SlideShare a Scribd company logo
1 of 18
Download to read offline
자연어처리 연구실
M2020064
조단비
Published in: Journal of KIISE, Vol. 48, No. 5, PP. 556-567, 2021.
URL: https://www.kiise.or.kr/e_journal/2021/5/JOK/pdf/11.pdf
Content
1. Introduction
2. Feature Extraction
3. Method
4. Experiments and results
5. Summary
#Kookmin_University #Natural_Language_Processing_lab. 1
Introduction
개체명 인식
> 주어진 문장 내의 고유한 의미가 있는 단어들을 미리 정의된 개체의 범주로 분류하는 문제
기존 연구
> 딥러닝 네트워크 (LSTM, CNN) 혹은 언어 모델 (BERT, ELECTRA) 이용
- 장점: 높은 성능
- 단점: 고성능 컴퓨팅 파워 필요, 느린 모델 학습 속도 (> 실용성 부족)
연구 목적
> 처리 속도와 정확률을 고려한 기계학습 방식의 CRF 기반 개체명 인식
> 자질 추출 방법론 제안
- 한국어 어휘 의미망 (의미 자질 및 구문적 자질) 활용
#Kookmin_University #Natural_Language_Processing_lab. 2
Feature Extraction
1. 의미 자질
- 한국어 어휘 의미망 (UWordMap)
- 한국어 어휘 의미망의 상위어를 이용한 자질
2. 구문적 자질
- 의존관계 자질
- 간접적 의존관계 자질
#Kookmin_University #Natural_Language_Processing_lab. 3
의미 자질
1. 한국어 어휘 의미망 (UWordMap)
: 표준국어대사전 기반으로 어휘들이 의미제약으로 상호연결된 어휘의미망 (다의어 수준)
- 명사 계층 구조 (상위어, 하위어 정보)
Ex) level of node = 5
: 계층이 5단계인 단어의 개수가 117,524개 존재
#Kookmin_University #Natural_Language_Processing_lab. 4
상위어
하위어
의미 자질
2. 한국어 어휘 의미망의 상위어를 이용한 자질
: OOV 문제를 보완하여 재현율을 높이기 위한 방법
- 1계층 상위어와 최상위어를 자질로 사용
(1) 학습 데이터를 확장하는 역할
> 만약 같은 상위어에 속할 경우, 하나의 범주로 인식
#Kookmin_University #Natural_Language_Processing_lab. 5
의미 자질
(2) 개체명 인식을 위한 키워드 역할
> 주변 단어의 상위어를 통해 키워드 파악
> 1계층 단어가 과도한 추상적 의미를 내포할 경우 (하위 단어들의 분포가 다르기 때문)
: 최상위어에 대한 하위분포에 따라 1계층 혹은 2계층 단어를 최상위어 자질로 사용
- 각 최상위어에 대한 명사의 평균 level: 𝑙𝑎𝑣𝑔 = σ𝑖=1
𝑛
𝑖 × 𝑟𝑎𝑡𝑖𝑜𝑖
- 𝑖: 현재 노드의 level, 𝑟𝑎𝑡𝑖𝑜𝑖: 현재 노드의 level에 대한 명사 분포 비율
- 생물_0101 = 6.49층 / 힘_0103 = 4.69층
- 𝑙𝑎𝑣𝑔 ≥ 6: 최상위어 자질로 2계층 단어 사용
> 상위 계층에 특정 개체명이 존재할 경우,
- 해당 개체 범주를 최상위 자질로 사용 (평균 계층과 무관)
#Kookmin_University #Natural_Language_Processing_lab. 6
Ex) ‘외국인 투수 바르가스가’
- 투수의 3계층 상위어: 사람
- ‘바르가스가’를 ‘인명’으로 예측
구문적 자질
1. 의존관계 자질
> 문장의 구조 파악을 위해 의존 관계 활용
: 의존 (dependent)-지배 (governor) 관계 표현
- 주변 정보 이외에 제한된 범위 밖의 키워드를 파악하는 데에 용이
- 말뭉치: 국립국어원의 모두의 말뭉치 (구문분석 말뭉치)
#Kookmin_University #Natural_Language_Processing_lab. 7
구문적 자질
2. 간접적 의존관계 자질
> 의존관계가 분석되어 있지 않은 경우, 명사 다음 격조사에 따라 지배소 추정
(1) 명사 + 관형격 조사(JKG)
: 해당 어절 다음의 어절
(2) 명사 + 격조사 (관형격 조사 제외)
: 해당 어절 이후에 가장 가까이에 존재하는 동사
(해당 동사 내에 관형형 전성 어미 (ETM)이 존재할 경우 이를 제외)
#Kookmin_University #Natural_Language_Processing_lab. 8
(1)
(2)
Method
> 학습 자질
- 기본 자질: 문장을 구성하는 형태소와 음절, 어절을 조합한 자질
- 형태소 어휘, 형태소 품사 자질, 어휘 형태 자질, 형태소가 속한 어절, 문장 내의 인덱스 등
#Kookmin_University #Natural_Language_Processing_lab. 9
기본 학습 자질
의미 자질
구문적 자질
Method
> 한국어 개체명 인식 시스템
#Kookmin_University #Natural_Language_Processing_lab. 10
형태소 분석 단계
자질 생성 단계
개체명 분석 단계
Experiments and Results
> 실험 설계
- 국립국어원 모두의 말뭉치 (개체명 분석 말뭉치, 150,082 문장)
- 15개의 개체명 태그 > 5개의 개체명 태그로 제한하여 실험 (PS, LC, OG, TI, DT)
- 5개의 개체명 태그가 포함되지 않은 문장은 삭제
- Utagger를 이용하여 형태소 태그 부착
- 10배수 교차검증 진행
(1) 기본 자질 성능 비교
: 기본 자질만을 학습한 CRF 개체명 모델 (Baseline)
- 성능: 10배수 교차검증 성능의 평균
- 속도: 자질 생성 시간을 제외한 개체명 태그 부착 시간
#Kookmin_University #Natural_Language_Processing_lab. 11
Experiments and Results
#Kookmin_University #Natural_Language_Processing_lab. 12
Experiments and Results
> 한국어 어휘 의미망을 활용한 자질 추출의 유의미성 검정
귀무가설: 유의미하지 않다.
대립가설: 유의미하다.
검정 통계량 𝑍 =
ෞ
𝑝1−ෞ
𝑝2
ො
𝑝(1− ො
𝑝)
1
𝑛1
+
1
𝑛2
(유의수준 1%)
- 재현율에 대한 기각역: 𝑍 = 9.047894 ≥ 𝑧0.01 = 2.33
- 정밀도에 대한 기각역: 𝑍 = 5.658423 ≥ 𝑧0.01 = 2.33
> 두 값 모두 기각역에 포함되기 때문에 “한국어 어휘 의미망을 활용하여 추출한 자질들은 유의미하다.”
#Kookmin_University #Natural_Language_Processing_lab. 13
Experiments and Results
(2) 자질 별 개체명 인식 성능 비교
- max iteration 130
#Kookmin_University #Natural_Language_Processing_lab. 14
평균과 가장 유사한fold7을 사용하여 성능 측정
Experiments and Results
(3) 개체명 사전 자질 추가에 따른 성능 비교
> 기존 연구: OOV 문제를 보완하기 위해 개체명 사전 구축
- 개체명 사전 구축: [A] 울산대 UWordMap / [B] 위키피디아 백과
- 매칭: BIEOS (Begin, Inside, End, Outside, Single) 표기법 사용
(4) 딥러닝 모델과의 성능 비교
- Stacked BiLSTM-CRF 모델 구현 및 비교
- data: fold7 / epochs: 13
#Kookmin_University #Natural_Language_Processing_lab. 15
Summary
- 기존 딥러닝 방법론들의 속도 및 학습 환경의 한계
- 사람의 지적 정보를 학습 자질로 사용한 CRF 기반의 개체명 인식 모델 제안
- 한국어 어휘 의미망 (UWordMap)을 활용하여 의미 자질 및 구문적 자질 추출
- F1-score: 90.54%
- Stacked BiLSTM-CRF와의 성능 비교 결과, 14배 빠른 학습 속도, 17개 빠른 처리 속도와 함께 성능 향상
- 실용성을 높은 개체명 인식 모델로써 제시
#Kookmin_University #Natural_Language_Processing_lab. 16
Thank You.
17
#Kookmin_University #Natural_Language_Processing_lab.

