CN113361277A - Medical named entity recognition modeling method based on attention mechanism - Google Patents

Medical named entity recognition modeling method based on attention mechanism Download PDF

Info

Publication number
CN113361277A
CN113361277A CN202110667423.9A CN202110667423A CN113361277A CN 113361277 A CN113361277 A CN 113361277A CN 202110667423 A CN202110667423 A CN 202110667423A CN 113361277 A CN113361277 A CN 113361277A
Authority
CN
China
Prior art keywords
vector
sentence
medical
sequence
bgru
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
Application number
CN202110667423.9A
Other languages
Chinese (zh)
Inventor
李天瑞
张世豪
贾真
杜圣东
滕飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southwest Jiaotong University
Original Assignee
Southwest Jiaotong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southwest Jiaotong University filed Critical Southwest Jiaotong University
Priority to CN202110667423.9A priority Critical patent/CN113361277A/en
Publication of CN113361277A publication Critical patent/CN113361277A/en
Pending legal-status Critical Current

Links

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
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)
  • Machine Translation (AREA)

Abstract

The invention discloses a medical named entity recognition modeling method based on an attention mechanism. Firstly, converting each word in an input medical text statement into a word vector by using a vector representation technology, then obtaining rich context information in the medical text statement by using BGRU, then selecting the importance degree of the context semantic information by using an attention mechanism, and finally obtaining a global optimal solution of a medical entity label sequence by CRF to complete the identification of the medical named entity. The invention constructs a medical named entity recognition model based on an attention mechanism, and adopts a network framework of RNN + CRF. Wherein, the RNN part uses BGRU network, compares with BLSTM commonly used, and its structure is simpler, and training speed is faster, and the effect is also better. An attention mechanism is introduced on the basis of the network framework of the RNN + CRF to select the importance degree of the context information, so that the entity identification effect is improved.

