• Proceedings of the Korean Society for Bioinformatics Conference
• /
• 2003.10a
• /
• pp.3-3
• /
• 2003

With biomedical literature expanding so rapidly, there is an urgent need to discover and organize knowledge extracted from texts. Although factual databases contain crucial information the overwhelming amount of new knowledge remains in textual form (e.g. MEDLINE). In addition, new terms are constantly coined as the relationships linking new genes, drugs, proteins etc. As the size of biomedical literature is expanding, more systems are applying a variety of methods to automate the process of knowledge acquisition and management. In my talk, I focus on the project, GENIA, of our group at the University of Tokyo, the objective of which is to construct an information extraction system of protein - protein interaction from abstracts of MEDLINE. The talk includes (1) Techniques we use fDr named entity recognition (1-a) SOHMM (Self-organized HMM) (1-b) Maximum Entropy Model (1-c) Lexicon-based Recognizer (2) Treatment of term variants and acronym finders (3) Event extraction using a full parser (4) Linguistic resources for text mining (GENIA corpus) (4-a) Semantic Tags (4-b) Structural Annotations (4-c) Co-reference tags (4-d) GENIA ontology I will also talk about possible extension of our work that links the findings of molecular biology with clinical findings, and claim that textual based or conceptual based biology would be a viable alternative to system biology that tends to emphasize the role of simulation models in bioinformatics.

• Annual Conference on Human and Language Technology
• /
• 2017.10a
• /
• pp.333-337
• /
• 2017

본 논문에서는 개체명 인식과 언어 모델의 다중 학습을 이용한 한국어 개체명 인식 방법을 제안한다. 다중 학습은 1 개의 모델에서 2 개 이상의 작업을 동시에 분석하여 성능 향상을 기대할 수 있는 방법이지만, 이를 적용하기 위해서 말뭉치에 각 작업에 해당하는 태그가 부착되어야 하는 문제가 있다. 본 논문에서는 추가적인 태그 부착 없이 정보를 획득할 수 있는 언어 모델을 개체명 인식 작업과 결합하여 성능 향상을 이루고자 한다. 또한 단순한 형태소 입력의 한계를 극복하기 위해 입력 표상을 자소 및 형태소 품사의 임베딩으로 확장하였다. 기계 학습 방법은 순차적 레이블링에서 높은 성능을 제공하는 Bi-directional LSTM CRF 모델을 사용하였고, 실험 결과 언어 모델이 개체명 인식의 오류를 효과적으로 개선함을 확인하였다.

• 한국어정보학회:학술대회논문집
• /
• 2017.10a
• /
• pp.143-146
• /
• 2017

심층학습 모델 중 LSTM-CRF는 개체명 인식, 품사 태깅과 같은 sequence labeling에서 우수한 성능을 보이고 있다. 한국어 개체명 인식에 대해서도 LSTM-CRF 모델을 기본 골격으로 단어, 형태소, 자모음, 품사, 기구축 사전 정보 등 다양한 정보와 외부 자원을 활용하여 성능을 높이는 연구가 진행되고 있다. 그러나 이런 방법은 언어 자원과 성능이 좋은 자연어 처리 모듈(형태소 세그먼트, 품사 태거 등)이 없으면 사용할 수 없다. 본 논문에서는 LSTM-CRF와 최소한의 언어 자원을 사용하여 다국어에 대한 개체명 인식에 대한 성능을 평가한다. LSTM-CRF의 입력은 문자 기반의 n-gram 표상으로, 성능 평가에는 unigram 표상과 bigram 표상을 사용했다. 한국어, 일본어, 중국어에 대해 개체명 인식 성능 평가를 한 결과 한국어의 경우 bigram을 사용했을 때 78.54%의 성능을, 일본어와 중국어는 unigram을 사용했을 때 각 63.2%, 26.65%의 성능을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.190-194
• /
• 2019

개체명 인식이란 주어진 문서에서 개체명의 범위를 찾고 개체명을 분류하는 것이다. 최근 많은 연구는 신경망 모델을 이용하며 하나 이상의 단어로 구성된 개체명을 BIO 표기법으로 표현한다. BIO 표기법은 개체명이 시작되는 단어의 표지에 B(Beginning)-를 붙이고, 개체명에 포함된 그 외의 단어의 표지에는 I(Inside)-를 붙이며, 개체명과 개체명 사이의 모든 단어의 표지를 O로 간주하는 방법이다. BIO 표기법으로 표현된 말뭉치는 O 표지가 90% 이상을 차지하므로 O 표지에 대한 혼잡도가 높아지는 문제와 불균형 학습 문제가 발생된다. 본 논문에서는 BIO 표기법 대신에 BIT 표기법을 제안한다. BIT 표기법이란 BIO 표기법에서 O 표지를 T(Tag) 표지로 변환하는 방법이며 본 논문에서 T 표지는 품사 표지를 나타낸다. 실험을 통해서 BIT 표기법이 거의 모든 경우에 성능이 향상됨을 확인할 수 있었다.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.337-342
• /
• 2020

