CN111178074A - Deep learning-based Chinese named entity recognition method - Google Patents

Deep learning-based Chinese named entity recognition method Download PDF

Info

Publication number
CN111178074A
CN111178074A CN201911271419.XA CN201911271419A CN111178074A CN 111178074 A CN111178074 A CN 111178074A CN 201911271419 A CN201911271419 A CN 201911271419A CN 111178074 A CN111178074 A CN 111178074A
Authority
CN
China
Prior art keywords
vector
layer
word
vectors
lstm
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201911271419.XA
Other languages
Chinese (zh)
Other versions
CN111178074B (en
Inventor
罗韬
冯爽
徐天一
赵满坤
于健
喻梅
于瑞国
李雪威
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.)
Tianjin University
Original Assignee
Tianjin 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 Tianjin University filed Critical Tianjin University
Priority to CN201911271419.XA priority Critical patent/CN111178074B/en
Publication of CN111178074A publication Critical patent/CN111178074A/en
Application granted granted Critical
Publication of CN111178074B publication Critical patent/CN111178074B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

The invention relates to a Chinese named entity recognition method based on deep learning, which is characterized by comprising the following steps: the identification method comprises the following steps: 1) embedding a word position information mixed vector into a data text; 2) inputting the vectors obtained in the step into a Bi-LSTM layer for vector coding, and simulating a long-term relation between time series capture vectors; 3) inputting the vector output by the Bi-LSTM layer into a self-attention layer, definitely learning the dependency relationship between any two characters in a sentence, and capturing the internal structure information of the sentence; 4) and inputting the output vector sequence into a CRF layer, making an independent marking decision, and decoding the label. The method is scientific and reasonable in design, capable of operating multiple data sets, high in applicability and accuracy and capable of being applied to the named entity recognition model of the multi-field text.

