[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN115238693A - Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory - Google Patents

Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory Download PDF

Info

Publication number
CN115238693A
CN115238693A CN202210809038.8A CN202210809038A CN115238693A CN 115238693 A CN115238693 A CN 115238693A CN 202210809038 A CN202210809038 A CN 202210809038A CN 115238693 A CN115238693 A CN 115238693A
Authority
CN
China
Prior art keywords
layer
model
output
bilstm
named entity
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
CN202210809038.8A
Other languages
Chinese (zh)
Inventor
张锋
程振宁
陈婕卿
曾可
姜会珍
李大伟
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.)
Beijing Anne Fox Information Consulting Co ltd
Peking Union Medical College Hospital Chinese Academy of Medical Sciences
Original Assignee
Beijing Anne Fox Information Consulting Co ltd
Peking Union Medical College Hospital Chinese Academy of Medical Sciences
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 Beijing Anne Fox Information Consulting Co ltd, Peking Union Medical College Hospital Chinese Academy of Medical Sciences filed Critical Beijing Anne Fox Information Consulting Co ltd
Priority to CN202210809038.8A priority Critical patent/CN115238693A/en
Publication of CN115238693A publication Critical patent/CN115238693A/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
    • G06N3/08—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)
  • Computational Linguistics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Machine Translation (AREA)

Abstract

The invention discloses a Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory, which improves the recognition precision of named entities by modifying a BERT-BILSTM-CRF model; determining input and output of the named entity recognition model: taking a medical text as a research object, taking a medical text data set with entity labels as the input of a named entity recognition model, wherein the output of the model is an entity label result given after medical entity prediction is carried out on the data set; according to the invention, by further enhancing the context feature extraction performance of the text of the model, on one hand, a multi-word segmentation method is considered to increase local context features, on the other hand, a multi-layer bidirectional long-term and short-term memory method is introduced, global context features are increased by setting BILSTM models with different depths, external knowledge of a medical dictionary is introduced, and the precision of the named entity recognition task is further improved by enriching semantic feature information in the model learning process.

