CN109388706A - A kind of problem fine grit classification method, system and device - Google Patents
A kind of problem fine grit classification method, system and device Download PDFInfo
- Publication number
- CN109388706A CN109388706A CN201710678652.4A CN201710678652A CN109388706A CN 109388706 A CN109388706 A CN 109388706A CN 201710678652 A CN201710678652 A CN 201710678652A CN 109388706 A CN109388706 A CN 109388706A
- Authority
- CN
- China
- Prior art keywords
- vector
- text
- fine grit
- word
- question text
- 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
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Abstract
The present invention provides a kind of problem fine grit classification method, system and device, comprising the following steps: semantic primitive extraction step extracts semantic primitive in former question text;Semantic primitive spread step is extended semantic primitive using vector space model, to obtain scaling problem text;Chinese word coding step, using two-way length, memory network carries out Chinese word coding to scaling problem text in short-term, to obtain encoded question text;Word focus steps carry out word focusing using attention mechanism to encoded question text, to obtain question text vector.Fine grit classification step carries out fine grit classification to question text vector using softmax classifier.The present invention realizes the automation of this process to the greatest extent under the premise of guaranteeing certain accuracy rate, improves the efficiency of problem fine grit classification, can save human cost to greatest extent.
Description
Technical field
The present invention relates to text fine grit classification field, especially a kind of problem fine grit classification method, system and device.
Background technique
With the development of social networks, the social tools such as community's question and answer, microblogging, wechat become more and more popular.Social question and answer net
It stands, such as Quora, Research Gate, Yahoo!Answers, know, bean cotyledon etc. cause the pass of many domestic and foreign scholars
Note.Question Classification is the core component of question and answer website, directly affects the retrieval rate of user.And in recent years, with depth
The development of study is spent, the fine grit classification of problem is increasingly by the favor of researcher.Problem fine grit classification is directed to
The classification of problem more refined is current research hotspot.
Generally, problem refers to sentence word number text the problem of 140-150 or so is needed by answer or explanation.It asks
Topic fine grit classification is under the jurisdiction of short text classification, and main task is that accurately explanation distinguishes the son for belonging to the same superclass
Classification.With short text classification compared with, problem fine grit classification is faced with following problem, mainly have: (1) segment classification it
Between global feature it is more similar, only had differences in some regional area;(2) it is difficult to find out the feature locally to differ greatly
Region;(3) question text itself is more short and small, and feature space is more sparse.Therefore traditional text representation model is directly applied
It is ineffective on to problem fine grit classification.
Summary of the invention
For the above existing issue, it is an object of the invention to overcome short text feature space itself more it is sparse not
Foot, proposes a kind of problem fine grit classification method.
According to the first aspect of the invention, a kind of problem fine grit classification method is provided, for former question text into
Row Question Classification, comprising the following steps: semantic primitive extraction step extracts semantic primitive in former question text;Semantic primitive
Spread step is extended semantic primitive using vector space model, to obtain scaling problem text;Chinese word coding step, makes
With two-way length, memory network carries out Chinese word coding to scaling problem text in short-term, to obtain encoded question text;Word focus steps are right
Encoded question text carries out word focusing using attention mechanism, to obtain question text vector.Fine grit classification step uses
Softmax classifier carries out fine grit classification to question text vector.
Preferably, semantic primitive extraction step includes: by finding out institute in former question text to interdependent syntax traversal of tree
Some noun phrase nodes and verb phrase node are as semantic primitive.
Preferably, semantic primitive spread step includes that semantic primitive is converted semanteme by generative semantics unit vector step
Unit vector, semantic primitive vector VuCoincidence formula:
Vu=Vu1+Vu2+…+Vum={ z1,z2,…,zd}
Wherein, Vu1, Vu2, VumUnit vector corresponding to multiple phrase nodes in semantic primitive is respectively indicated, m is derived from so
Number;Z1, Z2, ZdVector set is constituted, d takes natural number;Calculate cosine similarity step, calculate semantic primitive vector with
The cosine similarity of the vector of all words in Word2Vec model;Scaling problem text steps are obtained, select cosine similarity most
Big preceding several corresponding words or phrase of being worth are as scaling problem text.