Description

Medical named entity recognition modeling method based on attention mechanism
Technical Field
The invention belongs to the technical field of natural language processing, and particularly relates to a medical named entity recognition modeling method based on an attention mechanism.
Background
With the advance of medical informatization, the medical field has accumulated a huge amount of unstructured text data, which contains a large amount of valuable information. How to extract effective information from the medical texts and store and manage the effective information to construct a large-scale and high-quality medical knowledge map has great significance for the development of medical informatization, and is also a research hotspot in the field of natural language processing. Named entity recognition is one of the core tasks of medical text structured information extraction, and aims to identify entities with specific meanings from unstructured text.
Traditional named entity recognition methods mainly include rule-based, dictionary-based, and machine learning-based methods. The rule and dictionary based method requires a domain expert to manually write a rule template, or realizes the identification of an entity by means of character string matching with the help of a domain dictionary. The named entity recognition is regarded as a sequence labeling problem based on a machine learning method, so that the named entity is recognized by using targeted feature engineering and a proper machine learning model, and the commonly used machine learning model comprises maximum entropy, a Support Vector Machine (SVM), a Conditional Random Field (CRF) and the like. Although all the methods achieve certain effects, the methods need to rely on experts in the medical field to establish medical rules or medical dictionaries or rely on manually designed features to train machine learning models, which not only takes a lot of time and effort, but also limits the recognition effect to the quality of the manually designed rules, dictionaries or features. In recent years, with the development of deep learning, neural network-based methods have been applied to entity recognition tasks, and many research results have been obtained. The method does not depend on artificial design features, and all related features are automatically learned by a neural network.
At present, the network framework of 'RNN + CRF' combining a cyclic neural network and a conditional random field is a mainstream model in a named entity recognition task. Due to the particularity of the medical field, entities in the medical text have the characteristics of strong medical specialization, a large number of abbreviations and the like, and the required medical entities can be extracted more accurately and effectively only by depending on context information with high association degree and strong dependency. However, a simple "RNN + CRF" network framework can only learn the context information of a sentence from the RNN model, and cannot select the importance of the context information.
Disclosure of Invention
The invention provides a medical named entity recognition modeling method based on an attention mechanism, aiming at the problem that an entity in a medical text has poor entity recognition effect caused by strong medical specialization, a large number of abbreviations and the like. Firstly, converting each word in an input medical text statement into a word vector by using a vector representation technology, then obtaining rich context information in the medical text statement by using BGRU, then selecting the importance degree of the context semantic information by using an attention mechanism, and finally obtaining a global optimal solution of a medical entity label sequence by CRF to complete the identification of the medical named entity.
The medical named entity recognition modeling method based on the attention mechanism comprises the following steps:
step 1: vectorizing the medical text statement sequence X to obtain an input feature vector W, which specifically comprises the following steps:
the sentence sequence X with the length of n is equal to (X)1，x2，...，xn) Word x iniInto a low-dimensional dense real-valued vector wiWord vectors of words are embedded by words in a matrix WcharIs represented by a vector code ofcharIs | V | × d, where | V | is a fixed-size input word table and d is the dimension of the word vector; wherein i belongs to [1, 2]；
Representing an input feature vector of a medical text sentence as W ═ W (W ═ W1，w2，...，wn)；
Step 2: learning the context information of the medical text sentence from the input feature vector W by using a bidirectional gate control loop unit network BGRU to obtain a sentence vector H, which specifically comprises the following steps:
the BGRU obtains the state output of the hidden layer from the upper information and the lower information of the medical text sentence from the input feature vector W by a forward GRU network and a backward GRU network respectively
Figure BDA0003117428920000021
And
Figure BDA0003117428920000022
Figure BDA0003117428920000023
Figure BDA0003117428920000024
wherein
Figure BDA0003117428920000025
And
Figure BDA0003117428920000026
respectively representing hidden layer state output of a forward GRU network and a backward GRU network at the time t, wherein t belongs to [1, 2]；
BGRU splices hidden layer state outputs of forward GRU network and backward GRU network to obtain sentence vector H ═ (H)1，h2，...，hn) Wherein the hidden layer state output of BGRU at time t is:
Figure BDA0003117428920000027
and step 3: selecting the importance degree of the context information in the sentence vector H by using an attention mechanism to obtain a feature vector M of the sentence, wherein the method specifically comprises the following steps:
performing attention weight calculation on the sentence vector H to obtain an attention weight vector a:
a＝softmax(watanh(H))；
wherein waIs the weight vector to be learned, tanh (-) is the hyperbolic tangent function;
and the sentence vector H carries out weighted summation according to the attention weight vector a to obtain a feature vector M of the sentence:
M＝aH；
and 4, step 4: decoding the feature vector M by using a conditional random field CRF to obtain a final output sequence Y of the input statement X*The method specifically comprises the following steps:
the feature vector M of the sentence obtained is (M)1，m2，...，mn) Calculating the conditional probability of the possible output label sequence Y:
P(Y|M)＝CRF(M，Y)；
wherein Y ∈ Yx，YxRepresents all possible output tag sequences of the input sequence X;
finally, the output label sequence Y with the maximum conditional probability is used*As the final output sequence of input statement X:
Y*＝argmaxP(Y|M)。
the invention constructs a medical named entity recognition model based on an attention mechanism, and adopts a network framework of RNN + CRF. Wherein, the RNN part uses BGRU network, compares with BLSTM commonly used, and its structure is simpler, and training speed is faster, and the effect is also better. An attention mechanism is introduced on the basis of the network framework of the RNN + CRF to select the importance degree of the context information, so that the entity identification effect is improved.
Drawings
FIG. 1 is a structural diagram of a medical named entity recognition model based on an attention mechanism.
Detailed Description
The specific implementation steps are as follows:
step 1: vectorizing the medical text statement by using a vector representation technology to obtain an input feature vector:
the medical text sentence sequence X with the length of n is equal to (X)1，x2，...，xn) Word x iniInto a low-dimensional dense real-valued vector wiWord vectors of words are embedded by words in a matrix WcharIs represented by a vector code ofcharIs | V | × d, where | V | is a fixed-size input word table and d is the dimension of the word vector; wherein i belongs to [1, 2]；
Thus, the input feature vector of the medical text sentence may be expressed as W ═ (W ═ W1w2，...，wn)。
Step 2: and learning the context information of the medical text sentence from the input feature vector by using a bidirectional gating circulation unit network BGRU to obtain a sentence vector.
For named entity recognition, this sequence tagging problem is suitably learned using LSTM to solve the problem of dependency on sequence data. GRU as LSTM variant can learn sequence data dependence well, solve RNN gradient disappearance problem, and has simpler structure, faster training speed and better effect than LSTM. Therefore, here, the input sequence data is processed using the GRU.
The BGRU learns the context information of the text sentence by combining a forward GRU network and a backward GRU network; the forward GRU network and the backward GRU network control information flow by setting an update gate z and a reset gate r, so as to realize the update, the accept and the reject and the storage of historical information; wherein, the information flow of the forward GRU network comprises the input information w of the current time ttAnd hidden layer state output h of GRU at previous momentt-1；
