• KIPS Transactions on Computer and Communication Systems
• /
• v.12 no.3
• /
• pp.103-110
• /
• 2023

With the spread of the Internet, anonymous communities emerged along with the activation of communities for communication between people, and many users are doing harm to others, such as posting aggressive posts and leaving comments using anonymity. In the past, administrators directly checked posts and comments, then deleted and blocked them, but as the number of community users increased, they reached a level that managers could not continue to monitor. Initially, word filtering techniques were used to prevent malicious writing from being posted in a form that could not post or comment if a specific word was included, but they avoided filtering in a bypassed form, such as using similar words. As a way to solve this problem, deep learning was used to monitor posts posted by users in real-time, but recently, the community uses words that can only be understood by the community or from a human perspective, not from a general Korean word. There are various types and forms of characters, making it difficult to learn everything in the artificial intelligence model. Therefore, in this paper, we proposes a preprocessing technique in which each character of a sentence is imaged using a CNN model that learns the consonants, vowel and spacing images of Korean word and converts characters that can only be understood from a human perspective into characters predicted by the CNN model. As a result of the experiment, it was confirmed that the performance of the LSTM, BiLSTM and CNN-BiLSTM models increased by 3.2%, 3.3%, and 4.88%, respectively, through the proposed preprocessing technique.

• Journal of Broadcast Engineering
• /
• v.28 no.3
• /
• pp.333-336
• /
• 2023

Recently, the transformer model used in natural language processing is also applied to the image super resolution field, showing good performance. However, these transformer based models have a disadvantage that they are difficult to use in small mobile devices because they are complex and have many learning parameters and require high hardware resources. Therefore, in this paper, we propose a knowledge distillation technique that can effectively reduce the size of a transformer based super resolution model. As a result of the experiment, it was confirmed that by applying the proposed technique to the student model with reduced number of transformer blocks, performance similar to or higher than that of the teacher model could be obtained.

• The Journal of Korean Association of Computer Education
• /
• v.14 no.2
• /
• pp.83-94
• /
• 2011

In the biomedical domain, the longest matching method is frequently used for recognizing named entity written in the literature. This method uses a dictionary as a resource for named entity recognition. If there exist appropriated dictionary about target domain, the longest matching method has the advantage of being able to recognize the entities of target domain quickly and exactly. However, the longest matching method is difficult to recognize the enumerated named entities, because these entities are frequently expressed as being omitted some words. In order to resolve this problem, we propose the partial matching method using a dictionary. The proposed method makes several candidate entities on the assumption that the ellipses may be included. After that, the method selects the most valid one among candidate entities through the optimization algorithm. We tested the longest and partial matching method about HLA entities: HLA gene, antigen, and allele entities, which are frequently enumerated among biomedical entities. As preparing for named entity recognition, we built two new resource, extended dictionary and tag-based dictionary about HLA entities. And later, we performed the longest and partial matching method using each dictionary. According to our experiment result, the longest matching method was effective in recognizing HLA antigen entities, in which the ellipses are rare, and the partial matching method was effective in recognizing HLA gene and allele entities, in which the ellipses are frequent. Especially, the partial matching method had a high F-score 95.59% about HLA alleles.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.44-45
• /
• 2019

최근 CNN 등 인공신경망은 최근 이미지 분류, 객체 인식, 자연어 처리 등 다양한 분야에서 뛰어난 성능을 보이고 있다. 그러나, 대부분의 분야에서 보다 더 높은 성능을 얻기 위해 사용한 인공신경망 모델들은 파라미터 수 및 연산량 등이 방대하여, 모바일 및 IoT 디바이스 같은 연산량이나 메모리가 제한된 환경에서 추론하기에는 제한적이다. 따라서 연산량 및 모델 파라미터 수를 압축하기 위한 딥러닝 경량화 알고리즘이 연구되고 있다. 본 논문에서는 임베디트 보드에서의 압축된 CNN 모델의 성능을 검증한다. 인공지능 지원 맞춤형 칩인 QCS605 를 내장한 임베디드 보드에서 카메라로 입력한 영상에 대해서 원 CNN 모델과 압축된 CNN 모델의 분류 성능과 동작속도 비교 분석한다. 본 논문의 실험에서는 CNN 모델로 MobileNetV2, VGG16 을 사용했으며, 주어진 모델에서 가지치기(pruning) 기법, 양자화, 행렬 분해 등의 인공신경망 압축 기술을 적용하였을 때 원래의 모델 대비 추론 시간 및 분류의 정확도 성능을 분석하고 인공신경망 압축 기술의 유용성을 확인하였다.

• Korean Journal of Cognitive Science
• /
• v.22 no.4
• /
• pp.429-447
• /
• 2011

In this paper, we propose a modified deterministic Korean dependency parser using a projectivity. The modified parser is improved by finding errors, such as cross dependency, from the original parsing results and correcting them according to the projectivity and head-final principles. Our parser also uses parsing history information in addition to rich features, which only a deterministic algorithm can use. Results on the modified parser for ETRI(2005) corpus, that consists of complex sentences, show that our parser outperforms other parsers.