More Related Content

What's hot

Intent Classifier with Facebook fastText
Intent Classifier with Facebook fastTextIntent Classifier with Facebook fastText
Intent Classifier with Facebook fastTextBayu Aldi Yansyah
 
Warnik Chow - 2018 HCLT
Warnik Chow - 2018 HCLTWarnik Chow - 2018 HCLT
Warnik Chow - 2018 HCLTWarNik Chow
 
A Survey of Various Methods for Text Summarization
A Survey of Various Methods for Text SummarizationA Survey of Various Methods for Text Summarization
A Survey of Various Methods for Text SummarizationIJERD Editor
 
Pragmatic text mining: From literature to electronic health records
Pragmatic text mining: From literature to electronic health recordsPragmatic text mining: From literature to electronic health records
Pragmatic text mining: From literature to electronic health recordsLars Juhl Jensen
 
Dna data compression algorithms based on redundancy
Dna data compression algorithms based on redundancyDna data compression algorithms based on redundancy
Dna data compression algorithms based on redundancyijfcstjournal
 
Statistically-Enhanced New Word Identification
Statistically-Enhanced New Word IdentificationStatistically-Enhanced New Word Identification
Statistically-Enhanced New Word IdentificationAndi Wu
 
Fast and Accurate Preordering for SMT using Neural Networks
Fast and Accurate Preordering for SMT using Neural NetworksFast and Accurate Preordering for SMT using Neural Networks
Fast and Accurate Preordering for SMT using Neural NetworksSDL
 
Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...
Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...
Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...IJERA Editor
 

What's hot (12)

