word embedding), the vector characterization of word are an operations common in natural language processing, are
The common basic technology in the Internet services such as search engine, ad system, recommender system behind.
Term vector can simply be interpreted as word carrying out vectorization expression, and entity has been abstracted into mathematical description, such as one
A word: " apple ", be expressed as [0.4,0.5,0.9 ...], " banana ": [0.3,0.8,0.1 ...], the different dimensional of vector
Degree is used to characterize different characteristic, just represents different semantemes on different dimensions.
Natural language processing (NLP is made in natural language processing, abbreviation) is artificial intelligence and linguistics
The subdiscipline in field.Inquire into how to handle and with natural language, allow computer " to understand " language of the mankind, computer in this field
Data are converted into natural language, and natural language is converted into the form that computer program is more easily handled.
Present natural language processing, including various ways, wherein word2vec be it is now more common be used to carry out from
The series model of right Language Processing.Word2vec relies on skip-grams or continuous bag of words (CBOW) and is embedded in establish neural word,
Term vector is obtained using neural network model.It is more in line in daily communication compared to skip-grams, CBOW by nature language
The requirement that speech is interchangeable with machine language.
Although word2vec is able to carry out natural language processing, but word ambiguity and the obstructed feelings of sentence often occurs
Condition, to find out its cause, being because of the unsupervised mechanism of Word2vec, what word2vec considered is only the pass between word and surrounding word
System, when surrounding's word difference of two synonyms, the term vector that the two synonyms train also differs widely certainly.By big
The word2vec of corpus learns term vector out, has in term vector space from what given word was closer: synonym, same to position
Word, upper hyponym, related term etc., but word2vec can not distinguish these relationships.And many NLP tasks need this kind of word and word
Relationship, but the term vector obtained in existing learning method does not have such function.
Summary of the invention
The invention is intended to provide a kind of term vector learning method of supervision, can not only obtain the corresponding word of natural language to
Amount can also predict relationship between two term vectors.
The term vector learning method supervised in this programme, comprising the following steps:
Step 1 builds depth by increasing word relationship disaggregated model on the basis of word2vec neural network model
Learning network model；
Multiple a certain specified term vectors of adjacent input word vector sum are input in deep learning network model by step 2
Carry out multi-task learning；
Step 3 repeats step 4, is iterated calculating, and the word2vec neural network model and word after being optimized close
It is disaggregated model.
The present invention has the advantages that
The present invention proposes a kind of word-based and word relationship term vector generation method for having supervised.This method is existing
On the basis of word2vec, the word relationship disaggregated model for calculating word and word relationship is increased, using neural network multitask
The mechanism of habit learns term vector and word word relationship simultaneously.After the completion of training, term vector corresponding to word can not only obtain, and
It can predict the word relationship of two words.This word relationship is in multiple skills such as Text similarity computing, the information retrievals of natural language
There is very important effect in art field.
In addition, telling the priori knowledge of neural network word in the training process, the study for helping to eliminate low-frequency word is not filled
The case where dividing.
Further, before step 1, corpus text is segmented, establishes vocabulary and initial term vector corresponding with vocabulary.
By collecting corpus, vocabulary and initial term vector are established initially to be instructed to newly-built deep learning network model
Practice.
Further, it before step 1, according to vocabulary, marks in corpus text between each term vector and term vector
Relationship.
By be labelled with relationship term vector can output vector to deep learning network model and word relationship carry out it is anti-
To learning and give comments and criticisms, make the ginseng in the word2vec neural network model and word relationship disaggregated model in deep learning network model
Number can access optimization.
Wherein, corpus text is collected into internet and corpus ancient books and records using crawler.
Corpus text in corpus ancient books and records than more complete, but be not it is newest, crawled on internet by crawler
Supplement of the cyberspeak as corpus text in existing corpus ancient books and records, when the vocabulary established and initial term vector can be made all to meet
For language feature.