Description

Deep learning-based Chinese named entity recognition method
Technical Field
The invention belongs to the technical field of natural language processing, knowledge maps and sequence marking, relates to a deep learning technology and a sequence marking technology, and particularly relates to a Chinese named entity recognition method based on deep learning.
Background
Named entity recognition belongs to one of the field of sequence labeling, is a basic task of natural language processing, and mainly refers to finding out entities with specific meanings in texts, including names of people, places, names of organizations and some specific proper nouns. The identified task mainly comprises two parts: entity boundaries identify and determine their entity categories (person, place, and organizational names, etc.), where named entities serve as basic elements of text and also as basic units for understanding the content of articles. More, named entity recognition is used as an upper-layer basic task for processing text data such as a knowledge graph, wherein the accuracy of named entity recognition directly influences the final effect of knowledge graph construction. The knowledge graph is established on the relationship between the entities, and if the entity extraction is wrong, the subsequent entity relationship cannot be determined; the same applies to the automatic summarization and question-answering system, and when semantic analysis is performed on a sentence, related named entities in the text must be found. Thus, named entity recognition is extremely critical and important for text data processing, especially natural language processing.
At present, popular named entity recognition methods include CRF models, LSTM models, and models combining LSTM and CRF, which are popular named entity recognition models. Compared with an independent single model, the mixed model combining the LSTM and the CRF combines the advantages of the LSTM and the CRF, can memorize the dependency relationship between long-distance sequences and also utilizes the advantages of CRF labeling, so that the method is widely applied to the field of named entity recognition, and is optimized and improved on the basis of the method. Zhang et al studied a new dynamic meta-embedding method in 2019 and applied it to the chinese NER task. The method creates dynamic, data-specific and task-specific meta-embedding, since meta-embedding of the same character in different sentence sequences is different. Experiments on the MSRA and litteraurener data sets verified the validity of the model and the latest results were obtained on litteraurener.
Although many methods have been proposed in recent years, these methods generally do not produce good results on multiple datasets, and at the same time, there is no universal named entity recognition model which has strong applicability and high accuracy and can be applied to multiple fields.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a Chinese named entity recognition method based on deep learning, can operate multiple data sets, has strong applicability and high accuracy, and can be applied to a named entity recognition model of multi-field texts.
The technical problem to be solved by the invention is realized by the following technical scheme:
a Chinese named entity recognition method based on deep learning is characterized in that: the identification method comprises the following steps:
1) embedding a word position information mixed vector into a data text;
2) inputting the vectors obtained in the step into a Bi-LSTM layer for vector coding, and simulating a long-term relation between time series capture vectors;
3) inputting the vector output by the Bi-LSTM layer into a self-attention layer, definitely learning the dependency relationship between any two characters in a sentence, and capturing the internal structure information of the sentence;
4) and inputting the output vector sequence into a CRF layer, making an independent marking decision, and decoding the label.
Moreover, the specific operation of the step 1) is as follows:
a. establishing a dictionary according to a training data set, obtaining one-hot vectors of each word, wherein the length is the dictionary length V, and then mapping the one-hot vectors into low-dimensional dense vectors by using a pre-trained single word position vector matrix through a look-up layer;
b. carrying out vector splicing on three-part character vectors of the word vector, character vectors divided into words and word position vectors, wherein the vectors are used as the input of a network model and are used for a Chinese token sequence
X＝(x1，x2，x3，…xn，)
Checking whether a token X exists in the word lookup table and the character lookup table or not, and when the token X exists in all the two tables, namely the token consists of one character, combining vectors of two embedded items to serve as distributed representation of the token; otherwise, only the embedding in one lookup table is used as the output of the embedding layer, and the word position vector is initialized to the word vector of the word in which the word is located.
Moreover, the specific operation of the step 2) is as follows: inputting the word and word mixed vector of each word in an input sequence into a Bi-LSTM layer as each time step of the network, extracting global features, and obtaining an implicit output sequence (h) of the forward LSTM through a bidirectional LSTM network1，h2...hn) And implicit output sequence of inverse LSTM
Figure BDA0002314306680000021
Splicing the two groups of implicit sequences according to positions to obtain a complete implicit sequence
Figure BDA0002314306680000022
This implicit sequence is taken as input for the next layer.
Moreover, the specific operation of the step 3) is: for each time step input, H ═ H1，h3，...hnAnd (3) expressing the output of the B-iLSTM hidden layer, and performing linear transformation on the input vector once according to the principle of a multi-head attention mechanism, and performing scaling on the input vector to obtain a dotproduct, wherein the attention formula is as follows:
Figure DEST_PATH_3
wherein:
Figure BDA0002314306680000032
is a query matrix;
Figure BDA0002314306680000033
is a key matrix;
Figure BDA0002314306680000034
is a matrix of values;
d is the dimension of the hidden unit of Bi-LSTM, numerically equal to 2dh；