Preferably, in Chinese word coding step, two-way length memory network coincidence formula in short-term:
Wherein xiIndicate term vector,To unit LSTM before indicating, according to from the 1st word to the direction of the T word
Word is read,It indicates reversed unit LSTM, reads word according to from the T word to the direction of the 1st word,It indicates just
Output to i-th layer of hidden layer,Indicate the output of reversed i-th layer of hidden layer.
Preferably, word focus steps obtain question text vector, formula using formula are as follows:
ui=tanh (Ewhi+bw)
Wherein αiIndicate that normalized weighted value, s indicate finally obtained question text vector, EwAnd μwIndicate random first
The weight parameter value of beginningization, bwFor amount of bias, μiFor the output of Single layered perception neural networks, hiFor the output of hidden layer.
Preferably, in fine grit classification step, softmax classifier includes formula:
P=softmax (Wcs+bc)
Wherein j indicates the label of question text vector s, WcIndicate the weight parameter value of random initializtion, bcFor amount of bias, p
Value is the probability score value of each classification, and L indicates loss function value.
According to the second aspect of the invention, a kind of problem fine grit classification system is provided, for former question text into
Row Question Classification, comprising: semantic primitive extraction module, for extracting semantic primitive in former question text;Semantic primitive extension
Module, for being extended to semantic primitive using vector space model, to obtain scaling problem text;Chinese word coding device module,
For using two-way length, memory network carries out Chinese word coding to scaling problem text in short-term, to obtain encoded question text;Word focuses
Module, for carrying out word focusing using attention mechanism to encoded question text, to obtain question text vector;Softmax points
Class device module carries out fine grit classification to question text vector using softmax classifier.
According to the third aspect of the present invention, a kind of problem fine grit classification device is provided, for former question text into
Row Question Classification, comprising: storage unit, for storing program, program unit processed realizes above-mentioned first aspect when executing
The problem of fine grit classification method the step of;Execution unit, for executing program in storage unit.
With it is traditional based on statistical learning the problem of classification method compared with, the present invention extracts semantic single in former question text
Member is extended semantic primitive using vector space model, carries out Chinese word coding to scaling problem text and word focuses, utilize
The problem of softmax classifier indicates obtained vectorization text carries out fine grit classification, improves problem fine grit classification
Accuracy rate.Compared with existing the problem of being based on craft or being assisted by machine fine grit classification, the present invention is guaranteeing centainly
The automation for realizing this process under the premise of accuracy rate to the greatest extent, improves the efficiency of problem fine grit classification, can be most
The saving human cost of limits.
Detailed description of the invention
Technical solution of the present invention is described in detail below in conjunction with the drawings and specific embodiments, so that of the invention
Characteristics and advantages become apparent.
The flow diagram of the problem of Fig. 1 is embodiment of the present invention fine grit classification;
Fig. 2 is the detailed process schematic diagram of step S102 in Fig. 1;
Fig. 3 is the curve synoptic diagram of the accuracy rate of thresholding and problem fine grit classification;
Fig. 4 is the module diagram of problem fine grit classification system of the invention;
Fig. 5 is the schematic diagram of problem fine grit classification device of the invention.
Specific embodiment
Detailed description will be provided to the embodiment of the present invention below.Although the present invention will combine some specific embodiments
It is illustrated and illustrates, but should be noted that the present invention is not merely confined to these embodiments.On the contrary, to the present invention
The modification or equivalent replacement of progress, are intended to be within the scope of the claims of the invention.
Some exemplary embodiments are described as the processing or method described as flow chart.Although flow chart grasps items
It is described into the processing of sequence, but many of these operations can be implemented concurrently, concomitantly or simultaneously.In addition, each
The sequence of item operation can be rearranged.The processing can be terminated when its operations are completed, it is also possible to have not
Including additional step in the accompanying drawings.The processing can correspond to method, function, regulation, subroutine, subprogram etc..
The flow diagram of the problem of Fig. 1 is embodiment of the present invention fine grit classification method, as shown in Figure 1, of the invention
Problem fine grit classification method comprising steps of
S101: semantic primitive is extracted, and extracts semantic primitive in former question text.
S102: semantic primitive extension is extended semantic primitive using vector space model, to obtain scaling problem text
This.
S103: Chinese word coding, using two-way length, memory network carries out Chinese word coding to scaling problem text in short-term, to be encoded
Question text.
S104: word focus, to encoded question text using attention mechanism progress word focusing, with obtain question text to
Amount.
S105: fine grit classification carries out fine grit classification to question text vector using softmax classifier.