2224d_final
2224d_final2224d_final
2224d_final
 
Wsd final paper
Wsd final paperWsd final paper
Wsd final paper
 
Intent Classifier with Facebook fastText
Intent Classifier with Facebook fastTextIntent Classifier with Facebook fastText
Intent Classifier with Facebook fastText
 
Warnik Chow - 2018 HCLT
Warnik Chow - 2018 HCLTWarnik Chow - 2018 HCLT
Warnik Chow - 2018 HCLT
 
Word embedding
Word embedding Word embedding
Word embedding
 
Cl35491494
Cl35491494Cl35491494
Cl35491494
 
A Survey of Various Methods for Text Summarization
A Survey of Various Methods for Text SummarizationA Survey of Various Methods for Text Summarization
A Survey of Various Methods for Text Summarization
 
Pragmatic text mining: From literature to electronic health records
Pragmatic text mining: From literature to electronic health recordsPragmatic text mining: From literature to electronic health records
Pragmatic text mining: From literature to electronic health records
 
Dna data compression algorithms based on redundancy
Dna data compression algorithms based on redundancyDna data compression algorithms based on redundancy
Dna data compression algorithms based on redundancy
 
Statistically-Enhanced New Word Identification
Statistically-Enhanced New Word IdentificationStatistically-Enhanced New Word Identification
Statistically-Enhanced New Word Identification
 
Fast and Accurate Preordering for SMT using Neural Networks
Fast and Accurate Preordering for SMT using Neural NetworksFast and Accurate Preordering for SMT using Neural Networks
Fast and Accurate Preordering for SMT using Neural Networks
 
Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...
Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...
Duration for Classification and Regression Treefor Marathi Textto- Speech Syn...
 

Similar to Crf based named entity recognition using a korean lexical semantic network

Colloquium talk on modal sense classification using a convolutional neural ne...
Colloquium talk on modal sense classification using a convolutional neural ne...Colloquium talk on modal sense classification using a convolutional neural ne...
Colloquium talk on modal sense classification using a convolutional neural ne...Ana Marasović
 
DextMP: Text mining for finding moonlighting proteins
DextMP: Text mining for finding moonlighting proteinsDextMP: Text mining for finding moonlighting proteins
DextMP: Text mining for finding moonlighting proteinsPurdue University
 
Chunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali TextChunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali Textkevig
 
Chunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali TextChunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali Textkevig
 
生態文化ニッチモデリングによる分布推定
生態文化ニッチモデリングによる分布推定生態文化ニッチモデリングによる分布推定
生態文化ニッチモデリングによる分布推定Yasuhisa Kondo
 
Bio Scope
Bio ScopeBio Scope
Bio ScopeStartup
 
Learning to learn unlearned feature for segmentation
Learning to learn unlearned feature for segmentationLearning to learn unlearned feature for segmentation
Learning to learn unlearned feature for segmentationNAVER Engineering
 
Verb based manipuri sentiment analysis
Verb based manipuri sentiment analysisVerb based manipuri sentiment analysis
Verb based manipuri sentiment analysisijnlc
 
Kefed introduction 12-05-10-2224
Kefed introduction 12-05-10-2224Kefed introduction 12-05-10-2224
Kefed introduction 12-05-10-2224Gully Burns
 
Natural Language Processing Advancements By Deep Learning: A Survey
Natural Language Processing Advancements By Deep Learning: A SurveyNatural Language Processing Advancements By Deep Learning: A Survey
Natural Language Processing Advancements By Deep Learning: A SurveyRimzim Thube
 
Toward Single Neuron Gene Expression Analysis for Studying Behavior
Toward Single Neuron Gene Expression Analysis for Studying Behavior Toward Single Neuron Gene Expression Analysis for Studying Behavior
Toward Single Neuron Gene Expression Analysis for Studying Behavior Rayna Harris
 
Three's a crowd-source: Observations on Collaborative Genome Annotation
Three's a crowd-source: Observations on Collaborative Genome AnnotationThree's a crowd-source: Observations on Collaborative Genome Annotation
Three's a crowd-source: Observations on Collaborative Genome AnnotationMonica Munoz-Torres
 
Automatic Personality Prediction with Attention-based Neural Networks
Automatic Personality Prediction with Attention-based Neural NetworksAutomatic Personality Prediction with Attention-based Neural Networks
Automatic Personality Prediction with Attention-based Neural NetworksJinho Choi
 
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...kevig
 
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...ijnlc
 
Welch Wordifier Bosc2009
Welch Wordifier Bosc2009Welch Wordifier Bosc2009
Welch Wordifier Bosc2009bosc
 
Phrase Structure Identification and Classification of Sentences using Deep Le...
Phrase Structure Identification and Classification of Sentences using Deep Le...Phrase Structure Identification and Classification of Sentences using Deep Le...
Phrase Structure Identification and Classification of Sentences using Deep Le...ijtsrd
 
Chunking in manipuri using crf
Chunking in manipuri using crfChunking in manipuri using crf
Chunking in manipuri using crfijnlc
 
JBrowse and Inter-"Mine" Communication - IMDEV 2017
JBrowse and Inter-"Mine" Communication - IMDEV 2017JBrowse and Inter-"Mine" Communication - IMDEV 2017
JBrowse and Inter-"Mine" Communication - IMDEV 2017Vivek Krishnakumar
 