Setting Q-K-V-H, the multi-head attention first linearly projects the query, key and value H by using different linear projections, then H projects the scaled dot product attention in parallel, finally concatenates these attention results and projects again to get the new representation.
Moreover, the specific operation of the step 4) is as follows: and accessing the result to a CRF layer, wherein the CRF layer comprises a transfer matrix which represents transfer scores among all labels, and the score of each label corresponding to each word in the CRF layer consists of two parts: and adding legal constraint between predicted labels through a transition matrix in a CRF layer by the sum of the unary emission score output by the LSTM model and the binary transition score in the CRF, increasing the rationality of label grammar, and finally deducing a label sequence with the highest score by using Viterbi decoding on the prediction of the labels.
The invention has the advantages and beneficial effects that:
the Chinese named entity recognition method based on deep learning can operate multiple data sets, is high in applicability and accuracy, and can be applied to a named entity recognition model of multi-field texts.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a graph of the number of iterations of the present invention on a MSRA data set versus the change in value of model F1;
FIG. 3 is a graph of the number of iterations of the invention on a LiteratureNER dataset plotted against the change in value of model F1;
FIG. 4 is a graph of the number of iterations on the MSRA data set versus the change in model Accuracy values in accordance with the present invention;
FIG. 5 is a graph of the number of iterations of the invention on a LiteratureNER dataset versus the change in model Accuracy values.
Detailed Description
The present invention is further illustrated by the following specific examples, which are intended to be illustrative, not limiting and are not intended to limit the scope of the invention.
As shown in fig. 1, a method for identifying a named entity in chinese based on deep learning is characterized in that: the identification method comprises the following steps:
1) embedding a word position information mixed vector into a data text;
a. establishing a dictionary according to a training data set, obtaining one-hot vectors of each word, wherein the length is the dictionary length V, and then mapping the one-hot vectors into low-dimensional dense vectors by using a pre-trained single word position vector matrix through a look-up layer;
b. carrying out vector splicing on three-part character vectors of the word vector, character vectors divided into words and word position vectors, wherein the vectors are used as the input of a network model and are used for a Chinese token sequence
X＝(x1，x2，x3，…xn，)
Checking whether a token X exists in the word lookup table and the character lookup table or not, and when the token X exists in all the two tables, namely the token consists of one character, combining vectors of two embedded items to serve as distributed representation of the token; otherwise, only embedding in a lookup table is used as the output of the embedding layer, and the word position vector is initialized to the word vector of the word where the word is located;
2) inputting the vectors obtained in the step into a Bi-LSTM layer for vector coding, and simulating a long-term relation between time series capture vectors;
mixing words of individual words in an input sequenceThe vector is input to a Bi-LSTM layer as each time step of the network, global features are extracted, and an implicit output sequence (h) of the forward LSTM is obtained through a bidirectional LSTM network1，h2...hn) And implicit output sequence of inverse LSTM
Figure BDA0002314306680000041
Splicing the two groups of implicit sequences according to positions to obtain a complete implicit sequence
Figure BDA0002314306680000042
Taking the implicit sequence as the input of the next layer;
3) inputting the vector output by the Bi-LSTM layer into a self-attention layer, definitely learning the dependency relationship between any two characters in a sentence, and capturing the internal structure information of the sentence;
for each time step input, H ═ H1，h3，...hnAnd (3) expressing the output of the B-iLSTM hidden layer, and performing linear transformation on the input vector once according to the principle of a multi-head attention mechanism, and performing scaling on the input vector to obtain a dotproduct, wherein the attention formula is as follows:
Figure 389050DEST_PATH_3
wherein:
Figure BDA0002314306680000044
is a query matrix;
Figure BDA0002314306680000045
is a key matrix;
Figure BDA0002314306680000046
is a matrix of values;
d is the dimension of the hidden unit of Bi-LSTM, numerically equal to 2dh；
Setting Q-K-V-H, the multi-head attention firstly linearly projecting the query, the key and the value H by using different linear projections, then projecting H and performing scaled dot product attention in parallel, finally connecting the attention results, and projecting again to obtain a new representation;
4) inputting the output vector sequence into a CRF layer, making an independent marking decision, and performing label decoding; and accessing the result to a CRF layer, wherein the CRF layer comprises a transfer matrix which represents transfer scores among all labels, and the score of each label corresponding to each word in the CRF layer consists of two parts: and adding legal constraint between predicted labels through a transition matrix in a CRF layer by the sum of the unary emission score output by the LSTM model and the binary transition score in the CRF, increasing the rationality of label grammar, and finally deducing a label sequence with the highest score by using Viterbi decoding on the prediction of the labels.
5) Model training:
a. the network reads the training sample for training, iteration is carried out from 1, and when the maximum iteration number is larger than K, the training is stopped; and for each input training data set, calculating a loss difference value of current output according to a loss function, wherein the loss difference value is used for measuring the training degree of the model, if the loss is greater than a preset minimum loss value, the model needs to be trained and adjusted continuously, network parameters of each layer need to be updated in sequence by using a back propagation algorithm, if the loss is less than the preset minimum loss value, the model reaches a training standard, the training is finished, and the program exits.