Updating the door z at time ttAnd a reset gate rtThe calculation method is as follows:
zt＝σ(Wwzwt+Whzht-1+bz)；
rt＝σ(Wwrwt+Whrht-1+br)；
where σ (-) stands for sigmoid functionNumber, Wwz、WhzTo update the weight matrix to be learned in the gate, bzTo update the offset vector of the gate, Wwr、WhrTo reset the weight matrix to be learned in the gate, brA bias vector to reset the gate;
then, use the reset gate rtObtaining candidate information of GRU hidden layer at current time t
Figure BDA0003117428920000041
The calculation method is as follows:
Figure BDA0003117428920000042
wherein tanh (-) represents a hyperbolic tangent function,
Figure BDA0003117428920000043
representing the Hadamard (Hadamard) product, Wwh、WhhWeight matrix to be learned in the candidate information of the hidden layer at the current moment, bhA bias vector in the candidate information of the hidden layer at the current moment is obtained;
finally, with the refresh door ztRespectively carrying out Hadamard multiplication with hidden layer state output of GRU at the previous moment and candidate information of the hidden layer at the current moment to obtain hidden layer state output of GRU at the current moment
Figure BDA0003117428920000044
Figure BDA0003117428920000045
The forward GRU network is used for learning the above information of the medical text sentence, and the backward GRU network is used for learning the below information of the medical text sentence, and the information flow of the backward GRU network comprises the input information w of the current time ttAnd hidden layer state output h of GRU at the later timet+1The calculation mode is the same as that of the forward GRU network;
BGRU is prepared by mixingAnd splicing hidden layer state outputs of the GRU network and the backward GRU network to obtain a sentence vector H ═ H1，h2，...，hn) Wherein at time t, the hidden layer output of the BGRU is:
Figure BDA0003117428920000046
wherein
Figure BDA0003117428920000047
And
Figure BDA0003117428920000048
respectively representing hidden layer state output of the forward GRU network and the backward GRU network at the time t.
And step 3: and selecting the importance degree of the context information in the sentence vector by using an attention mechanism to obtain the feature vector of the sentence.
Context dependence information of medical text sentences can be learned relatively comprehensively by using the BGRU, so that current characters can be effectively recognized. However, each context information is not of the same importance for identifying the current character. Therefore, by using an attention mechanism behind the BGRU, the attention of the context information with higher relevance and stronger dependence on the current character can be enhanced, and the attention of the context information with lower relevance and weaker dependence is weakened, so that the recognition effect of the entity is improved.
Specifically, the attention weight calculation is performed on the sentence vector H output by the BGRU network in step 2, resulting in an attention weight vector a:
a＝softmax(watanh(H))，
wherein waIs the weight vector to be learned, tanh (-) is the hyperbolic tangent function;
and (3) carrying out weighted summation on the sentence vector H output by the BGRU network according to the attention weight vector a to obtain a feature vector M of the sentence:
M＝aH。
and 4, step 4: and performing joint decoding on the prediction labels by using the CRF to obtain a final output sequence of the input statement.
Named entity recognition is a sequence tagging problem, in which each tag in a predicted tag sequence is not predicted independently, and it needs to be combined with previous and next tags to predict more accurately. For example, for an entity composed of multiple characters, the labels of each character are consistent with the labels of the entity category. If the label predicts one of the characters independently, this information is not available, which may lead to prediction errors. Although the context information of the current character can be well learned by the BGRU, the prediction of the label is independent, and the label bias problem is generated. Therefore, the attention added BGRU is followed by a CRF to carry out joint decoding on the label sequence, so that the label sequence at the current position can be well predicted according to the labels at the front and rear positions, and the globally optimal label sequence can be obtained.
Specifically, the sentence obtained in step 3 has a feature vector M ═ M (M)1，m2，...，mn) And calculating the conditional probability of the possible output label sequence Y in the following way:
Figure BDA0003117428920000051
S(M，Y)＝∑i，kλktk(yi-1，yi，m，i)+∑i，lμlsl(yi，m，i)
wherein t iskAnd slAre all characteristic functions, tkIs a state feature function for extracting the features of the state sequence, its state y at the current momentiDependent on the state y at the previous momenti-1The influence of (a); stIs a transfer feature function for extracting the features of the observation sequence, its state y at the current momentiWill be observed by the current moment miThe influence of (c). The characteristic function can only take a value of 0 or 1, and takes a value of 1 when a certain characteristic is satisfied, and takes a value of 0 when the certain characteristic is not satisfied. Lambda [ alpha ]k、μlThe weights of the two characteristic functions are used for measuring the importance degree of the current characteristic. Y isXRepresents all possible output tag sequences of the input sequence X;
finally, the output label sequence Y with the maximum conditional probability is used*As the final output sequence of input statement X:
Y*＝argmaxP(Y|M)。
to verify the effectiveness of the present invention, the present invention evaluates task one at CHIP 2020: a comparison experiment is carried out on a data set identified by the named entity of the Chinese medical text and 3 named entity identification models in the medical field, wherein the 3 comparison models are as follows:
(1) CRF: and after converting the input medical text into a word vector, performing entity identification by adopting a CRF model.
(2) BLSTM-CRF: the context-dependent information is learned from the input feature vectors using BLSTM, and then entity identification is performed using CRF joint decoding.
(3) BLSTM-CRF: an attention mechanism is introduced on the basis of BLSTM-CRF, and the importance degree of the context information is selected by adding the attention mechanism after BLSTM.
The evaluation indexes adopt precision (P), Recall (R) and F1 (F1-score), wherein the precision represents the proportion of the quantity of a certain medical entity type which is correctly predicted to the quantity of the medical entity type, the Recall represents the proportion of the quantity of the certain medical entity type which is correctly predicted to the quantity of the medical entity type which is truly predicted to the quantity of the medical entity type, and the F1 is a harmonic mean of the precision and the Recall, and simultaneously takes into account the two contradictory indexes of the precision and the Recall, thereby being capable of carrying out more comprehensive and overall evaluation on the performance of the model.
Let J ═ J1，j2，...，jn) For the real label set of the medical entity, the label set of the medical entity predicted by the medical named entity recognition model of the invention is set as K ═ K (K1，k2，...，km). Any one element in the two sets represents oneEach element of the medical entity comprises four contents of a sentence sequence number, an entity starting position, an entity ending position and an entity type. The sentence sequence number represents the number of the sentence in which the entity is located in the data set, the start position of the entity represents the position of the first character of the entity in the sentence, the end position of the entity represents the position of the last character of the entity in the sentence, and the entity type is the type of the medical entity. For any two elements j in two setsi、klTwo elements are equivalent if and only if their sentence numbers, start positions of the entities, end positions of the entities, and types of the entities are the same.
Based on this, the accuracy, recall and F1 values were calculated as follows:
Figure BDA0003117428920000061
Figure BDA0003117428920000062
Figure BDA0003117428920000071
where # represents the intersection of sets, i.e., the element that both sets share;
the results of the experiment are shown in table 1:
TABLE 1 results of the experiment
Figure BDA0003117428920000072
Experimental results show that the BGRU-att-CRF model provided by the invention has better effect than all comparison models, and has the best effect on accuracy, recall rate and F1 value.