First such as step S101, semantic primitive is extracted in former question text.Former question text is the original proposition of user
Need the problem of being answered or being explained text.Former question text is analyzed using interdependent parsing tree, by according to
Syntax traversal of tree is deposited, finds out noun phrase node and verb phrase node all in the former question text as semantic single
Member.
Interdependent syntactic analysis is proposed at first by French linguist L.Tesniere.He is interdependent at one by the analysis of sentence
Syntax tree, is depicted the dependence between each word, that is, indicates in syntactical Matching Relation between word, this to take
It is associated with semanteme with relationship.It is generally acknowledged that noun phrase and verb phrase most can reflect the semantic letter an of sentence
Breath, therefore by interdependent syntax traversal of tree, finding out wherein all noun phrase node and verb phrase node as whole
The semantic primitive of a sentence.
Such as step S102, semantic primitive is extended using vector space model, to obtain scaling problem text.Fig. 2
For the detailed process schematic diagram of step S102 in Fig. 1.As shown in Fig. 2, step S1021 is first carried out, generative semantics unit vector,
Semantic primitive vector is converted by semantic primitive, converts semantic primitive vector for semantic primitive.Semantic primitive vector VuSymbol
Close formula:
Vu=Vu1+Vu2+…+Vum={ z1,z2,…,zd}
Wherein, Vu1, Vu2, VumUnit vector corresponding to multiple phrase nodes in semantic primitive is respectively indicated, m is derived from so
Number;Z1, Z2, ZdVector set is constituted, d takes natural number.
Such as step S1022, the cosine similarity of the vector of all words in semantic primitive vector and Word2Vec model is calculated
cosθ.Word2Vec model is the tool of a kind of open source that Google provides calculated for term vector, Word2Vec model
Including Skip-gram algorithm, Skip-gram algorithm be calculate input word input term vector and target word output word to
Cosine similarity between amount.In the present embodiment, calculating semantic primitive vector is input term vector, is owned in Word2Vec model
The vector of word is output term vector, calculates cosine similarity cos θ between the two, and calculation formula is as follows,
Wherein yiIndicate value of the word in i dimension.
Such as step S1023, scaling problem text is obtained.Before the cosine similarity calculated in selection step S1022 is maximum
Several target words being worth in corresponding Word2Vec model are as scaling problem text.
Preceding several values determine that the cosine similarity of i.e. target word needs to be greater than threshold value by given threshold.Fig. 3 is thresholding
With the curve synoptic diagram of the accuracy rate of problem fine grit classification, abscissa indicates thresholding, and ordinate indicates problem fine grit classification
Accuracy rate, curve negotiating EM algorithm (Expectation Maximization Algorithm, EM) passes through in Fig. 3
Training study fitting.EM algorithm is a kind of iterative algorithm, for the probability containing hidden variable (latent variable)
The maximal possibility estimation or maximum a posteriori estimate of parameter model.As shown in figure 3, threshold value value is 35 in the present embodiment, when
When threshold value is greater than 35, accuracy rate is higher.
Such as step S103 and step S104, Chinese word coding is carried out to scaling problem text and word focuses.S103 Chinese word coding
Step include by the two-way length of scaling problem text input in short-term memory network (Long Short-Term Memory, LSTM) to obtain
Obtain encoded question text.LSTM meets following formula:
Wherein xiIndicate term vector,To unit LSTM before indicating, according to from the 1st word to the direction of the T word
Word is read,It indicates reversed unit LSTM, reads word according to from the T word to the direction of the 1st word,It indicates just
Output to i-th layer of hidden layer,Indicate the output of reversed i-th layer of hidden layer.
In S104 word focus steps, word is carried out using attention mechanism to the encoded question text that S103 step obtains and is gathered
Coke, to obtain question text vector.Specifically, word, which focuses, uses formula following formula:
ui=tanh (Ewhi+bw)
Wherein αiIndicate that normalized weighted value, s indicate finally obtained question text vector, EwAnd μwIndicate random first
The weight parameter value of beginningization, bwFor amount of bias, μiFor the output of Single layered perception neural networks, hiFor the output of hidden layer.
After obtaining question text vector S, execute step S105, using softmax classifier to question text vector S into
Row fine grit classification, and output category result.