Deep learning for extracting protein-protein interactions from biomedical lit...
Deep learning for extracting protein-protein interactions from biomedical lit...Deep learning for extracting protein-protein interactions from biomedical lit...
Deep learning for extracting protein-protein interactions from biomedical lit...Yifan Peng
 

Similar to Crf based named entity recognition using a korean lexical semantic network (20)

Colloquium talk on modal sense classification using a convolutional neural ne...
Colloquium talk on modal sense classification using a convolutional neural ne...Colloquium talk on modal sense classification using a convolutional neural ne...
Colloquium talk on modal sense classification using a convolutional neural ne...
 
DextMP: Text mining for finding moonlighting proteins
DextMP: Text mining for finding moonlighting proteinsDextMP: Text mining for finding moonlighting proteins
DextMP: Text mining for finding moonlighting proteins
 
Chunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali TextChunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali Text
 
Chunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali TextChunker Based Sentiment Analysis and Tense Classification for Nepali Text
Chunker Based Sentiment Analysis and Tense Classification for Nepali Text
 
生態文化ニッチモデリングによる分布推定
生態文化ニッチモデリングによる分布推定生態文化ニッチモデリングによる分布推定
生態文化ニッチモデリングによる分布推定
 
Bio Scope
Bio ScopeBio Scope
Bio Scope
 
Learning to learn unlearned feature for segmentation
Learning to learn unlearned feature for segmentationLearning to learn unlearned feature for segmentation
Learning to learn unlearned feature for segmentation
 
Verb based manipuri sentiment analysis
Verb based manipuri sentiment analysisVerb based manipuri sentiment analysis
Verb based manipuri sentiment analysis
 
Kefed introduction 12-05-10-2224
Kefed introduction 12-05-10-2224Kefed introduction 12-05-10-2224
Kefed introduction 12-05-10-2224
 
Natural Language Processing Advancements By Deep Learning: A Survey
Natural Language Processing Advancements By Deep Learning: A SurveyNatural Language Processing Advancements By Deep Learning: A Survey
Natural Language Processing Advancements By Deep Learning: A Survey
 
Toward Single Neuron Gene Expression Analysis for Studying Behavior
Toward Single Neuron Gene Expression Analysis for Studying Behavior Toward Single Neuron Gene Expression Analysis for Studying Behavior
Toward Single Neuron Gene Expression Analysis for Studying Behavior
 
Three's a crowd-source: Observations on Collaborative Genome Annotation
Three's a crowd-source: Observations on Collaborative Genome AnnotationThree's a crowd-source: Observations on Collaborative Genome Annotation
Three's a crowd-source: Observations on Collaborative Genome Annotation
 
Automatic Personality Prediction with Attention-based Neural Networks
Automatic Personality Prediction with Attention-based Neural NetworksAutomatic Personality Prediction with Attention-based Neural Networks
Automatic Personality Prediction with Attention-based Neural Networks
 
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
 
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
BIDIRECTIONAL LONG SHORT-TERM MEMORY (BILSTM)WITH CONDITIONAL RANDOM FIELDS (...
 
Welch Wordifier Bosc2009
Welch Wordifier Bosc2009Welch Wordifier Bosc2009
Welch Wordifier Bosc2009
 
Phrase Structure Identification and Classification of Sentences using Deep Le...
Phrase Structure Identification and Classification of Sentences using Deep Le...Phrase Structure Identification and Classification of Sentences using Deep Le...
Phrase Structure Identification and Classification of Sentences using Deep Le...
 
Chunking in manipuri using crf
Chunking in manipuri using crfChunking in manipuri using crf
Chunking in manipuri using crf
 
JBrowse and Inter-"Mine" Communication - IMDEV 2017
JBrowse and Inter-"Mine" Communication - IMDEV 2017JBrowse and Inter-"Mine" Communication - IMDEV 2017
JBrowse and Inter-"Mine" Communication - IMDEV 2017
 
Deep learning for extracting protein-protein interactions from biomedical lit...
Deep learning for extracting protein-protein interactions from biomedical lit...Deep learning for extracting protein-protein interactions from biomedical lit...
Deep learning for extracting protein-protein interactions from biomedical lit...
 

More from Danbi Cho

Attention boosted deep networks for video classification
Attention boosted deep networks for video classificationAttention boosted deep networks for video classification
Attention boosted deep networks for video classificationDanbi Cho
 
A survey on deep learning based approaches for action and gesture recognition...
A survey on deep learning based approaches for action and gesture recognition...A survey on deep learning based approaches for action and gesture recognition...
A survey on deep learning based approaches for action and gesture recognition...Danbi Cho
 
ELECTRA_Pretraining Text Encoders as Discriminators rather than Generators
ELECTRA_Pretraining Text Encoders as Discriminators rather than GeneratorsELECTRA_Pretraining Text Encoders as Discriminators rather than Generators
ELECTRA_Pretraining Text Encoders as Discriminators rather than GeneratorsDanbi Cho
 