Claims (1)

1. The medical named entity recognition modeling method based on the attention mechanism is characterized by comprising the following steps:
step 1: vectorizing the medical text statement sequence X to obtain an input feature vector W, which specifically comprises the following steps:
the sentence sequence X with the length of n is equal to (X)1，x2，...，xn) Word x iniInto a low-dimensional dense real-valued vector wiWord vectors of words are embedded by words in a matrix WcharIs represented by a vector code ofcharIs | V | × d, where | V | is a fixed-size input word table and d is the dimension of the word vector; wherein i belongs to [1, 2]；
Representing an input feature vector of a medical text sentence as W ═ W (W ═ W1，w2，...，wn)；
Step 2: learning the context information of the medical text sentence from the input feature vector W by using a bidirectional gate control loop unit network BGRU to obtain a sentence vector H, which specifically comprises the following steps:
the BGRU obtains the state output of the hidden layer from the upper information and the lower information of the medical text sentence from the input feature vector W by a forward GRU network and a backward GRU network respectively
Figure FDA0003117428910000011
And
Figure FDA0003117428910000012
Figure FDA0003117428910000013
Figure FDA0003117428910000014
wherein
Figure FDA0003117428910000015
And
Figure FDA0003117428910000016
respectively representing hidden layer state output of a forward GRU network and a backward GRU network at the time t, wherein t belongs to [1, 2]；
BGRU splices hidden layer state outputs of forward GRU network and backward GRU network to obtain sentence vector H ═ (H)1，h2，...，hn) Wherein the hidden layer state output of BGRU at time t is:
Figure FDA0003117428910000017
and step 3: selecting the importance degree of the context information in the sentence vector H by using an attention mechanism to obtain a feature vector M of the sentence, wherein the method specifically comprises the following steps:
performing attention weight calculation on the sentence vector H to obtain an attention weight vector a:
a＝softmax(wataah(H))；
wherein waIs the weight vector to be learned, tanh (-) is the hyperbolic tangent function;
and the sentence vector H carries out weighted summation according to the attention weight vector a to obtain a feature vector M of the sentence:
M＝aH；
and 4, step 4: decoding the feature vector M by using a conditional random field CRF to obtain a final output sequence Y of the input statement X*The method specifically comprises the following steps:
the feature vector M of the sentence obtained is (M)1，m2，...，mn) Calculating the conditional probability of the possible output label sequence Y:
P(Y|M)＝CRF(M，Y)；
wherein Y ∈ YX，YXRepresents all possible output tag sequences of the input sequence X;
finally, the output label sequence Y with the maximum conditional probability is used*As the final output sequence of input statement X:
Y*＝argmaxP(Y|M)。
CN202110667423.9A 2021-06-16 2021-06-16 Medical named entity recognition modeling method based on attention mechanism Pending CN113361277A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110667423.9A CN113361277A (en) 2021-06-16 2021-06-16 Medical named entity recognition modeling method based on attention mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110667423.9A CN113361277A (en) 2021-06-16 2021-06-16 Medical named entity recognition modeling method based on attention mechanism