b. After the traversal of the current batch of the training data set is completed, the model training degree is verified by using the verification set, if the current verification result is superior to the best result of the historical verification, the current training is effective, the performance of the model is in an ascending stage, the training can be tried to continue, the current data is recorded, the best result of the historical is replaced by the current verification result, and the next training is continued. If the verification result is not improved in M times of continuous training, which may indicate that the span of the learning rate is too large and may just cross the extreme value part of the minimum loss, we may consider to reduce the learning rate appropriately, try to continue training, repeat iteration until the learning rate is lower than the preset value of the system, and end the training and exit.
c. After the model training is finished, we will test the training condition of the model, and the testing process of the model is as follows:
(1) and loading the network parameters obtained through training into the model, and inputting a test data set.
(2) And the network receives the test data set and obtains final test output through a forward propagation algorithm.
(3) And comparing the output sequence of the network model with the correct labeling sequence for calculation.
(4) And finally, counting accuracy, F1 value and recall rate.
The experiments of this example were performed on microsoft's MSRA news dataset and the published litetarur ner dataset, respectively.
MSRA is from SIGHAN 2006 and is a shared task for chinese named entity identification. The data set contains 3 entity types: personnel, organization, and location. Statistics show that the data set contains 48998 sentences for training and 4432 sentences for testing, and the present example takes one tenth of the training set as the validation set because the MSRA data set lacks a validation set.
The liteatureer dataset was constructed from hundreds of chinese literature articles, excluding too short, too cluttered articles. There are 9 entity types: people, organizations, locations, abstractions, times, things, and metrics. The specific contents of the data set segmentation are as follows: 26320 training statements, 2045 verification statements, and 3016 test statements.
In this example, to prove the superiority of the data, the experimental methods of multiple periodicals are selected as baseline results for comparison, and the final performance of the model is shown in tables 1 and 2 to be compared with the performance of the comparison model.
In the present example, evaluation indexes similar to those of the conventional work were adopted, namely Precision P, Recall R and F1-score (F1). The accuracy reflects the ratio of the number of correctly predicted tokens to the number of predicted tokens; the recall rate reflects the ratio of correctly predicted tokens to all tokens in the data used; f1 is the harmonic mean of precision and recall. The following formula gives the calculation formula of three indexes:
Figure BDA0002314306680000061
Figure BDA0002314306680000062
Figure BDA0002314306680000063
wherein: TP is the number of tokens which are judged as positive examples by the model and are actually positive examples;
FP is the number of tokens which are determined to be positive examples by the model but are actually negative examples;
TN is the number of tokens that the model determines as negative but actually positive;
FN is the number of tokens that are determined by the model to be negative and actually negative.
Finally, experiments show that the new model obtains a better result on the MSRA without any manually-made characteristic template, the F1 value reaches 91.37%, and the F1 value on the LiteratureNER data set reaches 73.23%, which is better than the results of other people before, and is improved by 0.5% and 0.2% compared with the Zhang, so that the current best performance of the task is achieved, and the method has the characteristics of capability of running multiple data sets, strong applicability, high accuracy and capability of being applied to multi-field texts.
Table 1 comparison of results on dataset MSRA
Figure BDA0002314306680000064
Figure BDA0002314306680000071
Table 2 comparison of results on dataset LiteratureNER
Figure BDA0002314306680000072
In the training process of the model, a researcher can judge the training state of the model through the iteration times, the effect change curve graphs of all labels of the model and the Accuracy graph. Wherein, the evaluation Accuracy refers to the ratio of the number of correctly predicted samples in the verification set to the total number of predicted samples, and it does not consider whether the predicted samples are positive or negative, and the calculation formula of Accuracy is as follows:
Figure BDA0002314306680000073
thus, this example scales out 33 of the experimental setup for display in 100 iterations of the experimental setup and plots the F1 values for the model on both datasets separately, as shown in fig. 2-5.
It can be seen from the figure that the convergence speed of the F1 value of the model from the iteration start of the MSRA data set is relatively fast, the MSRA data set tends to be stable after about 15 iterations and keeps floating in a small range, while the setaterurener data set tends to be in a stable state at the initial training stage, which is related to the composition and the data volume of the two data sets, because the data volume of the MSRA data set is relatively large, the training is required to be longer to reach the stable state, however, the curve change diagram of the F1 value of the model seen by the two data sets well illustrates that the model can be fast converged and cannot fall into an overfitting state, and the model is well suitable for the task of identifying the named entities in chinese.
Although the embodiments of the present invention and the accompanying drawings are disclosed for illustrative purposes, those skilled in the art will appreciate that: various substitutions, changes and modifications are possible without departing from the spirit and scope of the invention and the appended claims, and therefore the scope of the invention is not limited to the disclosure of the embodiments and the accompanying drawings.

