• KIISE Transactions on Computing Practices
• /
• v.23 no.5
• /
• pp.322-327
• /
• 2017

The classification problem in the field of Natural Language Processing has been studied for a long time. Continuing forward with our previous research, which classifies large-scale text using Convolutional Neural Networks (CNN), we implemented Recurrent Neural Networks (RNN), Long-Short Term Memory (LSTM) and Gated Recurrent Units (GRU). The experiment's result revealed that the performance of classification algorithms was Multinomial Naïve Bayesian Classifier < Support Vector Machine (SVM) < LSTM < CNN < GRU, in order. The result can be interpreted as follows: First, the result of CNN was better than LSTM. Therefore, the text classification problem might be related more to feature extraction problem than to natural language understanding problems. Second, judging from the results the GRU showed better performance in feature extraction than LSTM. Finally, the result that the GRU was better than CNN implies that text classification algorithms should consider feature extraction and sequential information. We presented the results of fine-tuning in deep neural networks to provide some intuition regard natural language processing to future researchers.

• Journal of Korean Library and Information Science Society
• /
• v.47 no.3
• /
• pp.95-114
• /
• 2016

The purpose of this preliminary study is to collect specific examples of book reports and understand semantic characteristics of them through semantic network. The analysis was conducted with 23 book reports which classified by three groups. The keywords were selected from the of book reports. Five types of keyword network were composed based on co-occurrence relations with keywords. The result of this study is following these. First, each keyword network of book reports of groups and individuals is shown to have different structural characteristics. Second, each network has different high centrality keywords according to the result analysis of 3 types of centrality(degree centrality, closeness centrality, betweenness centrality). These characteristic means that keyword network analysis is useful in recognizing the characteristics of not only groups' and but also individual's book reports.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.12
• /
• pp.3139-3149
• /
• 1997

E-mail system has been widely used for information transmission. As multimedia technologies are developed, many Internet services support multimedia. E-mail system only transmits 7bit ASCII data using simple mail transfer protocol(SMTP) specified in RFC 821. E-mail system can't support the authoring and the transmission of multimedia data. In this study multimedia mail authoring system has been designed and implemented in order to provide message integration. Multimedia messages can be authored and transformed into text types with this system. Transformed messages are transmitted from sender's mail transfer agent (MTA) to receiver's MTA by means of SMTP. Messages can be authored in hypertext markup language (HTML) including text, image, audio and video. This system provides integrated messages which consist or various media. BASE64 Encoding method is used in order to transform multimedia messages into text types. Received mails are managed by multimedia user agent (MUA) which is proposed in this study. Convenience for user is improved in this system.