Softmax classifier includes formula:
P=softmax (Wcs+bc)
Wherein j indicates the label of question text vector s, WcIndicate the weight parameter value of random initializtion, bcFor amount of bias, p
Value is the probability score value of each classification, and L indicates loss function value.
It can be seen that the present invention is in former question text by the description above to problem fine grit classification method of the invention
Extract semantic primitive, semantic primitive is extended using vector space model, thus improve problem semanteme is obtained it is comprehensive
Property and accuracy.Chinese word coding is carried out to scaling problem text and word focuses, the vectorization using softmax classifier to obtaining
The problem of expression text carries out fine grit classification, improves the accuracy rate of problem fine grit classification.Problem fine granularity of the invention
The step of classification method, completes essentially by machine, realizes this process to the greatest extent under the premise of guaranteeing certain accuracy rate
Automation, improve the efficiency of problem fine grit classification, human cost can be saved to greatest extent.
The present invention also provides a kind of problem fine grit classification system 100, Fig. 4 is problem fine grit classification system of the invention
The module diagram of system.As shown in figure 4, problem fine grit classification system 100 includes semantic primitive extraction module 101, it is semantic single
First expansion module 102, Chinese word coding device module 103, word focus module 104 and softmax classifier modules 105.
Semantic primitive extraction module 101 is for extracting semantic primitive in former question text.Semantic primitive expansion module 102
For being extended to semantic primitive using vector space model, to obtain scaling problem text.Chinese word coding device module 103 is used for
Using two-way length, memory network carries out Chinese word coding to scaling problem text in short-term, to obtain encoded question text.Word focus module
104 for carrying out word focusing using attention mechanism to encoded question text, to obtain question text vector.Softmax classification
Device module 105 carries out fine grit classification to question text vector using softmax classifier.
Problem fine grit classification system 100 implements each step of above problem fine grit classification method, and specific steps are such as
Above to the description of fine grit classification method, details are not described herein again.
The present invention also provides a kind of problem fine grit classification device 200, Fig. 5 is that problem fine grit classification of the invention fills
The schematic diagram set.As shown in figure 5, problem fine grit classification device 200 includes storage unit 201 and execution unit 202.Storage
Unit 201 is for storing program, each step of problem of implementation fine grit classification method when program unit 201 processed executes.
Execution unit 202 is for executing program in storage unit 201.Specific steps retouch problem fine grit classification method with reference to above-mentioned
It states, details are not described herein again.
It should be noted that the present invention is the application to problem fine grit classification technology.During realization of the invention,
It can be related to carrying out the knowledge of semantic study, the application to Word2Vec model by neural network.Such as reading over application
After file, accurate understanding realization principle and goal of the invention of the invention, in the case where combining existing well-known technique, this field
Technical staff can realize the present invention with the knowledge for the software programming technical ability and neural network that it is grasped completely.It is aforementioned
Word2Vec model, long memory network, the specific algorithm and step of softmax classifier can refer in the prior art in short-term
Structure and method, category this scope that all the present patent application files refer to, applicant will not enumerate.
The above is only specific application examples of the invention, are not limited in any way to protection scope of the present invention.Except above-mentioned
Outside embodiment, the present invention can also have other embodiment.All technical solutions formed using equivalent substitution or equivalent transformation,
It falls within scope of the present invention.
Claims (8)
1. a kind of problem fine grit classification method, for carrying out Question Classification to former question text, which is characterized in that including following
Step:
Semantic primitive extraction step extracts semantic primitive in the former question text;
Semantic primitive spread step is extended institute's meaning elements using vector space model, to obtain scaling problem text
This;
Chinese word coding step, using two-way length, memory network carries out Chinese word coding to the scaling problem text in short-term, to be encoded
Question text;
Word focus steps carry out word focusing using attention mechanism to the encoded question text, to obtain question text vector.
Fine grit classification step carries out fine grit classification to described problem text vector using softmax classifier.
2. problem fine grit classification method as described in claim 1, which is characterized in that
Semantic primitive extraction step includes:
By finding out noun phrase node and verb phrase section all in the former question text to interdependent syntax traversal of tree
Point is used as institute's meaning elements.