Further, in step 1, word relationship disaggregated model includes sequentially connected input layer, splicing layer, Quan Lian stratum
And probability layer；Wherein splicing layer by the output vector Wi being calculated by word2vec neural network model and is input to word
The specified vector Wk of relationship disaggregated model is spliced according to following formula: [Wi, Wk, Wi-Wk, Wi ° of Wk, Cos (Wi, Wk)].
By word relationship disaggregated model, the relationship between initial term vector is subjected to corresponding mark, is facilitated below in training
It is calculated together in calculating process with relationship.
Further, in step 2, input word vector sum is defined by initial term vector and specifies term vector.
All term vectors are both initialized to the vector of specified equal length.
Further, it in step 2, inputs and exports to the neural network model of word2vec using continuous pond bag model
The adjacent multiple term vectors of term vector are as input term vector.
Continuous pond bag model is the main models for being currently used to carry out natural language processing in word2vec, but every
Term vector in a pond bag does not carry out relationship each other and corresponds to, this term vector for allowing for finally being calculated also is difficult
Accurately correct relationship is established with other term vectors.The present invention efficiently solves this by increase word relationship disaggregated model and asks
Topic.And neural network model is superimposed continuous pond bag model, can greatly reduce the number of plies of calculating and the number of iteration, reduces and calculates
Amount enables natural language to be quickly processed into the term vector of standard, and then carries out subsequent applications.
Further, in step 2, when carrying out multi-task learning, word2vec neural network disk model is calculating output
While vector Wi, word relationship disaggregated model calculates the relationship label (Wk, Wi) of Wi and Wk.
While with initial vector training word2vec neural network, training word relationship disaggregated model, trained depth
Degree learning network model can obtain the relationship label (Wk, Wi) of Wi and Wk while obtaining output term vector Wi.
Further, in step 2, word2vec neural network is by error back propagation mechanism to neural network parameter
It optimizes, error includes the error in classification and word relationship error in classification of Hofman tree.
Optimize output vector Wi and the word2vec neural network model being calculated.
Further, in step 2, word relationship disaggregated model is by neural network error back propagation mechanism to full connection
Layer parameter optimizes.
Using the term vector of mark relationship, the relationship after calculating word relationship disaggregated model compares optimization, Jin Erxiu
It is positive to update full connection layer parameter, optimize the label (Wk, Wi) being calculated and word relationship disaggregated model.
Further, in step 3, the multiple input vectors selected at random and specified vector are separately input to
In word2vec neural network model and word relationship disaggregated model, an output term vector and the output term vector is calculated
With the relationship between specified term vector.
It is used after training word2vec neural network model and word relationship disaggregated model by successive ignition
When, it can also synchronize to obtain the relationship of the output term vector and specified term vector while obtaining output term vector.
Detailed description of the invention
Fig. 1 is the flow chart of the embodiment of the present invention.
Fig. 2 is the operation frame diagram of the embodiment of the present invention.
Specific embodiment
It is further described below by specific embodiment:
Embodiment is substantially as shown in Fig. 1: the term vector learning method supervised in the present embodiment, comprising the following steps:
The first step establishes corpus text library, to corpus text segment, segmenting method can using existing ltp, stammerer,
Even segment by hand；Participle establishes vocabulary later, and vocabulary is the set of word composition one by one；And randomly select initial word to
Amount.
When establishing corpus text library, pass through existing Chinese corpus ancient books and records such as " ccs " dictionary, " hownet ", " major term
Woods " and corpus text is collected by crawler on the internet, multiple big corpus texts are formed, and these big corpus texts are built
It stands into for receiving the corpus text library of rope.
After the foundation for completing corpus text library, the corpus text in library is segmented.Meanwhile by the initial of each word
Term vector is defined as W0={ w1 ..., wn }, wherein W0 is term vector, and w1 to wn is respectively term vector in n different dimensions
Characteristic value, wherein n is the term vector intrinsic dimensionality of word2vec setting.
Second step marks the relative of a word according to vocabulary, mark can according to existing dictionary, as major term woods,
Ccs etc. can also be marked by hand.