음절 기반 모델은 음절 하나가 모델의 입력이 되며, 형태소 분석을 기반으로 하는 모델에서 발생하는 에러 전파(error propagation)와 미등록어 문제를 회피할 수 있다. 개체명 인식은 주어진 문장에서 고유한 의미를 갖는 단어를 찾아 개체 범주로 분류하는 자연어처리 태스크이며, 슬롯 필링(slot filling)은 문장 안에서 의미 정보를 추출하는 자연어이해 태스크이다. 본 논문에서는 자동차 도메인 슬롯 필링 데이터셋을 구축하며, 음절 단위로 한국어 개체명 인식과 슬롯 필링을 수행하고, 성능 향상을 위하여 한국어 대용량 코퍼스를 음절 단위로 사전학습한 ELECTRA 모델 기반 학습방법을 제안한다. 실험 결과, 국립국어원 문어체 개체명 데이터셋에서 F1 88.93%, ETRI 데이터셋에서는 F1 94.85%, 자동차 도메인 슬롯 필링에서는 F1 94.74%로 우수한 성능을 보였다. 이에 따라, 본 논문에서 제안한 방법이 의미있음을 알 수 있다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.1
• /
• pp.407-420
• /
• 2015

Natural Language Question Answering (NLQA) and Prescriptive Analytics (PA) have been identified as innovative, emerging technologies in 2015 by the Gartner group. These technologies require knowledge bases that consist of data that has been extracted from unstructured texts. Every business requires a knowledge base for business analytics as it can enhance companies' competitiveness in their industry. Most intelligent or analytic services depend a lot upon on knowledge bases. However, building a qualified knowledge base is very time consuming and requires a considerable amount of effort, especially if it is to be manually created. Another problem that occurs when creating a knowledge base is that it will be outdated by the time it is completed and will require constant updating even when it is ready in use. For these reason, it is more advisable to create a computerized knowledge base. This research focuses on building a computerized knowledge base for business using a supervised learning and rule-based method. The method proposed in this paper is based on information extraction, but it has been specialized and modified to extract information related only to a business. The business knowledge base created by our system can also be used for advanced functions such as presenting the hierarchy of technologies and products, and the relations between technologies and products. Using our method, these relations can be expanded and customized according to business requirements.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.26 no.3
• /
• pp.37-51
• /
• 2023

Spatial information is essential for interpreting urban phenomena. Methodologies for spatializing urban information, especially when it lacks location details, have been consistently developed. Typical methods include Geocoding using structured address information or place names, spatial integration with existing geospatial data, and manual tasks utilizing reference data. However, a vast number of documents produced by administrative agencies have not been deeply dealt with due to their unstructured nature, even when there's demand for spatialization. This research utilizes the natural language processing model BERT to spatialize public documents related to urban planning. It focuses on extracting sentence elements containing addresses from documents and converting them into structured data. The study used 18 years of urban planning public announcement documents as training data to train the BERT model and enhanced its performance by manually adjusting its hyperparameters. After training, the test results showed accuracy rates of 96.6% for classifying urban planning facilities, 98.5% for address recognition, and 93.1% for address cleaning. When mapping the result data on GIS, it was possible to effectively display the change history related to specific urban planning facilities. This research provides a deep understanding of the spatial context of urban planning documents, and it is hoped that through this, stakeholders can make more effective decisions.

• Journal of the Korean Institute of Landscape Architecture
• /
• v.50 no.5
• /
• pp.90-102
• /
• 2022

In recent years, as interest in healing increases, outdoor spaces with the concept of healing have been created. For more professional and in-depth planning and design, the perception and characteristics of outdoor healing places through social media posts were analyzed using NER. Text mining was conducted using 88,155 blog posts, and frequency analysis and clique cohesion analysis were conducted. Six elements were derived through a literature review, and two elements were added to analyze the perception and the characteristics of healing places. As a result, visitors considered place elements, date and time, social elements, and activity elements more important than personnel, psychological elements, plants and color, and form and shape when visiting healing places. The analysis allowed the derivation of perceptions and characteristics of healing places through keywords. From the results of the Clique, keywords, such as places, date and time, and relationship, were clustered, so it was possible to know where, when, what time, and with whom people were visiting places for healing. Through the study, the perception and characteristics of healing places were derived by analyzing large-scale data written by visitors. It was confirmed that specific elements could be used in planning and marketing.

• Journal of the Korean Society for information Management
• /
• v.40 no.2
• /
• pp.115-135
• /
• 2023

The purpose of this study is to assess the effectiveness of using deep learning language models to extract references automatically and create a reference database for research reports in an efficient manner. Unlike academic journals, research reports present difficulties in automatically extracting references due to variations in formatting across institutions. In this study, we addressed this issue by introducing the task of separating references from non-reference phrases, in addition to the commonly used metadata extraction task for reference extraction. The study employed datasets that included various types of references, such as those from research reports of a particular institution, academic journals, and a combination of academic journal references and non-reference texts. Two deep learning language models, namely RoBERTa+CRF and ChatGPT, were compared to evaluate their performance in automatic extraction. They were used to extract metadata, categorize data types, and separate original text. The research findings showed that the deep learning language models were highly effective, achieving maximum F1-scores of 95.41% for metadata extraction and 98.91% for categorization of data types and separation of the original text. These results provide valuable insights into the use of deep learning language models and different types of datasets for constructing reference databases for research reports including both reference and non-reference texts.