3. problem fine grit classification method as described in claim 1, which is characterized in that institute's meaning elements spread step packet
It includes,
Institute's meaning elements are converted semantic primitive vector, institute meaning elements vector V by generative semantics unit vector stepuSymbol
Close formula:
Vu=Vu1+Vu2+…+Vum={ z1,z2,…,zd}
Wherein, Vu1, Vu2, VumIt respectively indicates unit vector corresponding to multiple phrase nodes, m in semantic primitive and takes natural number;Z1,
Z2, ZdVector set is constituted, d takes natural number;
Cosine similarity step is calculated, the cosine of the vector of all words in institute's meaning elements vector and Word2Vec model is calculated
Similarity;
Scaling problem text steps are obtained, selects the corresponding word of the maximum preceding several values of cosine similarity or phrase as expansion
Open up question text.
4. problem fine grit classification method as described in claim 1, which is characterized in that described double in the Chinese word coding step
To long memory network coincidence formula in short-term:
Wherein xiIndicate term vector,To unit LSTM before indicating, read according to from the 1st word to the direction of the T word
Word,It indicates reversed unit LSTM, reads word according to from the T word to the direction of the 1st word,Indicate positive the
The output of i layers of hidden layer,Indicate the output of reversed i-th layer of hidden layer.
5. problem fine grit classification method as described in claim 1, which is characterized in that institute's predicate focus steps using formula with
Obtain question text vector, the formula are as follows:
ui=tan h (Ewhi+bw)
Wherein αiIndicate that normalized weighted value, s indicate finally obtained question text vector, EwAnd μwIndicate random initializtion
Weight parameter value, bwFor amount of bias, μiFor the output of Single layered perception neural networks, hiFor the output of hidden layer.
6. problem fine grit classification method as described in claim 1, which is characterized in that in the fine grit classification step,
Softmax classifier includes formula:
P=softmax (Wcs+bc)
Wherein j indicates the label of question text vector s, WcIndicate the weight parameter value of random initializtion, bcFor amount of bias, p value is
The probability score value of each classification, L indicate loss function value.
7. a kind of problem fine grit classification system, for carrying out Question Classification to former question text characterized by comprising
Semantic primitive extraction module, for extracting semantic primitive in the former question text;
Semantic primitive expansion module is asked for being extended to institute's meaning elements using vector space model with obtaining extension
Inscribe text;
Chinese word coding device module, for using two-way length, memory network carries out Chinese word coding to the scaling problem text in short-term, to obtain
Obtain encoded question text;
Word focus module, for carrying out word focusing using attention mechanism to the encoded question text, to obtain question text
Vector;
Softmax classifier modules carry out fine grit classification to described problem text vector using softmax classifier.
8. a kind of problem fine grit classification device, for carrying out Question Classification to former question text characterized by comprising
Storage unit, for storing program, described program unit processed realizes any one of claim 1 to 6 when executing
The step of described problem fine grit classification method;
Execution unit, for executing program described in storage unit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710678652.4A CN109388706A (en) | 2017-08-10 | 2017-08-10 | A kind of problem fine grit classification method, system and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710678652.4A CN109388706A (en) | 2017-08-10 | 2017-08-10 | A kind of problem fine grit classification method, system and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109388706A true CN109388706A (en) | 2019-02-26 |
Family
ID=65414192
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710678652.4A Pending CN109388706A (en) | 2017-08-10 | 2017-08-10 | A kind of problem fine grit classification method, system and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109388706A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110059718A (en) * | 2019-03-18 | 2019-07-26 | 国网浙江省电力有限公司信息通信分公司 | Fine granularity detection method based on the more attention mechanism of multiclass |
CN110146420A (en) * | 2019-04-12 | 2019-08-20 | 中国石油大学(北京) | A kind of glutenite granularity intelligent analysis system and method |
CN110209824A (en) * | 2019-06-13 | 2019-09-06 | 中国科学院自动化研究所 | Text emotion analysis method based on built-up pattern, system, device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1804829A (en) * | 2006-01-10 | 2006-07-19 | 西安交通大学 | Semantic classification method for Chinese question |