Relative includes synonym, apposition, hypernym, hyponym, unrelated word etc..In the connection established between word and word
When being, first, in accordance with the prior art, the relationship between word and word, corpus allusion quotation are established using the word relationship in existing corpus ancient books and records
Nationality such as " ccs " dictionary, " hownet ", " major term woods ".If only existing corpus ancient books and records, the word word relationship provided is not
Completely.In the present embodiment, we using following manner construct word word relationship, for Wi all word relationships with label (Wi,
Wk it) indicates, i and k belong to { 1 ..., n }.Word relationship has { synonym, apposition, hypernym, hyponym, unrelated word, unknown }.
Unknown word relational tags are set as " unknown ", these word relationships are simultaneously not involved in training.
Third step builds deep learning network structure.
Output term vector is calculated using the neural network model of word2vec, passes through CBOW model insertion in calculating process
Initial term vector.Meanwhile by word relationship disaggregated model calculate export term vector while, synchronize calculate the output word to
Relationship between amount and specified term vector.
Specifically, as shown in Fig. 2, being input to using the CBOW model of word2vec by term vector Wi-m to Wi+m is inputted
Output term vector Wi is calculated in word2vec neural network type.Then Wi is input to huffman model
The probability of output vector, the probability exported according to the prior art by huffman model are calculated in hierarchicalsoftmax
It corrects the neural network parameter and output vector of neural network model, makes the output term vector obtained after neural network model
It is more accurate.
While calculating output term vector, pass through word relationship disaggregated model, i.e. word relationship disaggregated model calculates output word
Relationship between vector Wi and specified term vector Wk.
Specifically, word relationship disaggregated model, including sequentially connected input layer, splicing layer, Quan Lian stratum fully
Connectedlayer and probability layer softmax.
4th step, multi-task learning.
While inputting multiple term vectors to word2vec, classified mould by the input layer of word relational model to word relationship
Type inputs specified term vector Wk.Then, while neural network model exports Wi, by Wi and Wk by the splicing layer of input, two
A vector forms row vector according to basic mathematical formulae recombination feature, and the row vector after recombination is Wi Wk Wi-Wk Wi °
WkCos (Wi, Wk), then remapped by the network of full articulamentum, finally by softmax classifier realize the classification of word relationship and
Error calculation obtains the relationship between two term vectors being arranged according to predetermined dimension.
Assuming that word2vec selects Cbow, the window selection of Cbow is 2m+1.[Wi-m ..., Wi+m] it is one in addition to Wi
The corpus data of the vectorization of window.Wk is the relative of Wi, i.e., specified term vector, the relationship of the two term vectors is expressed as
label(Wi,Wk).This variable represents the relationship of Wi and Wk.In the present embodiment, label (Wi, Wk) is equal to { synonym, same to position
It is word, hypernym, hyponym, unrelated word, unknown } in the similarity probabilities that calculate of each characteristic dimension, label (Wi, Wk) passes through
Word relationship disaggregated model is calculated.
This method uses the output and mark joint training of neural network model and sorter model, the damage of two models
The Probability Forms expression of appraxia logarithmetics is added to obtain the loss function of whole network again, as follows:
Loss=logP (Wi | Wi-m ..., Wi+m)+s*logP (label (Wi, Wk) | Wi, Wk)
Wherein, s is preset coefficient, such as takes s=0.5.
After obtaining loss function, joined using loss function using the mechanism learning network of neural network error back propagation
Number, wherein network parameter is that neural network is included, by the continuous corrective networks parameter of loss function, obtains neural network model
The output term vector arrived is more accurate.Meanwhile utilizing the Quan Lian stratum in the mechanism Study strategies and methods model of error back propagation
Parameter makes to connect full level parameter in the gradually calculating process of term vector relationship by continuous training optimization, makes the word finally obtained
Relationship disaggregated model can accurately calculate the relationship between two term vectors.
5th step updates and network parameter and connects level parameter entirely, and the deep neural network model after being optimized is to get arriving
Updated word2vec neural network model and word relationship disaggregated model.