Claims (5)

1. A Chinese named entity recognition method based on deep learning is characterized in that: the identification method comprises the following steps:
1) embedding a word position information mixed vector into a data text;
2) inputting the vectors obtained in the step into a Bi-LSTM layer for vector coding, and simulating a long-term relation between time series capture vectors;
3) inputting the vector output by the Bi-LSTM layer into a self-attention layer, definitely learning the dependency relationship between any two characters in a sentence, and capturing the internal structure information of the sentence;
4) and inputting the output vector sequence into a CRF layer, making an independent marking decision, and decoding the label.
2. The method for recognizing Chinese named entities based on deep learning of claim 1, wherein: the specific operation of the step 1) is as follows:
a. establishing a dictionary according to a training data set, obtaining one-hot vectors of each word, wherein the length is the dictionary length V, and then mapping the one-hot vectors into low-dimensional dense vectors by using a pre-trained single word position vector matrix through a look-up layer;
b. carrying out vector splicing on three-part character vectors of the word vector, character vectors divided into words and word position vectors, wherein the vectors are used as the input of a network model and are used for a Chinese token sequence
X＝(x1，x2，x3，...xn，)
Checking whether a token X exists in the word lookup table and the character lookup table or not, and when the token X exists in all the two tables, namely the token consists of one character, combining vectors of two embedded items to serve as distributed representation of the token; otherwise, only the embedding in one lookup table is used as the output of the embedding layer, and the word position vector is initialized to the word vector of the word in which the word is located.
3. The method for recognizing Chinese named entities based on deep learning of claim 1, wherein: the specific operation of the step 2) is as follows: inputting the word-word mixed vector of each word in an input sequence into Bi-L as each time step of the networkSTM layer, extracting global characteristics, and obtaining hidden output sequence (h) of forward LSTM through bidirectional LSTM network1，h2...hn) And implicit output sequence of inverse LSTM
Figure FDA0002314306670000011
Splicing the two groups of implicit sequences according to positions to obtain a complete implicit sequence
Figure FDA0002314306670000012
This implicit sequence is taken as input for the next layer.
4. The method for recognizing Chinese named entities based on deep learning of claim 1, wherein: the specific operation of the step 3) is as follows:
for each time step input, H ═ H1，h3，...hnAnd (3) expressing the output of the B-iLSTM hidden layer, and performing linear transformation on the input vector once according to the principle of a multi-head attention mechanism, and performing scaling on the input vector to obtain a dotproduct, wherein the attention formula is as follows:
Figure 3
wherein:
Figure RE-FDA0002415787500000022
is a query matrix;
Figure RE-FDA0002415787500000023
is a key matrix;
Figure RE-FDA0002415787500000024
is a matrix of values;
d is the dimension of the hidden unit of Bi-LSTM, numerically equal to 2dh；
Setting Q-K-V-H, the multi-head attention first linearly projects the query, key and value H by using different linear projections, then H projects the scaled dot product attention in parallel, finally concatenates these attention results and projects again to get the new representation.
5. The method for recognizing Chinese named entities based on deep learning of claim 1, wherein: the specific operation of the step 4) is as follows:
and accessing the result to a CRF layer, wherein the CRF layer comprises a transfer matrix which represents transfer scores among all labels, and the score of each label corresponding to each word in the CRF layer consists of two parts: and adding legal constraint between predicted labels through a transition matrix in a CRF layer by the sum of the unary emission score output by the LSTM model and the binary transition score in the CRF, increasing the rationality of label grammar, and finally deducing a label sequence with the highest score by using Viterbi decoding on the prediction of the labels.
CN201911271419.XA 2019-12-12 2019-12-12 Chinese named entity recognition method based on deep learning Active CN111178074B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911271419.XA CN111178074B (en) 2019-12-12 2019-12-12 Chinese named entity recognition method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911271419.XA CN111178074B (en) 2019-12-12 2019-12-12 Chinese named entity recognition method based on deep learning

Publications (2)

Publication Number Publication Date
CN111178074A true CN111178074A (en) 2020-05-19
CN111178074B CN111178074B (en) 2023-08-25

Family

ID=70650181

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911271419.XA Active CN111178074B (en) 2019-12-12 2019-12-12 Chinese named entity recognition method based on deep learning

Country Status (1)