Description

Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory
Technical Field
The invention relates to the technical field of Chinese named entity recognition, in particular to a Chinese named entity recognition method based on multi-participle and multi-layer bidirectional long-short term memory.
Background
Named entity recognition is a challenging fundamental task in the task of natural language processing. Named entity recognition is used as a basic task in natural language processing, and plays a key role in information extraction, knowledge graph construction and the like. In some specific areas, research into named entity recognition has found widespread and mature applications. Current methods of named entity recognition focus mainly on dictionary and rule based methods, statistical machine learning based methods, and deep learning based methods.
Dictionary-based and rule-based methods for entity recognition by using string matching and manually constructed entity extraction rules can achieve better accuracy on small data sets, but are not applicable as data sets increase. The learning method based on the statistical machine comprises a hidden Markov model, a support vector machine, a conditional random field and the like. While these approaches reduce vocabulary and rule-based workload to some extent, the need for manual assignment of features and extrinsic knowledge information is inevitable. Therefore, the methods can only be generally suitable for the current field, and the problem of named entity identification in a brand-new field is difficult to directly solve. Deep learning-based methods have gained widespread use and breakthrough in recent years, including BERT models, CNN models, BILSTM models, and the like. Compared with a machine learning model, the deep learning method can learn high-dimensional and deep feature representation, and is beneficial to improving the generalization capability of entity recognition.
Although the deep learning method of the existing research achieves better effect in the identification of the medical named entity, the identification of the medical named entity still faces some difficulties and challenges:
(1) The text representation with single granularity of the existing method can only obtain the global context characteristics of the text and lacks the local context characteristic information, thereby hindering the further improvement of the model performance;
(2) The commonly adopted single BILSTM can only capture the context characteristics of a specific dimension, and the contribution of the context characteristics to the model performance under other dimensions is not considered;
therefore, it is necessary to design a Chinese named entity recognition method based on multi-segmented words and multi-layer bidirectional long-short term memory.
Disclosure of Invention
The invention solves the problem of providing a Chinese named entity recognition method based on multi-participle and multi-layer bidirectional long-short term memory, which modifies a BERT-BILSTM-CRF model widely used in the named entity recognition technology; a multi-Word segmentation module is introduced, all data sets can be segmented into a plurality of words in the module, and then local features are extracted through Word-Level BILSTM; introducing a multi-layer BILSTM module, wherein the module consists of BILSTM and Attention, and by setting different hidden layer parameters for the BILSTM, text context characteristics with different dimensionalities can be learned, and then important information is captured by the Attention; the two modules can enrich information in the model learning process, so that the accuracy of named entity identification is improved.
In order to achieve the purpose, the invention adopts the following technical scheme:
a Chinese named entity recognition method based on multi-word segmentation and multilayer bidirectional long-short term memory improves the recognition precision of named entities by modifying a BERT-BILSTM-CRF model; the method comprises the following steps:
step S1: determining input and output of the named entity recognition model: taking a medical text as a research object, taking a medical text data set with entity labels as the input of a named entity recognition model, wherein the output of the model is an entity label result given after medical entity prediction is carried out on the data set;
step S2: designing a medical named entity recognition model with multiple words and multi-layer bidirectional long and short term memory, wherein the model consists of an input layer, a word embedding layer, a semantic feature extraction layer, a CRF layer and an output layer; the model comprises a BERT pre-training language model, a bidirectional long-short term memory model BILSTM, an attention mechanism and a conditional random field CRF; the main method of the medical named entity recognition model comprises the following steps:
(1) an input layer: this layer is used to input the data set;
(2) word embedding layer: the layer encodes characters in a text into a vector representation form through a BERT pre-training language model; the output result after passing through the BERT model is represented as V = V 1 ，V 2 ，...，V n Wherein n represents the total number of characters contained in the current sentence;
(3) semantic feature extraction layer: the layer is formed by combining a multi-word segmentation module and a multi-layer BILSTM module. The multi-Word segmentation module mainly extracts features through a Word-LeveI BILSTM module, and the multi-layer BILSTM module acquires feature information of different dimensions by setting hidden layers of different sizes and captures important information by using an attention mechanism; the specific process is as follows:
1) Multi-word segmentation module
The Word-Level BILSTM module is formed based on a BILSTM model; BILSTM is formed by combining forward LSTM and backward LSTM; LSTM is expressed by mathematical expressions as shown in equations 1-6:
f t ＝σ(W f ·[h t-1 ，x t ]+b f )#(1)
i t ＝σ(W i ·[h t-1 ，x t ]+b i )#(2)
C t ＝tanh(W C ·[h t-1 ，x t ]+b c )#(3)
C t ＝f t *C t-1 +i t *C t #(4)
o t ＝σ(W o ·[h t-1 ，x t ]+b o )#(5)
h t ＝O t *tanh(C t )#(6)
wherein t and t-1 respectively represent the current moment and the last moment, h represents a hidden state, and sigma and tanh respectively represent a sigmoid activation function and a tanh activation function; w represents a weight matrix, b represents a bias vector; * Represents a dot product;
the output of BILSTM is a forward LSTM output
Figure BDA0003739733730000041
And negative LSTM output
Figure BDA0003739733730000042
Is represented by
Figure BDA0003739733730000043
2) Multilayer BILSTM module
The module integrates a BILSTM and an Attention mechanism; setting hidden layers with different sizes for the BILSTM to extract context features with different dimensions; the Attention mechanism is used to distinguish different degrees of importance of different features;
feature vector h output by attention mechanism layer to BILSTM layer t Carrying out weight distribution, and calculating to obtain a common output feature vector W of the t-th word in a BILSTM layer and an attention layer t(k) Expressed mathematically as in equations 7-9:
Figure BDA0003739733730000044
Figure BDA0003739733730000045
score(s t ，h i )＝vtanh(w[s t ，h i ])#(9)
wherein a is t，i For the attention function, the score function is an alignment model that assigns a score based on how well the inputs and outputs match at time i, defining each output as hidden to each inputHow much weight the hidden state is; w t(k) Representing the output of the tth word passing through the kth MBA model, wherein the value of k is 1,2;
the final output O of the semantic feature extraction layer is obtained by fusing the output of the multi-word segmentation module and the output of the multi-layer BILSTM, and is expressed by a mathematical expression as formula 10:
Figure BDA0003739733730000051
the final output sequence of the layer model is [ O ] 1 ，O 2 ...，O n ]；
(4) CRF layer:
the main role of this layer is to predict the label; in the process of training data, the layer automatically learns the constraint between the labels and ensures that the predicted labels are legal; the matrix P is a scoring matrix, P i，j Is a probability value, A, that classifies the ith character as the jth token i，j Is the state transition score from the ith marker to the jth marker; if the input sentence x = (x) 1 ，x 2 …，x n ) The mark sequence is y = (y) 1 ，y 2 ，...，y n ) The score is as follows:
Figure BDA0003739733730000052
for Score (x, y), the normalization process was performed using the Softmax function, as follows:
Figure BDA0003739733730000053
in training, for training sample (x, y) x ) Maximizing the log probability of the marker sequence using the following formula;
Figure BDA0003739733730000054
the experiment adopts the Viterbi algorithm to solve the probability maximum path of the dynamic programming, and the formula is as follows:
Figure BDA0003739733730000055
Y * is the sequence with the highest score in the scoring function, i.e. the expected output of the model,
Figure BDA0003739733730000061
is a maximize score function;
(5) and (3) an output layer: the layer is used for inputting labeling results of all texts in the data set; the evaluation index is measured by the accuracy P, recall R and F1 values, as shown in equations 15, 16 and 17:
Figure BDA0003739733730000062
Figure BDA0003739733730000063
Figure BDA0003739733730000064
wherein T is p Number of correctly identified medical entities of the representation model, F p Representing the number of unrelated medical entities identified by the model, F N The number of relevant medical entities which cannot be identified by the model; f1 is the weighted harmonic mean of P and R.
The invention has the beneficial effects that: according to the invention, by further enhancing the context feature extraction performance of the text of the model, on one hand, a multi-word segmentation method is considered to increase local context features and obtain feature information of different dimensions, and an attention mechanism is utilized to capture important information, on the other hand, a multi-layer bidirectional long-short term memory method is introduced, global context features are increased by setting BILSTM models of different depths, external knowledge of a medical dictionary is introduced, and the precision of the named entity recognition task is further improved by enriching semantic feature information in the model learning process.
Drawings
FIG. 1 is an overall process flow of the medical named entity recognition model of the present invention;
FIG. 2 is the overall processing procedure of the multi-segmentation module of the present invention;
FIG. 3 is an overall process of the multi-layer BILSTM module of the present invention;
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Specific examples are given below.
Referring to fig. 1-3, a method for recognizing a named entity in chinese based on multi-segmentation and multi-layer bidirectional long-short term memory, which improves recognition accuracy of the named entity by modifying a BERT-blst (tm) -CRF model; with reference to fig. 1, the method comprises the following steps:
step S1: determining input and output of the named entity recognition model: taking a medical text as a research object, taking a medical text data set with entity labels as the input of a named entity recognition model, wherein the output of the model is an entity label result given after medical entity prediction is carried out on the data set;
step S2: designing a medical named entity recognition model with multiple word segments and multi-layer bidirectional long-term and short-term memory, wherein the model consists of an input layer, a word embedding layer, a semantic feature extraction layer, a CRF layer and an output layer; the model comprises a BERT pre-training language model, a bidirectional long-short term memory model BILSTM, an attention mechanism and a conditional random field CRF; the main method of the medical named entity recognition model sequentially comprises the following steps:
(1) an input layer: this layer is used to input the data set;
(2) word embedding layer: the layer encodes characters in a text into a vector representation form through a BERT pre-training language model; the output result after passing through the BERT model is represented as V = V 1 ，V 2 ，...，V n Wherein n represents the total number of characters contained in the current sentence;
(3) semantic feature extraction layer: the layer is formed by combining a multi-word segmentation module and a multi-layer BILSTM module. The multi-Word segmentation module mainly extracts features through a Word-LeveI BILSTM module, and the multi-layer BILSTM module acquires feature information of different dimensions by setting hidden layers of different sizes and captures important information by using an attention mechanism; the specific process is as follows:
1) Multi-word segmentation module
The Word-Level BILSTM module is formed based on a BILSTM model; the BILSTM is formed by combining a forward LSTM and a backward LSTM; LSTM is expressed by mathematical expressions as shown in equations 1-6:
f t ＝σ(W f ·[h t-1 ，x t ]+b f )#(1)
i t ＝σ(W i ·[h t-1 ，x t ]+b i )#(2)
C t ＝tanh(W C ·[h t-1 ，x t ]+b c )#(3)
C t ＝f t *C t-1 +i t *C t #(4)
o t ＝σ(W o ·[h t-1 ，x t ]+b o )#(5)
h t ＝O t *tanh(C t )#(6)
wherein t and t-1 respectively represent the current moment and the last moment, h represents a hidden state, and sigma and tanh respectively represent a sigmoid activation function and a tanh activation function; w represents a weight matrix, b represents an offset vector; * Represents a dot product;
the output of BILSTM is a forward LSTM output
Figure BDA0003739733730000081
And negative LSTM output
Figure BDA0003739733730000082
Is represented by
Figure BDA0003739733730000083
The multi-segmentation module integrates a plurality of BILSTMs as shown in figure 2; taking the text of 'postpartum diagnosis as diabetes' as an example, the sequence of the text after full word segmentation is expressed as [ 'postpartum', 'diagnosis as', 'diabetes' ], and each word is subjected to word local context feature capture through a BILSTM model;
v in the diagram of FIG. 2 i The representative word embedding layer generates a character vector for the ith word; words i Represents the ith word after the multi-word segmentation,
Figure BDA0003739733730000091
represents the output representation of the ith word after the word-level local feature analysis when the jth word appears in a certain word, W i Representing the output of the ith Word after passing through a Word-LeveI BILSTM module;
2) Multilayer BILSTM module
The module integrates the BILSTM and the Attention mechanism, see FIG. 3; hidden layers with different sizes are arranged on the BILSTM to extract context features with different dimensions; the Attention mechanism is used to distinguish different degrees of importance of different features;
feature vector h output by attention mechanism layer to BILSTM layer t Carrying out weight distribution, and calculating to obtain a common output feature vector W of the t-th word in a BILSTM layer and an attention layer t(k) Expressed mathematically as in equations 7-9:
Figure BDA0003739733730000092
Figure BDA0003739733730000093
score(s t ，h i )＝vtanh(w[s t ，h i ])#(9)
wherein a is t，i For the attention function, the score function is an alignment model, which assigns a score based on the degree of matching of the input and output at time i, defining how much weight each output gives to each input hidden state; w is a group of t(k) Representing the output of the tth word passing through the kth MBA model, wherein the value of k is 1,2;
the final output 0 of the semantic feature extraction layer is obtained by fusing the multi-word segmentation module output and the multi-layer BILSTM output, and is expressed by a mathematical expression as formula 10:
Figure BDA0003739733730000094
the final output sequence of the layer model is [ O ] 1 ，O 2 ...，O n ]；
(4) CRF layer:
the main role of this layer is to predict the label; in the process of training data, the layer automatically learns the constraint between the labels and ensures that the predicted labels are legal; the matrix P is a scoring matrix, P i，j Is a probability value, A, that classifies the ith character as the jth token i，j Is the state transition score from the ith marker to the jth marker; if the input sentence x = (x) 1 ，x 2 …，x n ) The mark sequence is y = (y) 1 ，y 2 ，...，y n ) The score is as follows:
Figure BDA0003739733730000101
for Score (x, y), the normalization process was performed using the Softmax function, as follows:
Figure BDA0003739733730000102
in the training, for trainingSample (x, y) x ) Maximizing the log probability of the marker sequence using the following formula;
Figure BDA0003739733730000103
the experiment adopts the Viterbi algorithm to solve the probability maximum path of the dynamic programming, and the formula is as follows:
Figure BDA0003739733730000104
Y * is the sequence of the scoring function with the highest score, i.e. the expected output of the model,
Figure BDA0003739733730000105
is a maximize score function;
(5) an output layer: the layer is used for inputting labeling results of all texts in the data set; the evaluation index is measured by the accuracy P, recall R and F1 values, as shown in equations 15, 16 and 17:
Figure BDA0003739733730000106
Figure BDA0003739733730000111
Figure BDA0003739733730000112
wherein T is p Representing the number of correctly identified medical entities of the model, F p The FN is the number of the relevant medical entities which can not be identified by the model; f1 is the weighted harmonic mean of P and R.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered as the technical solutions and the inventive concepts of the present invention within the technical scope of the present invention.