When concrete application, the adjacent term vector of some term vector of stochastic inputs is obtained by neural network model
To output term vector Wi, while specified term vector Wk is inputted, obtains term vector Wi and word by the iterative calculation of sorter model
The relationship label (Wi, Wk) of the Wk of vector correlation between the two.
By above step, after the completion of training, the corresponding term vector of word can not only be obtained, while can be according to classification
Device model calculates the relationship between the term vector and specified term vector.
This method increases the classifier of word and word relationship on the basis of existing word2vec, more using neural network
The mechanism of tasking learning learns term vector and word word relationship simultaneously, during CBOW term vector model is learnt, passes through
The relationship of each term vector and other term vectors is predicted and is defined by word relationship disaggregated model.As shown in Fig. 2, this method has
The output and mark joint training of two networks are used to body, left network is the vec2vec CBOW net based on Hofman tree
Network, the right are word relationship sorter networks.The Probability Forms of the loss logarithmetics of two networks of left and right are indicated to be added again, as net
The loss function of network.After the completion of training, term vector corresponding to word can not only obtain, and can predict that the word of two words closes
System.This word relationship has very important in multiple technical fields such as Text similarity computing, the information retrievals of natural language
Effect.
In addition, telling the priori knowledge of neural network word in the training process, the study for helping to eliminate low-frequency word is not filled
The case where dividing.Such as: " Zhang San " and " Li Si " is synonym, and there are many frequency that " Li Si " occurs in training text, it is believed that can be with
It is trained up, and the frequency that " Zhang San " occurs is seldom, can not all train up according to traditional word2vec.In the present invention
Network in, when training Zhang San, the term vector of word-based word sorter network and " Li Si " can be with by mistake backpropagation mechanism
" Zhang San " term vector is updated, so inventive network helps to eliminate the insufficient situation of study of low-frequency word.
Similarly, the priori knowledge of neural network word is told in the training process, and word-based word sorter network enhances input
Two term vectors for having a priori interest distinguish and connection, overcome in original word2vec network model term vector only and dependence
Deficiency in the relevant mechanism of text.
What has been described above is only an embodiment of the present invention, and the common sense such as well known specific structure and characteristic are not made herein in scheme
Excessive description, technical field that the present invention belongs to is all before one skilled in the art know the applying date or priority date
Ordinary technical knowledge can know the prior art all in the field, and have using routine experiment hand before the date
The ability of section, one skilled in the art can improve and be implemented in conjunction with self-ability under the enlightenment that the application provides
This programme, some typical known features or known method should not become one skilled in the art and implement the application
Obstacle.It should be pointed out that for those skilled in the art, without departing from the structure of the invention, can also make
Several modifications and improvements out, these also should be considered as protection scope of the present invention, these all will not influence the effect that the present invention is implemented
Fruit and patent practicability.The scope of protection required by this application should be based on the content of the claims, the tool in specification
The records such as body embodiment can be used for explaining the content of claim.
Claims (10)
1. a kind of term vector learning method of supervision, it is characterised in that: the following steps are included:
Step 1 builds deep learning by increasing word relationship disaggregated model on the basis of word2vec neural network model
Network model；
Multiple a certain specified term vectors of adjacent input word vector sum are input in deep learning network model and carry out by step 2
Multi-task learning；
Step 3 repeats step 4, is iterated calculating, word2vec neural network model and word relation after being optimized
Class model.
2. the term vector learning method of supervision according to claim 1, it is characterised in that: before step 1, by corpus
Text participle, establishes vocabulary and initial term vector corresponding with vocabulary.
3. the term vector learning method of supervision according to claim 2, it is characterised in that: according to vocabulary, mark corpus text
Relationship in this between each term vector and term vector.
4. the term vector learning method of supervision according to claim 1, it is characterised in that: in step 1, word relation
Class model includes sequentially connected input layer, splicing layer, Quan Lian stratum and probability layer；Wherein splicing layer will pass through word2vec
The output vector Wi and be input to the specified vector Wk of word relationship disaggregated model according to following public affairs that neural network model is calculated
Formula is spliced: [Wi, Wk, Wi-Wk, Wi ° of Wk, Cos (Wi, Wk)].