Country Link
CN (1) CN111178074B (en)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111666427A (en) * 2020-06-12 2020-09-15 长沙理工大学 Entity relationship joint extraction method, device, equipment and medium
CN111967265A (en) * 2020-08-31 2020-11-20 广东工业大学 Chinese word segmentation and entity identification combined learning method capable of automatically generating data set
CN112037179A (en) * 2020-08-11 2020-12-04 深圳大学 Method, system and equipment for generating brain disease diagnosis model
CN112069823A (en) * 2020-09-17 2020-12-11 华院数据技术(上海)有限公司 Information processing method and device
CN112084336A (en) * 2020-09-09 2020-12-15 浙江综合交通大数据中心有限公司 Entity extraction and event classification method and device for expressway emergency
CN112084783A (en) * 2020-09-24 2020-12-15 中国民航大学 Entity identification method and system based on civil aviation non-civilized passengers
CN112464663A (en) * 2020-12-01 2021-03-09 小牛思拓(北京)科技有限公司 Multi-feature fusion Chinese word segmentation method
CN112685549A (en) * 2021-01-08 2021-04-20 昆明理工大学 Method and system for identifying entity of affair-related news element integrated with chapter semantics
CN113033206A (en) * 2021-04-01 2021-06-25 重庆交通大学 Bridge detection field text entity identification method based on machine reading understanding
CN113255294A (en) * 2021-07-14 2021-08-13 北京邮电大学 Named entity recognition model training method, recognition method and device
CN113283243A (en) * 2021-06-09 2021-08-20 广东工业大学 Entity and relation combined extraction method
CN113449524A (en) * 2021-04-01 2021-09-28 山东英信计算机技术有限公司 Named entity identification method, system, equipment and medium
CN113486173A (en) * 2021-06-11 2021-10-08 南京邮电大学 Text labeling neural network model and labeling method thereof
CN116151241A (en) * 2023-04-19 2023-05-23 湖南马栏山视频先进技术研究院有限公司 Entity identification method and device

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107145483A (en) * 2017-04-24 2017-09-08 北京邮电大学 A kind of adaptive Chinese word cutting method based on embedded expression
CN109388807A (en) * 2018-10-30 2019-02-26 中山大学 The method, apparatus and storage medium of electronic health record name Entity recognition
CN109614614A (en) * 2018-12-03 2019-04-12 焦点科技股份有限公司 A kind of BILSTM-CRF name of product recognition methods based on from attention
CN110083710A (en) * 2019-04-30 2019-08-02 北京工业大学 It is a kind of that generation method is defined based on Recognition with Recurrent Neural Network and the word of latent variable structure
CN110334339A (en) * 2019-04-30 2019-10-15 华中科技大学 It is a kind of based on location aware from the sequence labelling model and mask method of attention mechanism
CN110377686A (en) * 2019-07-04 2019-10-25 浙江大学 A kind of address information Feature Extraction Method based on deep neural network model
CN110399492A (en) * 2019-07-22 2019-11-01 阿里巴巴集团控股有限公司 The training method and device of disaggregated model aiming at the problem that user's question sentence

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107145483A (en) * 2017-04-24 2017-09-08 北京邮电大学 A kind of adaptive Chinese word cutting method based on embedded expression
CN109388807A (en) * 2018-10-30 2019-02-26 中山大学 The method, apparatus and storage medium of electronic health record name Entity recognition
CN109614614A (en) * 2018-12-03 2019-04-12 焦点科技股份有限公司 A kind of BILSTM-CRF name of product recognition methods based on from attention
CN110083710A (en) * 2019-04-30 2019-08-02 北京工业大学 It is a kind of that generation method is defined based on Recognition with Recurrent Neural Network and the word of latent variable structure
CN110334339A (en) * 2019-04-30 2019-10-15 华中科技大学 It is a kind of based on location aware from the sequence labelling model and mask method of attention mechanism
CN110377686A (en) * 2019-07-04 2019-10-25 浙江大学 A kind of address information Feature Extraction Method based on deep neural network model
CN110399492A (en) * 2019-07-22 2019-11-01 阿里巴巴集团控股有限公司 The training method and device of disaggregated model aiming at the problem that user's question sentence

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
刘晓俊等: "基于Bi-LSTM和注意力机制的命名实体识别", 《洛阳理工学院学报》 *