Claims (1)

1. A Chinese named entity recognition method based on multi-word segmentation and multilayer bidirectional long-short term memory is characterized in that the recognition precision of a named entity is improved by modifying a BERT-BILSTM-CRF model; the method comprises the following steps:
step S1: determining input and output of the named entity recognition model: taking a medical text as a research object, taking a medical text data set with entity labels as the input of a named entity recognition model, wherein the output of the model is an entity label result given after medical entity prediction is carried out on the data set;
step S2: designing a medical named entity recognition model with multiple words and multi-layer bidirectional long and short term memory, wherein the model consists of an input layer, a word embedding layer, a semantic feature extraction layer, a CRF layer and an output layer; the model comprises a BERT pre-training language model, a bidirectional long-short term memory model BILSTM, an attention mechanism and a conditional random field CRF; the main method of the medical named entity recognition model sequentially comprises the following steps:
(1) an input layer: this layer is used to input the data set;
(2) word embedding layer: the layer encodes characters in a text into a vector representation form through a BERT pre-training language model; the output result after passing through the BERT model is represented as V = V 1 ，V 2 ，...，V n Wherein n represents the total number of characters contained in the current sentence;
(3) semantic feature extraction layer: the layer is formed by combining a multi-word segmentation module and a multi-layer BILSTM module. The multi-Word segmentation module mainly extracts features through a Word-LeveI BILSTM module, and the multi-layer BILSTM module acquires feature information of different dimensions by setting hidden layers of different sizes and captures important information by using an attention mechanism; the specific process is as follows:
1) Multi-word segmentation module
The Word-Level BILSTM module is formed based on a BILSTM model; the BILSTM is formed by combining a forward LSTM and a backward LSTM; LSTM is expressed by mathematical expressions as shown in equations 1-6:
f t ＝σ(W f ·[h t-1 ，x t ]+b f )#(1)
i t ＝σ(W i ·[h t-1 ，x t ]+b i )#(2)
C t ＝tanh(W C ·[h t-1 ，x t ]+b c )#(3)
C t ＝f t *C t-1 +i t *C t #(4)
o t ＝σ(W o ·[h t-1 ，x t ]+b o )#(5)
h t ＝O t *tanh(C t )#(6)
wherein t and t-1 respectively represent the current moment and the last moment, h represents a hidden state, and sigma and tanh respectively represent a sigmoid activation function and a tanh activation function; w represents a weight matrix, b represents a bias vector; * Represents a dot product;
the output of BILSTM is a forward LSTM output
Figure FDA0003739733720000021
And negative LSTM output
Figure FDA0003739733720000022
Is represented by
Figure FDA0003739733720000023
2) Multilayer BILSTM module
The module integrates a BILSTM and an Attention mechanism; setting hidden layers with different sizes for the BlLSTM to extract context features with different dimensions; the Attention mechanism is used to distinguish different degrees of importance of different features;
feature vector h output by attention mechanism layer to BILSTM layer t Weight distribution is carried out, and common output feature vectors of the t-th word in a BILSTM layer and an attention layer are obtained through calculationW t(k) Expressed by mathematical formulas such as 7-9:
Figure FDA0003739733720000024
Figure FDA0003739733720000031
score(s t ，h i )＝vtanh(w[s t ，h i ])#(9)
wherein a is t，j For the attention function, the score function is an alignment model, which assigns a score based on the degree of matching of the input and output at time i, defining how much weight each output gives to each input hidden state; w t(k) Representing the output of the tth word passing through the kth MBA model, wherein the value of k is 1,2;
the final output 0 of the semantic feature extraction layer is obtained by fusing the output of the multi-word segmentation module and the output of the multi-layer BILSTM, and is expressed by a mathematical expression as formula 10:
Figure FDA0003739733720000032
the final output sequence of the layer model is [ O ] 1 ，O 2 ...，O n ]；
(4) CRF layer:
the main role of this layer is to predict the label; in the process of training data, the layer automatically learns the constraint between the labels and ensures that the predicted labels are legal; the matrix P is a scoring matrix, P i，j Is a probability value, A, that classifies the ith character as the jth token i，j Is the state transition score from the ith marker to the jth marker; if the input sentence x = (x) 1 ，x 2 …，x n ) The mark sequence is y = (y) 1 ，y 2 ，...，y n ) The score is as follows:
Figure FDA0003739733720000033
for Score (x, y), the normalization process was performed using the Softmax function, as follows:
Figure FDA0003739733720000034
in training, for training sample (x, y) x ) Maximizing the log probability of the marker sequence using the following formula;
Figure FDA0003739733720000041
the experiment adopts the Viterbi algorithm to solve the probability maximum path of the dynamic programming, and the formula is as follows:
Figure FDA0003739733720000042
Y * is the sequence of the scoring function with the highest score, i.e. the expected output of the model,
Figure FDA0003739733720000047
is a maximize score function;
(5) an output layer: the layer is used for inputting labeling results of all texts in the data set; the evaluation index is measured by the accuracy P, recall R and F1 values, as shown in equations 15, 16 and 17:
Figure FDA0003739733720000043
Figure FDA0003739733720000044
Figure FDA0003739733720000045
wherein T is p Representing the number of correctly identified medical entities of the model, F p Number of unrelated medical entities identified by the representation model, F N The number of relevant medical entities which cannot be identified by the model; f1 is the weighted harmonic mean of P and R.
CN202210809038.8A 2022-07-11 2022-07-11 Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory Pending CN115238693A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210809038.8A CN115238693A (en) 2022-07-11 2022-07-11 Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210809038.8A CN115238693A (en) 2022-07-11 2022-07-11 Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory

Publications (1)

Publication Number Publication Date
CN115238693A true CN115238693A (en) 2022-10-25

Family

ID=83671477

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210809038.8A Pending CN115238693A (en) 2022-07-11 2022-07-11 Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory

Country Status (1)

Country Link
CN (1) CN115238693A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116501884A (en) * 2023-03-31 2023-07-28 重庆大学 Medical entity identification method based on BERT-BiLSTM-CRF
CN117933259A (en) * 2024-03-25 2024-04-26 成都中医药大学 Named entity recognition method based on local text information
CN118278507A (en) * 2024-06-04 2024-07-02 南京大学 Method for constructing knowledge graph of biological medicine industry

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116501884A (en) * 2023-03-31 2023-07-28 重庆大学 Medical entity identification method based on BERT-BiLSTM-CRF
CN117933259A (en) * 2024-03-25 2024-04-26 成都中医药大学 Named entity recognition method based on local text information
CN118278507A (en) * 2024-06-04 2024-07-02 南京大学 Method for constructing knowledge graph of biological medicine industry
CN118278507B (en) * 2024-06-04 2024-10-01 南京大学 Method for constructing knowledge graph of biological medicine industry

Similar Documents

Publication Publication Date Title
CN110083831B (en) Chinese named entity identification method based on BERT-BiGRU-CRF
CN112733541A (en) Named entity identification method of BERT-BiGRU-IDCNN-CRF based on attention mechanism
CN111401061A (en) Method for identifying news opinion involved in case based on BERT and Bi L STM-Attention
CN110489750A (en) Burmese participle and part-of-speech tagging method and device based on two-way LSTM-CRF
CN115238693A (en) Chinese named entity recognition method based on multi-word segmentation and multi-layer bidirectional long-short term memory
CN106980609A (en) A kind of name entity recognition method of the condition random field of word-based vector representation
CN113673254B (en) Knowledge distillation position detection method based on similarity maintenance
CN111414481A (en) Chinese semantic matching method based on pinyin and BERT embedding
CN107797987B (en) Bi-LSTM-CNN-based mixed corpus named entity identification method
CN110555084A (en) remote supervision relation classification method based on PCNN and multi-layer attention
Xing et al. A convolutional neural network for aspect-level sentiment classification
CN114153971B (en) Error correction recognition and classification equipment for Chinese text containing errors
CN115687626A (en) Legal document classification method based on prompt learning fusion key words
CN114386417A (en) Chinese nested named entity recognition method integrated with word boundary information
CN113987183A (en) Power grid fault handling plan auxiliary decision-making method based on data driving
CN115600597A (en) Named entity identification method, device and system based on attention mechanism and intra-word semantic fusion and storage medium
CN110134950B (en) Automatic text proofreading method combining words
CN113239663B (en) Multi-meaning word Chinese entity relation identification method based on Hopkinson
CN114398488A (en) Bilstm multi-label text classification method based on attention mechanism
CN115238697A (en) Judicial named entity recognition method based on natural language processing
CN116029305A (en) Chinese attribute-level emotion analysis method, system, equipment and medium based on multitask learning
CN114417851A (en) Emotion analysis method based on keyword weighted information
CN108536781B (en) Social network emotion focus mining method and system
CN115169349A (en) Chinese electronic resume named entity recognition method based on ALBERT
CN115098673A (en) Business document information extraction method based on variant attention and hierarchical structure

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