5. the term vector learning method of supervision according to claim 2, it is characterised in that: in step 2, by initial
Term vector specifies term vector to define input word vector sum.
6. the term vector learning method of supervision according to claim 5, it is characterised in that: in step 2, using continuous
Pond bag model to the input of the neural network model of word2vec with export the adjacent multiple term vectors of term vector as input word to
Amount.
7. the term vector learning method of supervision according to claim 4, it is characterised in that: more in progress in step 2
When tasking learning, while calculating output vector Wi, word relationship disaggregated model calculates word2vec neural network disk model
The relationship label (Wk, Wi) of Wi and Wk.
8. the term vector learning method of supervision according to claim 1, it is characterised in that: in step 2, word2vec
Neural network optimizes neural network parameter by error back propagation mechanism, and error includes the error in classification of Hofman tree
With word relationship error in classification.
9. the term vector learning method of supervision according to claim 1, it is characterised in that: in step 2, word relation
Class model optimizes full connection layer parameter by neural network error back propagation mechanism.
10. the term vector learning method of supervision according to claim 1, it is characterised in that: in step 3, will select at random
Multiple input vectors and specified vector out are separately input in word2vec neural network model and word relationship disaggregated model, meter
It calculates and obtains the relationship between an output term vector and the output term vector and specified term vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811075603.2A CN109271632B (en) | 2018-09-14 | 2018-09-14 | Supervised word vector learning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811075603.2A CN109271632B (en) | 2018-09-14 | 2018-09-14 | Supervised word vector learning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109271632A true CN109271632A (en) | 2019-01-25 |
| CN109271632B CN109271632B (en) | 2023-05-26 |
Family
ID=65188340
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811075603.2A Active CN109271632B (en) | 2018-09-14 | 2018-09-14 | Supervised word vector learning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109271632B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110825875A (en) * | 2019-11-01 | 2020-02-21 | 科大讯飞股份有限公司 | Text entity type identification method and device, electronic equipment and storage medium |
| CN110852077A (en) * | 2019-11-13 | 2020-02-28 | 泰康保险集团股份有限公司 | Method, device, medium and electronic equipment for dynamically adjusting Word2Vec model dictionary |
| CN111444346A (en) * | 2020-03-31 | 2020-07-24 | 广州大学 | Word vector confrontation sample generation method and device for text classification |
| CN112989032A (en) * | 2019-12-17 | 2021-06-18 | 医渡云（北京）技术有限公司 | Entity relationship classification method, apparatus, medium and electronic device |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105975478A (en) * | 2016-04-09 | 2016-09-28 | 北京交通大学 | Word vector analysis-based online article belonging event detection method and device |
| CN106557462A (en) * | 2016-11-02 | 2017-04-05 | 数库（上海）科技有限公司 | Name entity recognition method and system |
| CN106682220A (en) * | 2017-01-04 | 2017-05-17 | 华南理工大学 | Online traditional Chinese medicine text named entity identifying method based on deep learning |
| US20170139899A1 (en) * | 2015-11-18 | 2017-05-18 | Le Holdings (Beijing) Co., Ltd. | Keyword extraction method and electronic device |
| CN106933806A (en) * | 2017-03-15 | 2017-07-07 | 北京大数医达科技有限公司 | The determination method and apparatus of medical synonym |
| CN107145503A (en) * | 2017-03-20 | 2017-09-08 | 中国农业大学 | Remote supervision non-categorical relation extracting method and system based on word2vec |
| CN107247780A (en) * | 2017-06-12 | 2017-10-13 | 北京理工大学 | A kind of patent document method for measuring similarity of knowledge based body |
| CN107291693A (en) * | 2017-06-15 | 2017-10-24 | 广州赫炎大数据科技有限公司 | A kind of semantic computation method for improving term vector model |