• Journal of the Korea Computer Industry Society
• /
• v.4 no.12
• /
• pp.1103-1111
• /
• 2003

Information retrieval system with indexing system matching single keyword is simple and popular. But with single keyword matching it is very hard to represent the exact meaning of documents and the set of documents from retrieval is very large, therefore it can't satisfy the user of the information retrieval systems. This paper proposes a phrase-based indexing system based on the phrase, the larger syntax unit than a single keyword. Web documents include lots of syntactic errors, the natural language parser with high Quality cannot be expected in Web. Partial trees, even not a full tree, from fully bottom-up parsing is still useful for extracting phrases, and they are much more discriminative than single keyword for index. It helps the information retrieval system enhance the efficiency and reduce the processing overhead.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.689-691
• /
• 2018

의존 구문 분석은 자연어 이해 영역의 대표적인 과제 중 하나이다. 본 논문에서는 한국어 의존 구분 분석의 성능 향상을 위해 Deep Bi-affine Network 와 스택 포인터 네트워크의 앙상블 모델을 제안한다. Bi-affine 모델은 그래프 기반 방식, 스택 포인터 네트워크의 경우 그래프 기반과 전이 기반의 장점을 모두 사용하는 모델로 서로 다른 모델의 앙상블을 통해 성능 향상을 기대할 수 있다. 두 모델 모두 한국어 어절의 특성을 고려한 자질을 사용하였으며 세종 의존 구문 분석 데이터에 대해 UAS 90.60 / LAS 88.26(Deep Bi-affine Network), UAS 92.17 / LAS 90.08(스택 포인터 네트워크) 성능을 얻었다. 두 모델에 대한 앙상블 기법 적용시 추가적인 성능 향상을 얻을 수 있었다.

• The Journal of the Korea Contents Association
• /
• v.15 no.10
• /
• pp.636-646
• /
• 2015

In information society, the need to detect document duplication and plagiarism is increasing. Many studies have progressed to meet such need, but there are limitations in increasing document duplication detection quality due to technological problem of natural language processing. Recently, some studies tried to increase the quality by applying syntatic-semantic analysis technique. But, the studies have the problem comparing syntactic-semantic trees. This paper develops a syntactic-semantic tree comparator, designs and implements a discriminator of similar documents using the comparator. To evaluate the system, we analyze the correlation between human discrimination and system discrimination with the comparator. This analysis shows that the proposed discrimination has good performance. We need to define the document type and improve the processing technique appropriate for each type.

• Health Policy and Management
• /
• v.30 no.1
• /
• pp.15-25
• /
• 2020

While healthcare data sets include extensive information about patients, many researchers have limitations in analyzing them due to their intrinsic characteristics such as heterogeneity, longitudinal irregularity, and noise. In particular, since the majority of medical history information is recorded in text codes, the use of such information has been limited due to the high dimensionality of explanatory variables. To address this problem, recent studies applied word embedding techniques, originally developed for natural language processing, and derived positive results in terms of dimensional reduction and accuracy of the prediction model. This paper reviews the deep learning-based natural language processing techniques (word embedding) and summarizes research cases that have used those techniques in the health care field. Then we finally propose a research framework for applying deep learning-based natural language process in the analysis of domestic health insurance data.

• Annual Conference on Human and Language Technology
• /
• 2002.10e
• /
• pp.163-168
• /
• 2002

오늘날 자연어 구문 분석 기술은 만족할 만한 수준에 도달하지 못하고 있고 한국어 구문분석 기술 역시 만족할만한 수준과는 거리가 멀다. 특히 문장의 길이가 긴 문장의 경우 구문분석기가 너무 많은 계산 량으로 인해 제대로 동작하지 못하는 경우가 빈번히 발생하고, 비록 구문구조 결과를 내더라도 정확도가 낮은 경우가 많다. 그 이유는 문장의 길이가 길어질수록 중의성이 매우 증가하여 많은 수의 구문분석 결과가 가능하기 때문이다. 이 중에서 정확한 구문구조를 선택하는 문제는 매우 어려워서 기존의 긴 전체 문장에 대한 구문구조를 한번에 계산하려는 시도는 앞으로도 계속 좋은 결과를 기대하기 어렵다. 따라서 우리는 문장의 길이에 상관없이 항상 안정적으로 결과를 내며, 구문분석에 소요되는 시간이 비교적 짧고, 정확도 역시 높은 구문분석기를 개발하고자 한다. 이를 위하여 전체 문장을 여러 개의 구간으로 분할하여 각 구간을 독립적으로 구문 분석한다. 그 다음 각 구간의 결과를 통합하여 전체 문장에 대한 결과를 생성하는 기법을 택하였다.