• 한국HCI학회:학술대회논문집
• /
• 2007.02a
• /
• pp.591-597
• /
• 2007

웨어러블 컴퓨팅의 지속적인 발전과 함께, 이들이 지원하는 기능은 점점 다양화되고 전문화되었다. 하지만 기기의 종류와 기능이 늘어나면서 각 기기의 상태를 확인하는 작업 자체가 복잡하고 불필요한 시간을 소요하기도 한다. 또한 이를 시각 디스플레이에만 의존하면, 디스플레이 크기의 한계와 더불어 시각적인 과부하가 발생할 수 있다. 본 연구에서는 이러한 문제를 극복하기 위해 웨어러블 각 기기의 상태정보 피드백으로서 소리를 이용하였다. 실험의 순서는 먼저 시각정보에 더하여 청각정보를 추가적으로 제시하는 방식의 유용성을 검증한 다음, 기기 특정적인 청각 아이콘(Auditory Icon)의 효과를 확인하였다. 청각정보의 유용성을 확인하는 단계에서는 참가자에게 특정 기기의 위치와 그 상태에 관한 정보를 시각 디스플레이를 통해 제시하면서, 조건에 따라 다양한 형태의 청각 디스플레이를 제시하고 기기 및 상태정보 확인에 소요되는 시간을 측정하였다. 시각디스플레이의 방식은 다양하지만, 본 연구에서는 사람의 몸을 나타내는 그림 위에 기기가 장착된 위치를 붉은 점으로 표시하고, 그 상태를 텍스트로 표시하는 방식(실험 1)과 텍스트를 통해 기기의 종류와 상태를 알려주는 방식(실험 2)을 적용하였다. 기기 확인에 소요되는 반응시간을 기록한 실험결과는 시각디스플레이를 통해서만 제시하는 조건보다 시각디스플레이와 청각디스플레이를 동시에 제시하는 조건에서 나은 수행을 보여주며, 정보의 내용이 일정 범위 내에서 구체적으로 제시될 수록 수행이 증가하는 경향을 보여주고 있다. 이어지는 기기 특정적인 청각 아이콘의 효과를 확인하는 단계에서는, 소리의 높낮이를 통해 기기의 위치를 표현하는 대신, 소리만 듣고 기기가 무엇인가를 알려줄 수 있도록, 참가자 평정을 통해 기기 특정적인 청각 아이콘을 선정하고, 이를 상태정보 디스플레이에 추가하였다. 기기의 수와 제시조건은 앞선 실험과 동일하였으며, 역시 기기 및 상태정보 확인에 소요되는 반응시간을 체크하였다. 실험결과는 이전의 실험과 비슷하게 시각과 청각 디스플레이의 중복제시 조건에서 높은 수행을 보였으며, 청각 아이콘의 사용으로 인한 수행의 증가도 확인할 수 있었다. 실험 2에서는 추가적으로 청각 디스플레이의 내용에 대한 주관적인 평정을 실시 하였는데, 기기 특정적인 청각 아이콘은 참가자에게 편의성, 효용성, 적합성, 확신도에서도 높은 점수를 얻었다. 이러한 결과는 청각 피드백을 이용한 정보의 디스플레이는 착용자의 시선을 자유롭게 함으로써 일상생활의 수행을 높이고, 웨어러블 컴퓨터 디스플레이 크기의 제약을 해결하는 방법이 될 수 있는 동시에 주관적인 사용성 측면에서 발전가능성이 있음을 보여준다.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.12 no.4
• /
• pp.37-42
• /
• 2012

LED is just a semiconductor which can produce light. Currently, there are active research works on LED lighting technologies according to the growth of energy-saving environmental industry. Especially, LED communication is one of the active research works in these fields. In this paper, we design a LED communication-based multi-hop transmission wireless network system. The designed system consists of a transmission circuit system(transmitter) using LED and a receiving circuit system(receiver) using PD(photo detector) and OP-Amp, and relay system which can support multi-hop wireless network service with PD, OP-Amp, and LED, respectively. The experiments for the designed system are performed as follows. One computer is connected at the end of transmitter and receiver, respectively. There are two relays between transmitter and receiver, and text files are transmitted continuously by using text transmission programming. In this experiment, we test the performance with various baud rates, transmission ranges.

• Journal of Korea Game Society
• /
• v.8 no.3
• /
• pp.3-14
• /
• 2008

Classifying digital games' genres is quite ambiguous work, as they have been using other media's genre classification such as film genre. Therefore this paper tries to propose a new way of classifying genres of digital games. There are two kinds of approaches to digital games genres: one is based on gamers' activities and the other is on game text. The former turned out to be limitative because the meanings of gamers' activities are sometimes overlapped and this overlapping causes lack of objectivity in deciding which activity belongs to which genre. On the other hand, the latter is comparatively clear and accurate as it is based on what a game text provides. After all. a new classification of digital games based on game text makes total of 7 digital game types, which are Physical obstacles, Mainly Physical & partially intellectual obstacles, Intellectual obstacles, Mainly intellectual & partially physical obstacles, Self-supporting games, Confronting games, Ranking games, and each type has its own convention and characteristics.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.23 no.5
• /
• pp.406-411
• /
• 2013

Big data issue has been considered in diverse fields. Also, big data learning has been required in all areas such as engineering and social science. Statistics and machine learning algorithms are representative tools for big data learning. In this paper, we study learning tools for big data and propose an efficient methodology for big data learning via legacy data to practical application. We apply our big data learning to patent analysis, because patent is one of big data. Also, we use patent analysis result for technology forecasting. To illustrate how the proposed methodology could be applied in real domain, we will retrieve patents related to big data from patent databases in the world. Using searched patent data, we perform a case study by text mining preprocessing and multiple linear regression of statistics.