| CN107895000A (en) * | 2017-10-30 | 2018-04-10 | 昆明理工大学 | A kind of cross-cutting semantic information retrieval method based on convolutional neural networks |
| CN107895051A (en) * | 2017-12-08 | 2018-04-10 | 宏谷信息科技（珠海）有限公司 | A kind of stock news quantization method and system based on artificial intelligence |
| CN108280058A (en) * | 2018-01-02 | 2018-07-13 | 中国科学院自动化研究所 | Relation extraction method and apparatus based on intensified learning |
| US10037362B1 (en) * | 2017-07-24 | 2018-07-31 | International Business Machines Corpoation | Mining procedure dialogs from source content |
| CN108388654A (en) * | 2018-03-01 | 2018-08-10 | 合肥工业大学 | A kind of sensibility classification method based on turnover sentence semantic chunk partition mechanism |
| CN108388914A (en) * | 2018-02-26 | 2018-08-10 | 中译语通科技股份有限公司 | A kind of grader construction method, grader based on semantic computation |
-
2018
- 2018-09-14 CN CN201811075603.2A patent/CN109271632B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170139899A1 (en) * | 2015-11-18 | 2017-05-18 | Le Holdings (Beijing) Co., Ltd. | Keyword extraction method and electronic device |
| CN105975478A (en) * | 2016-04-09 | 2016-09-28 | 北京交通大学 | Word vector analysis-based online article belonging event detection method and device |
| CN106557462A (en) * | 2016-11-02 | 2017-04-05 | 数库（上海）科技有限公司 | Name entity recognition method and system |
| CN106682220A (en) * | 2017-01-04 | 2017-05-17 | 华南理工大学 | Online traditional Chinese medicine text named entity identifying method based on deep learning |
| CN106933806A (en) * | 2017-03-15 | 2017-07-07 | 北京大数医达科技有限公司 | The determination method and apparatus of medical synonym |
| CN107145503A (en) * | 2017-03-20 | 2017-09-08 | 中国农业大学 | Remote supervision non-categorical relation extracting method and system based on word2vec |
| CN107247780A (en) * | 2017-06-12 | 2017-10-13 | 北京理工大学 | A kind of patent document method for measuring similarity of knowledge based body |
| CN107291693A (en) * | 2017-06-15 | 2017-10-24 | 广州赫炎大数据科技有限公司 | A kind of semantic computation method for improving term vector model |
| US10037362B1 (en) * | 2017-07-24 | 2018-07-31 | International Business Machines Corpoation | Mining procedure dialogs from source content |
| CN107895000A (en) * | 2017-10-30 | 2018-04-10 | 昆明理工大学 | A kind of cross-cutting semantic information retrieval method based on convolutional neural networks |
| CN107895051A (en) * | 2017-12-08 | 2018-04-10 | 宏谷信息科技（珠海）有限公司 | A kind of stock news quantization method and system based on artificial intelligence |
| CN108280058A (en) * | 2018-01-02 | 2018-07-13 | 中国科学院自动化研究所 | Relation extraction method and apparatus based on intensified learning |
| CN108388914A (en) * | 2018-02-26 | 2018-08-10 | 中译语通科技股份有限公司 | A kind of grader construction method, grader based on semantic computation |
| CN108388654A (en) * | 2018-03-01 | 2018-08-10 | 合肥工业大学 | A kind of sensibility classification method based on turnover sentence semantic chunk partition mechanism |
Non-Patent Citations (1)
| Title |
|---|
| ALEXIS CONNEAU ET AL: ""Supervised Learning of Universal Sentence Representations from"", 《COMPUTER SCIENCE>COMPUTATION AND LANGUAGE》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110825875A (en) * | 2019-11-01 | 2020-02-21 | 科大讯飞股份有限公司 | Text entity type identification method and device, electronic equipment and storage medium |
| CN110825875B (en) * | 2019-11-01 | 2022-12-06 | 科大讯飞股份有限公司 | Text entity type identification method and device, electronic equipment and storage medium |
| CN110852077A (en) * | 2019-11-13 | 2020-02-28 | 泰康保险集团股份有限公司 | Method, device, medium and electronic equipment for dynamically adjusting Word2Vec model dictionary |
| CN110852077B (en) * | 2019-11-13 | 2023-03-31 | 泰康保险集团股份有限公司 | Method, device, medium and electronic equipment for dynamically adjusting Word2Vec model dictionary |
| CN112989032A (en) * | 2019-12-17 | 2021-06-18 | 医渡云（北京）技术有限公司 | Entity relationship classification method, apparatus, medium and electronic device |