A survey on automatic detection of hate speech in text
A survey on automatic detection of hate speech in textA survey on automatic detection of hate speech in text
A survey on automatic detection of hate speech in textDanbi Cho
 
Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...
Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...
Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...Danbi Cho
 
Decision tree and ensemble
Decision tree and ensembleDecision tree and ensemble
Decision tree and ensembleDanbi Cho
 
Can recurrent neural networks warp time
Can recurrent neural networks warp timeCan recurrent neural networks warp time
Can recurrent neural networks warp timeDanbi Cho
 
Man is to computer programmer as woman is to homemaker debiasing word embeddings
Man is to computer programmer as woman is to homemaker debiasing word embeddingsMan is to computer programmer as woman is to homemaker debiasing word embeddings
Man is to computer programmer as woman is to homemaker debiasing word embeddingsDanbi Cho
 
Situation recognition visual semantic role labeling for image understanding
Situation recognition visual semantic role labeling for image understandingSituation recognition visual semantic role labeling for image understanding
Situation recognition visual semantic role labeling for image understandingDanbi Cho
 
Mitigating unwanted biases with adversarial learning
Mitigating unwanted biases with adversarial learningMitigating unwanted biases with adversarial learning
Mitigating unwanted biases with adversarial learningDanbi Cho
 

More from Danbi Cho (11)

Gpt models
Gpt modelsGpt models
Gpt models
 
Attention boosted deep networks for video classification
Attention boosted deep networks for video classificationAttention boosted deep networks for video classification
Attention boosted deep networks for video classification
 
A survey on deep learning based approaches for action and gesture recognition...
A survey on deep learning based approaches for action and gesture recognition...A survey on deep learning based approaches for action and gesture recognition...
A survey on deep learning based approaches for action and gesture recognition...
 
ELECTRA_Pretraining Text Encoders as Discriminators rather than Generators
ELECTRA_Pretraining Text Encoders as Discriminators rather than GeneratorsELECTRA_Pretraining Text Encoders as Discriminators rather than Generators
ELECTRA_Pretraining Text Encoders as Discriminators rather than Generators
 
A survey on automatic detection of hate speech in text
A survey on automatic detection of hate speech in textA survey on automatic detection of hate speech in text
A survey on automatic detection of hate speech in text
 
Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...
Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...
Zero wall detecting zero-day web attacks through encoder-decoder recurrent ne...
 
Decision tree and ensemble
Decision tree and ensembleDecision tree and ensemble
Decision tree and ensemble
 
Can recurrent neural networks warp time
Can recurrent neural networks warp timeCan recurrent neural networks warp time
Can recurrent neural networks warp time
 
Man is to computer programmer as woman is to homemaker debiasing word embeddings
Man is to computer programmer as woman is to homemaker debiasing word embeddingsMan is to computer programmer as woman is to homemaker debiasing word embeddings
Man is to computer programmer as woman is to homemaker debiasing word embeddings
 
Situation recognition visual semantic role labeling for image understanding
Situation recognition visual semantic role labeling for image understandingSituation recognition visual semantic role labeling for image understanding
Situation recognition visual semantic role labeling for image understanding
 
Mitigating unwanted biases with adversarial learning
Mitigating unwanted biases with adversarial learningMitigating unwanted biases with adversarial learning
Mitigating unwanted biases with adversarial learning
 

Recently uploaded

Driving Innovation: Scania's API Revolution with WSO2
Driving Innovation: Scania's API Revolution with WSO2Driving Innovation: Scania's API Revolution with WSO2
Driving Innovation: Scania's API Revolution with WSO2WSO2
 
WSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital Transformation
WSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital TransformationWSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital Transformation
WSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital TransformationWSO2
 
WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...
WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...
WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...WSO2
 
WSO2CON 2024 Slides - Unlocking Value with AI
WSO2CON 2024 Slides - Unlocking Value with AIWSO2CON 2024 Slides - Unlocking Value with AI
WSO2CON 2024 Slides - Unlocking Value with AIWSO2
 
AzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdf
AzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdfAzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdf
AzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdfryanfarris8
 
OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...
OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...
OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...Shane Coughlan
 
WSO2CON 2024 - How CSI Piemonte Is Apifying the Public Administration
WSO2CON 2024 - How CSI Piemonte Is Apifying the Public AdministrationWSO2CON 2024 - How CSI Piemonte Is Apifying the Public Administration
WSO2CON 2024 - How CSI Piemonte Is Apifying the Public AdministrationWSO2
 
WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...
WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...
WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...WSO2
 
WSO2CON 2024 - Designing Event-Driven Enterprises: Stories of Transformation
WSO2CON 2024 - Designing Event-Driven Enterprises: Stories of TransformationWSO2CON 2024 - Designing Event-Driven Enterprises: Stories of Transformation
WSO2CON 2024 - Designing Event-Driven Enterprises: Stories of TransformationWSO2
 
WSO2Con2024 - Hello Choreo Presentation - Kanchana
WSO2Con2024 - Hello Choreo Presentation - KanchanaWSO2Con2024 - Hello Choreo Presentation - Kanchana
WSO2Con2024 - Hello Choreo Presentation - KanchanaWSO2
 