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111666427B (en) * 2020-06-12 2023-05-12 长沙理工大学 Entity relationship joint extraction method, device, equipment and medium
CN111666427A (en) * 2020-06-12 2020-09-15 长沙理工大学 Entity relationship joint extraction method, device, equipment and medium
CN112037179A (en) * 2020-08-11 2020-12-04 深圳大学 Method, system and equipment for generating brain disease diagnosis model
CN111967265A (en) * 2020-08-31 2020-11-20 广东工业大学 Chinese word segmentation and entity identification combined learning method capable of automatically generating data set
CN111967265B (en) * 2020-08-31 2023-09-15 广东工业大学 Chinese word segmentation and entity recognition combined learning method for automatic generation of data set
CN112084336A (en) * 2020-09-09 2020-12-15 浙江综合交通大数据中心有限公司 Entity extraction and event classification method and device for expressway emergency
CN112069823A (en) * 2020-09-17 2020-12-11 华院数据技术(上海)有限公司 Information processing method and device
CN112084783B (en) * 2020-09-24 2022-04-12 中国民航大学 Entity identification method and system based on civil aviation non-civilized passengers
CN112084783A (en) * 2020-09-24 2020-12-15 中国民航大学 Entity identification method and system based on civil aviation non-civilized passengers
CN112464663A (en) * 2020-12-01 2021-03-09 小牛思拓(北京)科技有限公司 Multi-feature fusion Chinese word segmentation method
CN112685549B (en) * 2021-01-08 2022-07-29 昆明理工大学 Document-related news element entity identification method and system integrating discourse semantics
CN112685549A (en) * 2021-01-08 2021-04-20 昆明理工大学 Method and system for identifying entity of affair-related news element integrated with chapter semantics
CN113449524A (en) * 2021-04-01 2021-09-28 山东英信计算机技术有限公司 Named entity identification method, system, equipment and medium
CN113033206A (en) * 2021-04-01 2021-06-25 重庆交通大学 Bridge detection field text entity identification method based on machine reading understanding
CN113283243A (en) * 2021-06-09 2021-08-20 广东工业大学 Entity and relation combined extraction method
CN113283243B (en) * 2021-06-09 2022-07-26 广东工业大学 Entity and relationship combined extraction method
CN113486173A (en) * 2021-06-11 2021-10-08 南京邮电大学 Text labeling neural network model and labeling method thereof
CN113486173B (en) * 2021-06-11 2023-09-12 南京邮电大学 Text labeling neural network model and labeling method thereof
CN113255294A (en) * 2021-07-14 2021-08-13 北京邮电大学 Named entity recognition model training method, recognition method and device
CN116151241A (en) * 2023-04-19 2023-05-23 湖南马栏山视频先进技术研究院有限公司 Entity identification method and device

Also Published As

Publication number Publication date
CN111178074B (en) 2023-08-25

Similar Documents

Publication Publication Date Title
CN111178074B (en) Chinese named entity recognition method based on deep learning
CN109871535B (en) French named entity recognition method based on deep neural network
CN109960724B (en) Text summarization method based on TF-IDF
CN109344236B (en) Problem similarity calculation method based on multiple characteristics
CN107391486B (en) Method for identifying new words in field based on statistical information and sequence labels
Wang et al. Integrating extractive and abstractive models for long text summarization
US10891427B2 (en) Machine learning techniques for generating document summaries targeted to affective tone
CN111738003B (en) Named entity recognition model training method, named entity recognition method and medium
Ayana et al. Recent advances on neural headline generation
CN112035730B (en) Semantic retrieval method and device and electronic equipment
CN111241294A (en) Graph convolution network relation extraction method based on dependency analysis and key words
CN105068997B (en) The construction method and device of parallel corpora
CN110704621A (en) Text processing method and device, storage medium and electronic equipment
CN109271524B (en) Entity linking method in knowledge base question-answering system
CN110688836A (en) Automatic domain dictionary construction method based on supervised learning
CN108363691B (en) Domain term recognition system and method for power 95598 work order
CN110096572B (en) Sample generation method, device and computer readable medium
CN112101027A (en) Chinese named entity recognition method based on reading understanding
CN110569505A (en) text input method and device
CN110688450A (en) Keyword generation method based on Monte Carlo tree search, keyword generation model based on reinforcement learning and electronic equipment
CN115146629A (en) News text and comment correlation analysis method based on comparative learning
CN114818717A (en) Chinese named entity recognition method and system fusing vocabulary and syntax information
CN113723103A (en) Chinese medical named entity and part-of-speech combined learning method integrating multi-source knowledge
CN112232055A (en) Text detection and correction method based on pinyin similarity and language model
CN115906805A (en) Long text abstract generating method based on word fine granularity

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
GR01 Patent grant
GR01 Patent grant