• 인지과학
• /
• 제12권1_2호
• /
• pp.35-42
• /
• 2001

본 논문에서 문서는 문서관계도라고 하는 가중치 그래프로 표현된다. 노드는 문서의 구성요소인 문장을 명사벡터로 표현하고, 링크는 노드들 간의 의미적인 관계를 표현하며 의미적 유사도를 가중치로 가지고 있다. 한 노드의 인접한 노드를 사이의 유사도 합을 도합유사도라고 하며, 이를 문서에서 문장의 중요도로 간주한다. 본 논문에서는도합유사도를 이용한 한국어 문서요약 시스템을 기술한다. 실험에 사용된 평가용 요약문서는 정보처리관련 분야에서 수집된 논문 100편과 KORDIC에서 구축한 신문기사 105건을 이용하였다. 문서요약 시스템에 의해서 생상된 요약문서와 크기가 본문의 20%이고 평가용 요약문서가 논문(서론과 결론)일 경우, 재현율과 정확률은 각각 46.6%와 76.9%를 보였으며, 또한 평가용 요약문서가 신문기사일 경우, 재현율과 정확률은 각각 30.5%과 42.3%를 보였다.

• 정보처리학회논문지D
• /
• 제16D권5호
• /
• pp.683-690
• /
• 2009

소셜 네트워크(Social Network)는 웹 환경에서 개인 중심의 네트워크로 구성되어 웹 사용자별 프로파일을 탐색하고 새로운 연결을 형성함으로써 정보의 소통을 지원한다. 따라서 유사한 내재적 정보를 가진 웹 사용자들로 구성 된 소셜 네트워크를 찾아서 검색에 적용한다면 검색의 효율성과 검색 결과에 대한 웹 사용자의 만족도를 향상 시킬 수 있다. 본 논문에서는 첫째, 웹 사용자간 직접 또는 간접적인 연결로 구성된 소셜 네트워크를 구성 한다. 둘째, 사용자들의 속성(Feature)에 내재된 정보를 이용하여 주제(topic)별 웹 사용자 간 유사성(Similarity)을 산정한 후, 주제(Topic)별 변화되는 유사성에 따라 소셜 네트워크를 재구성한다. 마지막으로 산정된 유사성과 웹 사용자들의 검색결과에 대한 만족도, 즉 검색 패턴(Search Pattern)을 비교 실험 한다. 실험 결과 주제별 유사성이 높은 웹 사용자 간에는 검색 패턴 또한 유사함을 확인 하였다. 이와 같은 사실을 검색에 적용한다면 개인화 검색(Personalized Search) 및 소셜 검색(Social Search)의 효율성 및 신뢰성 향상에 기여 할 수 있다.

• 인터넷정보학회논문지
• /
• 제6권1호
• /
• pp.51-64
• /
• 2005

본 논문에서는 서로 다른 두 개의 C 프로그램의 구문트리를 이용하여 유사도를 평가하는 시스템을 제시한다. 구문 트리를 이용하는 방법은 기존의 유사도 평가 방법과는 달리 들여쓰기, 여백, 설명문 등 프로그램과 무관한 프로그램 스타일의 변화에 민감하지 않으며, 문장, 코드 블록, 함수 등의 순서 바꾸기 같은 제어 구조의 변경에 민감하지 않은 특징을 가지고 있다. 그리고 프로그램을 파싱함으로써 구문 오류도 함께 검사찬 수 있는 장점을 제공한다. 논문에서는 유사도를 평가하기 위한 알고리즘과 함께 프로그램의 비교횟수를 줄이기 위한 그룹 짓기 알고리즘도 같이 제공한다. 실험부분에서는 구문트리 비교방법을 이용한 프로그램의 유사도 평가 결과와, 그룹 짓기를 수행한 후에 많은 비교 횟수를 줄일 수 있다는 것을 보여준다.

• 한국도로학회논문집
• /
• 제19권4호
• /
• pp.37-44
• /
• 2017

PURPOSES : Conductive and convective heat transfer simulations for an asphalt mixture were made by using discrete element method (DEM) and similarity principle. METHODS : In this research, virtual specimens composed of discrete element method particles were generated according to four different predetermined particle size distribution curves. Temperature variations of the four different particles for a given condition were estimated and were compared with measurements and analytical solutions. RESULTS : The virtual specimen with mixed particles and with the smallest particle show very good agreement with laboratory test results and analytical solutions. As particle size decreases, better heat transfer simulation can be performed due to smaller void ratio and more contact points and areas. In addition, by utilizing the similarity principle of thermal properties and corresponding time unit, analytical time can be drastically reduced. CONCLUSIONS : It is concluded that the DEM asphalt mixture specimens with similarity principle could be used to predict the temperature variation for a given condition. It is observed that the void ratio has critical effect on prediction of temperature variation. Comparing the prediction for a 4 mm particle specimen with a mixed particle specimen, it is also concluded that predicting the mixed particle specimen temperature is much more efficient considering the number of particles that are directly associated with computational time in DEM analysis.