WSO2CON 2024 - Software Engineering for Digital Businesses
WSO2CON 2024 - Software Engineering for Digital BusinessesWSO2CON 2024 - Software Engineering for Digital Businesses
WSO2CON 2024 - Software Engineering for Digital BusinessesWSO2
 
WSO2CON 2024 - How to Run a Security Program
WSO2CON 2024 - How to Run a Security ProgramWSO2CON 2024 - How to Run a Security Program
WSO2CON 2024 - How to Run a Security ProgramWSO2
 
Evolving Data Governance for the Real-time Streaming and AI Era
Evolving Data Governance for the Real-time Streaming and AI EraEvolving Data Governance for the Real-time Streaming and AI Era
Evolving Data Governance for the Real-time Streaming and AI Eraconfluent
 
WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...
WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...
WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...WSO2
 
WSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & Innovation
WSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & InnovationWSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & Innovation
WSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & InnovationWSO2
 
WSO2CON 2024 - Building a Digital Government in Uganda
WSO2CON 2024 - Building a Digital Government in UgandaWSO2CON 2024 - Building a Digital Government in Uganda
WSO2CON 2024 - Building a Digital Government in UgandaWSO2
 
Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024
Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024
Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024VictoriaMetrics
 
WSO2Con2024 - Unleashing the Financial Potential of 13 Million People
WSO2Con2024 - Unleashing the Financial Potential of 13 Million PeopleWSO2Con2024 - Unleashing the Financial Potential of 13 Million People
WSO2Con2024 - Unleashing the Financial Potential of 13 Million PeopleWSO2
 
What Goes Wrong with Language Definitions and How to Improve the Situation
What Goes Wrong with Language Definitions and How to Improve the SituationWhat Goes Wrong with Language Definitions and How to Improve the Situation
What Goes Wrong with Language Definitions and How to Improve the SituationJuha-Pekka Tolvanen
 
WSO2CON 2024 - Architecting AI in the Enterprise: APIs and Applications
WSO2CON 2024 - Architecting AI in the Enterprise: APIs and ApplicationsWSO2CON 2024 - Architecting AI in the Enterprise: APIs and Applications
WSO2CON 2024 - Architecting AI in the Enterprise: APIs and ApplicationsWSO2
 

Recently uploaded (20)

Driving Innovation: Scania's API Revolution with WSO2
Driving Innovation: Scania's API Revolution with WSO2Driving Innovation: Scania's API Revolution with WSO2
Driving Innovation: Scania's API Revolution with WSO2
 
WSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital Transformation
WSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital TransformationWSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital Transformation
WSO2Con2024 - WSO2's IAM Vision: Identity-Led Digital Transformation
 
WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...
WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...
WSO2CON 2024 - Building the API First Enterprise – Running an API Program, fr...
 
WSO2CON 2024 Slides - Unlocking Value with AI
WSO2CON 2024 Slides - Unlocking Value with AIWSO2CON 2024 Slides - Unlocking Value with AI
WSO2CON 2024 Slides - Unlocking Value with AI
 
AzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdf
AzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdfAzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdf
AzureNativeQumulo_HPC_Cloud_Native_Benchmarks.pdf
 
OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...
OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...
OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...
 
WSO2CON 2024 - How CSI Piemonte Is Apifying the Public Administration
WSO2CON 2024 - How CSI Piemonte Is Apifying the Public AdministrationWSO2CON 2024 - How CSI Piemonte Is Apifying the Public Administration
WSO2CON 2024 - How CSI Piemonte Is Apifying the Public Administration
 
WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...
WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...
WSO2CON 2024 - Navigating API Complexity: REST, GraphQL, gRPC, Websocket, Web...
 
WSO2CON 2024 - Designing Event-Driven Enterprises: Stories of Transformation
WSO2CON 2024 - Designing Event-Driven Enterprises: Stories of TransformationWSO2CON 2024 - Designing Event-Driven Enterprises: Stories of Transformation
WSO2CON 2024 - Designing Event-Driven Enterprises: Stories of Transformation
 
WSO2Con2024 - Hello Choreo Presentation - Kanchana
WSO2Con2024 - Hello Choreo Presentation - KanchanaWSO2Con2024 - Hello Choreo Presentation - Kanchana
WSO2Con2024 - Hello Choreo Presentation - Kanchana
 
WSO2CON 2024 - Software Engineering for Digital Businesses
WSO2CON 2024 - Software Engineering for Digital BusinessesWSO2CON 2024 - Software Engineering for Digital Businesses
WSO2CON 2024 - Software Engineering for Digital Businesses
 
WSO2CON 2024 - How to Run a Security Program
WSO2CON 2024 - How to Run a Security ProgramWSO2CON 2024 - How to Run a Security Program
WSO2CON 2024 - How to Run a Security Program
 
Evolving Data Governance for the Real-time Streaming and AI Era
Evolving Data Governance for the Real-time Streaming and AI EraEvolving Data Governance for the Real-time Streaming and AI Era
Evolving Data Governance for the Real-time Streaming and AI Era
 
WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...
WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...
WSO2CON 2024 - Cloud Native Middleware: Domain-Driven Design, Cell-Based Arch...
 
WSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & Innovation
WSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & InnovationWSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & Innovation
WSO2CON 2024 - OSU & WSO2: A Decade Journey in Integration & Innovation
 
WSO2CON 2024 - Building a Digital Government in Uganda
WSO2CON 2024 - Building a Digital Government in UgandaWSO2CON 2024 - Building a Digital Government in Uganda
WSO2CON 2024 - Building a Digital Government in Uganda
 
Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024
Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024
Large-scale Logging Made Easy: Meetup at Deutsche Bank 2024
 
WSO2Con2024 - Unleashing the Financial Potential of 13 Million People
WSO2Con2024 - Unleashing the Financial Potential of 13 Million PeopleWSO2Con2024 - Unleashing the Financial Potential of 13 Million People
WSO2Con2024 - Unleashing the Financial Potential of 13 Million People
 
What Goes Wrong with Language Definitions and How to Improve the Situation
What Goes Wrong with Language Definitions and How to Improve the SituationWhat Goes Wrong with Language Definitions and How to Improve the Situation
What Goes Wrong with Language Definitions and How to Improve the Situation
 
WSO2CON 2024 - Architecting AI in the Enterprise: APIs and Applications
WSO2CON 2024 - Architecting AI in the Enterprise: APIs and ApplicationsWSO2CON 2024 - Architecting AI in the Enterprise: APIs and Applications
WSO2CON 2024 - Architecting AI in the Enterprise: APIs and Applications
 