| CN111444346A (en) * | 2020-03-31 | 2020-07-24 | 广州大学 | Word vector confrontation sample generation method and device for text classification |
| CN111444346B (en) * | 2020-03-31 | 2023-04-18 | 广州大学 | Word vector confrontation sample generation method and device for text classification |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109271632B (en) | 2023-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110298037B (en) | Convolutional neural network matching text recognition method based on enhanced attention mechanism | |
| CN109992783B (en) | Chinese word vector modeling method | |
| CN112163426B (en) | Relationship extraction method based on combination of attention mechanism and graph long-time memory neural network | |
| CN111241294B (en) | Relationship extraction method of graph convolution network based on dependency analysis and keywords | |
| CN110377903B (en) | Sentence-level entity and relation combined extraction method | |
| CN109271632A (en) | A kind of term vector learning method of supervision | |
| CN106326212B (en) | A kind of implicit chapter relationship analysis method based on level deep semantic | |
| CN109657239A (en) | The Chinese name entity recognition method learnt based on attention mechanism and language model | |
| CN110222163A (en) | A kind of intelligent answer method and system merging CNN and two-way LSTM | |
| CN109344399A (en) | A kind of Text similarity computing method based on the two-way lstm neural network of stacking | |
| CN110555084B (en) | Remote supervision relation classification method based on PCNN and multi-layer attention | |
| CN107273426B (en) | A kind of short text clustering method based on deep semantic route searching | |
| CN111177383B (en) | Text entity relation automatic classification method integrating text grammar structure and semantic information | |
| CN111522965A (en) | Question-answering method and system for entity relationship extraction based on transfer learning | |
| CN104615767A (en) | Searching-ranking model training method and device and search processing method | |
| CN109933792B (en) | Viewpoint type problem reading and understanding method based on multilayer bidirectional LSTM and verification model | |
| CN108647191A (en) | It is a kind of based on have supervision emotion text and term vector sentiment dictionary construction method | |
| CN110866542A (en) | Depth representation learning method based on feature controllable fusion | |
| CN111241303A (en) | Remote supervision relation extraction method for large-scale unstructured text data | |
| Chen et al. | Deep neural networks for multi-class sentiment classification | |
| CN114417851A (en) | Emotion analysis method based on keyword weighted information | |
| CN111428481A (en) | Entity relation extraction method based on deep learning | |
| CN103678318A (en) | Multi-word unit extraction method and equipment and artificial neural network training method and equipment | |
| CN110569355B (en) | Viewpoint target extraction and target emotion classification combined method and system based on word blocks | |
| CN114048314A (en) | Natural language steganalysis method |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Qin Hong Hui Inventor after: Du Ruo Inventor after: Xiang Hai Inventor after: Hou Cong Inventor after: Liu Ke Inventor before: Qin Hong Hui |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231127 Address after: Room 1208-208, 12th Floor, Building 2, Fuhai Center, Daliushu, Haidian District, Beijing, 100081 Patentee after: Beijing Zhicheng Excellence Technology Co.,Ltd. Address before: 401120 No. 1, Floor 3, Building 11, Internet Industrial Park, No. 106, West Section of Jinkai Avenue, Yubei District, Chongqing Patentee before: CHONGQING XIEZHI TECHNOLOGY Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231220 Address after: No. 1-0286, Juhe 6th Street, Jufuyuan Industrial Park, Tongzhou Economic Development Zone, Beijing, 101127 Patentee after: Beijing Star Cube Digital Technology Co.,Ltd. Address before: Room 1208-208, 12th Floor, Building 2, Fuhai Center, Daliushu, Haidian District, Beijing, 100081 Patentee before: Beijing Zhicheng Excellence Technology Co.,Ltd. |
|
| TR01 | Transfer of patent right |