• 지식경영연구
• /
• 제16권3호
• /
• pp.149-169
• /
• 2015

In this study, we try to understand the role of online social recommendation and the similarity of preferences between the recommender and the recommendee on consumer decisions in the framework of the two stage purchase decision-making process. Applying construal level theory to our context, we expect that the role of social recommendation and the similarity of preferences would vary over the stages in the two-stage decision making process. To test our hypotheses, we collected the data through an incentive compatible experiment, and analyzed the data with nested logit model. As a result, we found that the role of online social recommendation varies over the stages. Consumers take recommendation from similar others at the stage of consideration set formation, but no longer consider it at the stage of final choice. Consumers take recommendation from dissimilar others at the stage of consideration set formation. At the stage of final choice, however, consumers avoid choosing the option recommended by dissimilar others. The results of our study enrich the understanding about the role of social recommendation, and have implication to marketing practitioners who attempt to make online social recommendation system more efficient.

• 제어로봇시스템학회논문지
• /
• 제5권5호
• /
• pp.550-557
• /
• 1999

This paper describes a road-following controller using the proposed neural network for autonomous vehicle. Road-following with visual sensor like camera requires intelligent control algorithm because analysis of relation from road image to steering control is complex. The proposed neural network, relative similarity modular network(RSMN), is composed of some learning networks and a partitioniing network. The partitioning network divides input space into multiple sections by similarity of input data. Because divided section has simlar input patterns, RSMN can learn nonlinear relation such as road-following with visual control easily. Visual control uses two criteria on road image from camera; one is position of vanishing point of road, the other is slope of vanishing line of road. The controller using neural network has input of two criteria and output of steering angle. To confirm performance of the proposed neural network controller, a software is developed to simulate vehicle dynamics, camera image generation, visual control, and road-following. Also, prototype autonomous electric vehicle is developed, and usefulness of the controller is verified by physical driving test.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제11권9호
• /
• pp.4476-4490
• /
• 2017

The aim of image annotation is to determine labels that can accurately describe the semantic information of images. Many approaches have been proposed to automate the image annotation task while achieving good performance. However, in most cases, the semantic similarities of images are ignored. Towards this end, we propose a novel Visual-Semantic Nearest Neighbor (VS-KNN) method by collectively exploring visual and semantic similarities for image annotation. First, for each label, visual nearest neighbors of a given test image are constructed from training images associated with this label. Second, each neighboring subset is determined by mining the semantic similarity and the visual similarity. Finally, the relevance between the images and labels is determined based on maximum a posteriori estimation. Extensive experiments were conducted using three widely used image datasets. The experimental results show the effectiveness of the proposed method in comparison with state-of-the-arts methods.

• 유통과학연구
• /
• 제19권2호
• /
• pp.53-64
• /
• 2021

Purpose: Even though arranging images of products is a common practice in the online retail context, relatively little attention has been paid to the distance effect among alternatives, that is, how distance among displayed products can impact consumers' responses. Drawing on contagion theory, the primary goal of the current study is to investigate how spatial distance between two products in a product display can influence consumers' perceived similarity. Research design, data and methodology: This study used a 2(spatial distance: close vs. far) experimental design and collected data from undergraduate students in Korea through an online survey using Qualtrics. ANOVA was conducted to test the proposed effect, in which the dependent variables are the perceived similarity of usage occasion/purpose (Study 1) and the indexed differences of perceived brand statuses between two products (Study 2). Results: The results of both experiments indicated that the displayed products were perceived to be more similar to one another when products were presented close together (vs. far). Conclusions: The results help to fill a research gap and provide a better understanding of the role of physical distance in diverse marketing communications. This is especially useful when designing online shopping websites to form perceptions of brand images.

• 한국인터넷방송통신학회논문지
• /
• 제22권5호
• /
• pp.171-176
• /
• 2022

최근 딥러닝 기술의 발전과 함께 추천 시스템의 영역도 다양해졌다. 본 논문은 학습률 향상을 위한 알고리즘을 연구하였으며 Word2Vec 모델의 성능 특징과 비교를 통해 단어에 따른 유의어 결과를 연구하였다. 문제 추천 알고리즘은 Word2Vec 모델의 특징인 텍스트 간 의미 반영 및 유사성 테스트를 통해 표현된 값으로 구현됐다. Word2Vec 의 학습 결과를 통해 텍스트 유사도 값을 이용해 문제 추천을 진행하였으며 유사도가 높은 문제를 추천할 수 있다. 실험 과정에서 정량적인 데이터양으로는 정확성이 낮아지는 결과를 보았으며 데이터 셋의 데이터양이 방대할수록 정확성을 높일 수 있음을 확인하였다.

• 한국해양공학회지
• /
• 제31권4호
• /
• pp.257-265
• /
• 2017

In this study, a series of numerical simulations was conducted in order to design a truncated mooring line with a static similarity to the prototype. A finite element method based on minimizing the potential energy was utilized to describe the dynamics of mooring lines. The prototype mooring lines considered were installed at a water depth of 1,000 m, whereas the KRISO ocean engineering basin (OEB) in Daejeon has a water depth of 3.2 m, which represents 192 m using a scaling of 1:60. First, an investigation for the design of the truncated mooring line was carried out to match the static characteristics of the KRISO Daejeon OEB environment. Then, the same procedure was performed with the KRISO new deepwater ocean engineering basin (DOEB) that is under construction in Busan. This new facility has a water depth of 15 m, which reflects a real scale depth of 900 m considering the 1:60 scaling factor. A finite element method was used to model the mooring line dynamics. It was found that the targeted truncated mooring line could not be designed under the circumstances of the KRISO OEB with any material properties, whereas several mooring lines were easily matched to the prototype under the circumstances of the KRISO DOEB.