• The Transactions of the Korea Information Processing Society
• /
• v.6 no.6
• /
• pp.1481-1492
• /
• 1999

As Internet is popularized by the advent of Web concept having characteristics such as open network, user-friendly, and easy-usage, there are many changes in Information systems providing various information. Web is rapidly transferred traditional Information systems to Web-based Information systems, because it provides not only text information but also multimedia information including image, audio, video, and etc. Also, as information contents were changed from text-based simple abstract information to full-text information, there was appeared various document formats processing Full-text information. But, as they naturally demand large systems memory, long processing time, broader transmission bandwidth, and etc, estimating of these factors is necessary when constructing information systems. This paper focuses on how to design and construct information system processing full-text information and providing function of an integrated document. Primarily, we should review standard document format which is used or developed, and any document format is appropriate to process full-text information in review with viewpoint of information system. Also, practically we should construct information system providing full-text information based on PDF document.

• Annual Conference of KIPS
• /
• 2003.11b
• /
• pp.853-856
• /
• 2003

바이오 텍스트 마이닝을 위한 정보 추출의 첫 단계는 생물학적 문헌으로부터의 유전자, 단백질, 세포조직 등과 같은 생물학적 개체명의 인식이다. 생물학적 개체명의 명명법상 특징이 매우 다양하고 저자의 개성에 의해 쉽게 좌우되어 단순히 규칙이나 학습 방법 만으로는 쉽게 개체명들을 인식할 수 없다. 또한, 생물학 관련 문헌에 나오는 가능한 모든 개체명과 이들의 모든 변형을 수록하는 것은 현실적으로 불가능하므로 이를 해결하기 위해 이미 알려진 개체명에 대해서 기본적으로 사전을 탐색하고 알려지지 않은 용어들을 규칙과 통계 기반 방법을 통하여 인식하는 것이 효과적이다. 그러나 만족할 만한 수준의 양질의 사전을 구축하는 것은 쉽지 않을 뿐만 아니라 많은 비용이 소요되며, 어느 순간 만족할 만한 성능을 낼 수 있는 사전을 구축했다. 할지라도 유지 관리 하는 것이 결코 쉬운 일이 아니며 마찬가지로 많은 비용을 필요로 하게 된다. 따라서, 잘 구축된 자원으로부터 필요한 정보를 추출하여 적절한 사전을 자동으로 구축하여 활용하는 방법을 사용할 경우, 사전 구축 및 관리에 드는 많은 비용을 줄이면서도 상당히 효과적인 성능을 얻을 수 있을 것이다. 본 연구에서는 바이오 텍스트 마이닝 엔진을 위한 생물학적 개체명 사전을 자동으로 구축하고 이를 쉽게 관리하도록 하는 도구를 개발하였다.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.6
• /
• pp.727-731
• /
• 2010

Recently, mobile life-log data are collected by mobile devices and used to recode one's life. In order to help a user search data, a mobile life-log semantic network is introduced for storing logs and retrieving associative information. However, associative search systems on common semantic networks in previous studies provide for a user with only found data as text to users. This paper proposes an associative search system for mobile life-log semantic network that supports selection and keyword associative search of which a process and result are a visualized graph representing associative data and their relationships when a user inputs a keyword for search. In addition, by using semantic abstraction, the system improves user's understanding of search result and simplifies the resulting graph. The system's usability was tested by an experiment comparing the system and a text-based search system.