Crf based named entity recognition using a korean lexical semantic network

  • 1. 자연어처리 연구실 M2020064 조단비 Published in: Journal of KIISE, Vol. 48, No. 5, PP. 556-567, 2021. URL: https://www.kiise.or.kr/e_journal/2021/5/JOK/pdf/11.pdf
  • 2. Content 1. Introduction 2. Feature Extraction 3. Method 4. Experiments and results 5. Summary #Kookmin_University #Natural_Language_Processing_lab. 1
  • 3. Introduction 개체명 인식 > 주어진 문장 내의 고유한 의미가 있는 단어들을 미리 정의된 개체의 범주로 분류하는 문제 기존 연구 > 딥러닝 네트워크 (LSTM, CNN) 혹은 언어 모델 (BERT, ELECTRA) 이용 - 장점: 높은 성능 - 단점: 고성능 컴퓨팅 파워 필요, 느린 모델 학습 속도 (> 실용성 부족) 연구 목적 > 처리 속도와 정확률을 고려한 기계학습 방식의 CRF 기반 개체명 인식 > 자질 추출 방법론 제안 - 한국어 어휘 의미망 (의미 자질 및 구문적 자질) 활용 #Kookmin_University #Natural_Language_Processing_lab. 2
  • 4. Feature Extraction 1. 의미 자질 - 한국어 어휘 의미망 (UWordMap) - 한국어 어휘 의미망의 상위어를 이용한 자질 2. 구문적 자질 - 의존관계 자질 - 간접적 의존관계 자질 #Kookmin_University #Natural_Language_Processing_lab. 3
  • 5. 의미 자질 1. 한국어 어휘 의미망 (UWordMap) : 표준국어대사전 기반으로 어휘들이 의미제약으로 상호연결된 어휘의미망 (다의어 수준) - 명사 계층 구조 (상위어, 하위어 정보) Ex) level of node = 5 : 계층이 5단계인 단어의 개수가 117,524개 존재 #Kookmin_University #Natural_Language_Processing_lab. 4 상위어 하위어
  • 6. 의미 자질 2. 한국어 어휘 의미망의 상위어를 이용한 자질 : OOV 문제를 보완하여 재현율을 높이기 위한 방법 - 1계층 상위어와 최상위어를 자질로 사용 (1) 학습 데이터를 확장하는 역할 > 만약 같은 상위어에 속할 경우, 하나의 범주로 인식 #Kookmin_University #Natural_Language_Processing_lab. 5
  • 7. 의미 자질 (2) 개체명 인식을 위한 키워드 역할 > 주변 단어의 상위어를 통해 키워드 파악 > 1계층 단어가 과도한 추상적 의미를 내포할 경우 (하위 단어들의 분포가 다르기 때문) : 최상위어에 대한 하위분포에 따라 1계층 혹은 2계층 단어를 최상위어 자질로 사용 - 각 최상위어에 대한 명사의 평균 level: 𝑙𝑎𝑣𝑔 = σ𝑖=1 𝑛 𝑖 × 𝑟𝑎𝑡𝑖𝑜𝑖 - 𝑖: 현재 노드의 level, 𝑟𝑎𝑡𝑖𝑜𝑖: 현재 노드의 level에 대한 명사 분포 비율 - 생물_0101 = 6.49층 / 힘_0103 = 4.69층 - 𝑙𝑎𝑣𝑔 ≥ 6: 최상위어 자질로 2계층 단어 사용 > 상위 계층에 특정 개체명이 존재할 경우, - 해당 개체 범주를 최상위 자질로 사용 (평균 계층과 무관) #Kookmin_University #Natural_Language_Processing_lab. 6 Ex) ‘외국인 투수 바르가스가’ - 투수의 3계층 상위어: 사람 - ‘바르가스가’를 ‘인명’으로 예측
  • 8. 구문적 자질 1. 의존관계 자질 > 문장의 구조 파악을 위해 의존 관계 활용 : 의존 (dependent)-지배 (governor) 관계 표현 - 주변 정보 이외에 제한된 범위 밖의 키워드를 파악하는 데에 용이 - 말뭉치: 국립국어원의 모두의 말뭉치 (구문분석 말뭉치) #Kookmin_University #Natural_Language_Processing_lab. 7
  • 9. 구문적 자질 2. 간접적 의존관계 자질 > 의존관계가 분석되어 있지 않은 경우, 명사 다음 격조사에 따라 지배소 추정 (1) 명사 + 관형격 조사(JKG) : 해당 어절 다음의 어절 (2) 명사 + 격조사 (관형격 조사 제외) : 해당 어절 이후에 가장 가까이에 존재하는 동사 (해당 동사 내에 관형형 전성 어미 (ETM)이 존재할 경우 이를 제외) #Kookmin_University #Natural_Language_Processing_lab. 8 (1) (2)
  • 10. Method > 학습 자질 - 기본 자질: 문장을 구성하는 형태소와 음절, 어절을 조합한 자질 - 형태소 어휘, 형태소 품사 자질, 어휘 형태 자질, 형태소가 속한 어절, 문장 내의 인덱스 등 #Kookmin_University #Natural_Language_Processing_lab. 9 기본 학습 자질 의미 자질 구문적 자질
  • 11. Method > 한국어 개체명 인식 시스템 #Kookmin_University #Natural_Language_Processing_lab. 10 형태소 분석 단계 자질 생성 단계 개체명 분석 단계
  • 12. Experiments and Results > 실험 설계 - 국립국어원 모두의 말뭉치 (개체명 분석 말뭉치, 150,082 문장) - 15개의 개체명 태그 > 5개의 개체명 태그로 제한하여 실험 (PS, LC, OG, TI, DT) - 5개의 개체명 태그가 포함되지 않은 문장은 삭제 - Utagger를 이용하여 형태소 태그 부착 - 10배수 교차검증 진행 (1) 기본 자질 성능 비교 : 기본 자질만을 학습한 CRF 개체명 모델 (Baseline) - 성능: 10배수 교차검증 성능의 평균 - 속도: 자질 생성 시간을 제외한 개체명 태그 부착 시간 #Kookmin_University #Natural_Language_Processing_lab. 11
  • 13. Experiments and Results #Kookmin_University #Natural_Language_Processing_lab. 12
  • 14. Experiments and Results > 한국어 어휘 의미망을 활용한 자질 추출의 유의미성 검정 귀무가설: 유의미하지 않다. 대립가설: 유의미하다. 검정 통계량 𝑍 = ෞ 𝑝1−ෞ 𝑝2 ො 𝑝(1− ො 𝑝) 1 𝑛1 + 1 𝑛2 (유의수준 1%) - 재현율에 대한 기각역: 𝑍 = 9.047894 ≥ 𝑧0.01 = 2.33 - 정밀도에 대한 기각역: 𝑍 = 5.658423 ≥ 𝑧0.01 = 2.33 > 두 값 모두 기각역에 포함되기 때문에 “한국어 어휘 의미망을 활용하여 추출한 자질들은 유의미하다.” #Kookmin_University #Natural_Language_Processing_lab. 13
  • 15. Experiments and Results (2) 자질 별 개체명 인식 성능 비교 - max iteration 130 #Kookmin_University #Natural_Language_Processing_lab. 14 평균과 가장 유사한fold7을 사용하여 성능 측정
  • 16. Experiments and Results (3) 개체명 사전 자질 추가에 따른 성능 비교 > 기존 연구: OOV 문제를 보완하기 위해 개체명 사전 구축 - 개체명 사전 구축: [A] 울산대 UWordMap / [B] 위키피디아 백과 - 매칭: BIEOS (Begin, Inside, End, Outside, Single) 표기법 사용 (4) 딥러닝 모델과의 성능 비교 - Stacked BiLSTM-CRF 모델 구현 및 비교 - data: fold7 / epochs: 13 #Kookmin_University #Natural_Language_Processing_lab. 15
  • 17. Summary - 기존 딥러닝 방법론들의 속도 및 학습 환경의 한계 - 사람의 지적 정보를 학습 자질로 사용한 CRF 기반의 개체명 인식 모델 제안 - 한국어 어휘 의미망 (UWordMap)을 활용하여 의미 자질 및 구문적 자질 추출 - F1-score: 90.54% - Stacked BiLSTM-CRF와의 성능 비교 결과, 14배 빠른 학습 속도, 17개 빠른 처리 속도와 함께 성능 향상 - 실용성을 높은 개체명 인식 모델로써 제시 #Kookmin_University #Natural_Language_Processing_lab. 16