US20120253792A1 (en) * | 2011-03-30 | 2012-10-04 | Nec Laboratories America, Inc. | Sentiment Classification Based on Supervised Latent N-Gram Analysis |
CN104572892A (en) * | 2014-12-24 | 2015-04-29 | 中国科学院自动化研究所 | Text classification method based on cyclic convolution network |
CN104991891A (en) * | 2015-07-28 | 2015-10-21 | 北京大学 | Short text feature extraction method |
CN106126596A (en) * | 2016-06-20 | 2016-11-16 | 中国科学院自动化研究所 | A kind of answering method based on stratification memory network |
CN106294684A (en) * | 2016-08-06 | 2017-01-04 | 上海高欣计算机系统有限公司 | The file classification method of term vector and terminal unit |
-
2017
- 2017-08-10 CN CN201710678652.4A patent/CN109388706A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1804829A (en) * | 2006-01-10 | 2006-07-19 | 西安交通大学 | Semantic classification method for Chinese question |
US20120253792A1 (en) * | 2011-03-30 | 2012-10-04 | Nec Laboratories America, Inc. | Sentiment Classification Based on Supervised Latent N-Gram Analysis |
CN104572892A (en) * | 2014-12-24 | 2015-04-29 | 中国科学院自动化研究所 | Text classification method based on cyclic convolution network |
CN104991891A (en) * | 2015-07-28 | 2015-10-21 | 北京大学 | Short text feature extraction method |
CN106126596A (en) * | 2016-06-20 | 2016-11-16 | 中国科学院自动化研究所 | A kind of answering method based on stratification memory network |
CN106294684A (en) * | 2016-08-06 | 2017-01-04 | 上海高欣计算机系统有限公司 | The file classification method of term vector and terminal unit |
Non-Patent Citations (5)
Title |
---|
ZEYNEP AKATA等: "Evaluation of output embeddings for fine-grained image classification", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
奚雪峰等: "面向自然语言处理的深度学习研究", 《自动化学报》 * |
熊回香: "面向Web3.0的大众分类研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
谢雨飞等: "基于语义扩展与注意力网络的问题细粒度分类", 《计算机工程》 * |
钱晓东: "基于神经网络等技术的数据与文本聚分类研究", 《中国优秀博硕士学位论文全文数据库 (博士) 信息科技辑》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110059718A (en) * | 2019-03-18 | 2019-07-26 | 国网浙江省电力有限公司信息通信分公司 | Fine granularity detection method based on the more attention mechanism of multiclass |
CN110146420A (en) * | 2019-04-12 | 2019-08-20 | 中国石油大学(北京) | A kind of glutenite granularity intelligent analysis system and method |
CN110209824A (en) * | 2019-06-13 | 2019-09-06 | 中国科学院自动化研究所 | Text emotion analysis method based on built-up pattern, system, device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108829719B (en) | Non-fact question-answer selection method and system | |
Young et al. | Augmenting end-to-end dialogue systems with commonsense knowledge | |
Huang et al. | Deep sentiment representation based on CNN and LSTM | |
Fu et al. | Zero-shot object recognition by semantic manifold distance | |
Ji et al. | Representation learning for text-level discourse parsing | |
CN102662931B (en) | Semantic role labeling method based on synergetic neural network | |
CN110321563B (en) | Text emotion analysis method based on hybrid supervision model | |
Sen et al. | Approximate computing for long short term memory (LSTM) neural networks | |
Yan et al. | Joint learning of response ranking and next utterance suggestion in human-computer conversation system | |
CN111274790B (en) | Chapter-level event embedding method and device based on syntactic dependency graph | |
CN113743097B (en) | Emotion triplet extraction method based on span sharing and grammar dependency relationship enhancement | |
CN112015868A (en) | Question-answering method based on knowledge graph completion | |
CN112883193A (en) | Training method, device and equipment of text classification model and readable medium | |
CN114722820A (en) | Chinese entity relation extraction method based on gating mechanism and graph attention network | |
CN109388706A (en) | A kind of problem fine grit classification method, system and device | |
CN113204611A (en) | Method for establishing reading understanding model, reading understanding method and corresponding device | |
CN113705196A (en) | Chinese open information extraction method and device based on graph neural network | |
CN114036303A (en) | Remote supervision relation extraction method based on double-granularity attention and confrontation training | |
CN112528136A (en) | Viewpoint label generation method and device, electronic equipment and storage medium | |
CN114492451B (en) | Text matching method, device, electronic equipment and computer readable storage medium | |
CN116127099A (en) | Combined text enhanced table entity and type annotation method based on graph rolling network | |
CN114781380A (en) | Chinese named entity recognition method, equipment and medium fusing multi-granularity information | |
CN110728135A (en) | Text theme indexing method and device, electronic equipment and computer storage medium | |
Chen et al. | Gaussian mixture embeddings for multiple word prototypes | |
Kang et al. | A short texts matching method using shallow features and deep features |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190226 |