Publications (1)

Publication Number Publication Date
CN113361277A true CN113361277A (en) 2021-09-07

Family

ID=77534506

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110667423.9A Pending CN113361277A (en) 2021-06-16 2021-06-16 Medical named entity recognition modeling method based on attention mechanism

Country Status (1)

Country Link
CN (1) CN113361277A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115630649A (en) * 2022-11-23 2023-01-20 南京邮电大学 Medical Chinese named entity recognition method based on generative model

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109284361A (en) * 2018-09-29 2019-01-29 深圳追科技有限公司 A kind of entity abstracting method and system based on deep learning
CN112733541A (en) * 2021-01-06 2021-04-30 重庆邮电大学 Named entity identification method of BERT-BiGRU-IDCNN-CRF based on attention mechanism

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109284361A (en) * 2018-09-29 2019-01-29 深圳追科技有限公司 A kind of entity abstracting method and system based on deep learning
CN112733541A (en) * 2021-01-06 2021-04-30 重庆邮电大学 Named entity identification method of BERT-BiGRU-IDCNN-CRF based on attention mechanism

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115630649A (en) * 2022-11-23 2023-01-20 南京邮电大学 Medical Chinese named entity recognition method based on generative model

Similar Documents

Publication Publication Date Title
CN109657239B (en) Chinese named entity recognition method based on attention mechanism and language model learning
CN110502749B (en) Text relation extraction method based on double-layer attention mechanism and bidirectional GRU
CN109800437B (en) Named entity recognition method based on feature fusion
CN110765775B (en) Self-adaptive method for named entity recognition field fusing semantics and label differences
CN111858931B (en) Text generation method based on deep learning
CN111967266A (en) Chinese named entity recognition model and construction method and application thereof
CN111738007B (en) Chinese named entity identification data enhancement algorithm based on sequence generation countermeasure network
CN112487820B (en) Chinese medical named entity recognition method
CN113095415B (en) Cross-modal hashing method and system based on multi-modal attention mechanism
CN113190656B (en) Chinese named entity extraction method based on multi-annotation frame and fusion features
CN109284361A (en) A kind of entity abstracting method and system based on deep learning
CN111400494B (en) Emotion analysis method based on GCN-Attention
CN111881256B (en) Text entity relation extraction method and device and computer readable storage medium equipment
CN111400455A (en) Relation detection method of question-answering system based on knowledge graph
CN111581970B (en) Text recognition method, device and storage medium for network context
CN110162789A (en) A kind of vocabulary sign method and device based on the Chinese phonetic alphabet
CN111222318A (en) Trigger word recognition method based on two-channel bidirectional LSTM-CRF network
CN113360667B (en) Biomedical trigger word detection and named entity identification method based on multi-task learning
CN111145914B (en) Method and device for determining text entity of lung cancer clinical disease seed bank
CN112699685A (en) Named entity recognition method based on label-guided word fusion
CN117371523A (en) Education knowledge graph construction method and system based on man-machine hybrid enhancement
CN114841151A (en) Medical text entity relation joint extraction method based on decomposition-recombination strategy
CN111444720A (en) Named entity recognition method for English text
CN114925205A (en) GCN-GRU text classification method based on comparative learning
CN116720519B (en) Seedling medicine named entity identification 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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210907

WD01 Invention patent application deemed withdrawn after publication