HOME > 상세정보

상세정보

Contextualized representation learning for named entity recognition and event scheduling

Contextualized representation learning for named entity recognition and event scheduling

자료유형
학위논문
개인저자
김동현, 金東玄
서명 / 저자사항
Contextualized representation learning for named entity recognition and event scheduling / Donghyeon Kim
발행사항
Seoul :   Graduate School, Korea University,   2020  
형태사항
v, 66장 : 도표 ; 26 cm
기타형태 저록
Contextualized representation learning for named entity recognition and event scheduling   (DCOLL211009)000000127336  
학위논문주기
학위논문(박사)-- 고려대학교 대학원: 컴퓨터·전파통신공학과, 2020. 2
학과코드
0510   6YD36   375  
일반주기
지도교수: 강재우  
서지주기
참고문헌: 장 54-66
이용가능한 다른형태자료
PDF 파일로도 이용가능;   Requires PDF file reader(application/pdf)  
비통제주제어
named entity recognition , event scheduling , language representation model , contextual representation learning,,
000 00000nam c2200205 c 4500
001 000046025837
005 20200428132138
007 ta
008 191226s2020 ulkd bmAC 000c eng
040 ▼a 211009 ▼c 211009 ▼d 211009
085 0 ▼a 0510 ▼2 KDCP
090 ▼a 0510 ▼b 6YD36 ▼c 375
100 1 ▼a 김동현, ▼g 金東玄
245 1 0 ▼a Contextualized representation learning for named entity recognition and event scheduling / ▼d Donghyeon Kim
260 ▼a Seoul : ▼b Graduate School, Korea University, ▼c 2020
300 ▼a v, 66장 : ▼b 도표 ; ▼c 26 cm
500 ▼a 지도교수: 강재우
502 1 ▼a 학위논문(박사)-- ▼b 고려대학교 대학원: ▼c 컴퓨터·전파통신공학과, ▼d 2020. 2
504 ▼a 참고문헌: 장 54-66
530 ▼a PDF 파일로도 이용가능; ▼c Requires PDF file reader(application/pdf)
653 ▼a named entity recognition ▼a event scheduling ▼a language representation model ▼a contextual representation learning
776 0 ▼t Contextualized representation learning for named entity recognition and event scheduling ▼w (DCOLL211009)000000127336
900 1 0 ▼a Kim, Dong-hyeon, ▼e
900 1 0 ▼a 강재우, ▼g 姜在雨, ▼e 지도교수
945 ▼a KLPA

전자정보

No. 원문명 서비스
1
Contextualized representation learning for named entity recognition and event scheduling (35회 열람)
PDF 초록 목차

소장정보

No. 소장처 청구기호 등록번호 도서상태 반납예정일 예약 서비스
No. 1 소장처 과학도서관/학위논문서고/ 청구기호 0510 6YD36 375 등록번호 123063719 도서상태 대출가능 반납예정일 예약 서비스 B M
No. 2 소장처 과학도서관/학위논문서고/ 청구기호 0510 6YD36 375 등록번호 123063720 도서상태 대출가능 반납예정일 예약 서비스 B M

컨텐츠정보

초록

In a classification task, considering a context around a target helps to improve inference by providing additional clues to a model. For example, it is possible to improve the word-level classification accuracy by obtaining a contextualized word representation considering words before and after each word in a sentence. Bidirectional models such as Bidirectional Long Short-Term Memory (BiLSTM) networks, Embeddings from Language Models (ELMo), and Bidirectional Encoder Representations from Transformers (BERT) have shown their effectiveness in obtaining contextualized representations for words, and recently, transfer learning using a model pre-trained on a large dataset has become the mainstream. 
In this paper, we use and propose models which learn contextualized representations for named entity recognition (NER) and event scheduling. For NER, we propose a neural biomedical NER and multi-type normalization tool called Biomedical Entity Recognition and Normalization (BERN) which uses BioBERT pre-trained on 19M PubMed abstracts and 2M PubMed Central full-texts. Using deep contextualized representations, BioBERT NER models outperform existing NER models in F1 and even discover new entities that are not in entity dictionaries, using their language model and WordPiece embeddings. Also, we developed probability-based decision rules to identify the entity types of overlapping entities in multi-type NER results. The BERN provides a Web service for tagging entities in PubMed articles or raw text. For event scheduling, we propose Neural Event Scheduling Assistant (NESA) which learns user preferences and understands calendar contexts, directly from raw online calendars for fully automated and highly effective event scheduling. We leverage over 593K calendar events for NESA to learn scheduling personal events, and we further utilize NESA for multi-attendee event scheduling. NESA successfully incorporates deep neural networks for event scheduling based on natural languages, users, durations, and pre-registered events. The experimental results show that NESA significantly outperforms previous baseline models on both personal and multi-attendee event scheduling tasks. 

목차

1   Introduction 1
 1.1    Named entity recognition   2
 1.2    Event scheduling    5
2   Contextualized word representations for named entity recognition 9
 2.1    Background  9
  2.1.1 Named entity recognition for biomedical text mining   9
  2.1.2 Resolving overlapping entities 10
  2.1.3 Named entity normalization models for biomedical text mining  10
 2.2    Methods   11
  2.2.1 BioBERT for named entity recognition   12
  2.2.2 Decision rules for overlapping entities    14
  2.2.3 The multi-type normalization model  17
 2.3    Implementation   19
  2.3.1 Demonstrations  21
  2.3.2 Application programming interfaces   22
 2.4    Discussion    23
  2.4.1 Use cases    23
  2.4.2 Advantages and limitations of having a separate NER model for each entity type  25
  2.4.3 Additional dependencies of BERN    25
  2.4.4 Overcoming the network distance between a server and a client  26
3   Contextualized event representation learning for scheduling 27
 3.1    Background  27
  3.1.1 Preference learning for event scheduling 27
  3.1.2 Multi-attendee event scheduling  28
  3.1.3 Representation learning using deep neural networks  29
 3.2    Problem formulation   29
  3.2.1 Attributes of calendar data   29
  3.2.2 Personal event scheduling   30
  3.2.3 Multi-attendee event scheduling  31
 3.3    Methodology   31
  3.3.1 Title layer   32
  3.3.2 Intention layer    34
  3.3.3 Context layer   35
  3.3.4 Output layer  36
 3.4    Experiment   37
  3.4.1 Dataset    37
  3.4.2 Experimental settings   39
  3.4.3 Quantitative analysis  42
  3.4.4 Qualitative analysis   46
4   Conclusion 52
Bibliography 54