가상현실 기술의 발전으로 360도 영상 콘텐츠가 각광을 받으면서, 360도 비디오를 이용한 영상처리, 객체인식 연구가 활발히 진행되고 있다. 그러나 360도 영상은 극점에서의 왜곡이 존재하기 때문에 객체를 정확하게 인식하기 어렵다. 본 논문에서는 3차 투영법을 이용해 360도 비디오의 객체인식성능을 향상시키고자 한다. 실험 결과 본 논문에서 제안하는 알고리즘을 이용했을 때 360도 비디오에서 객체를 인식하는 최신 연구보다 13.68% 더 높은 성능을 보였다.

최근 딥러닝 기반의 컨볼루션 신경망은 컴퓨터비전 분야의 여러 문제에서 좋은 성능을 보여주었으며 다양한 형태의 컨볼루션 신경망 아키텍처가 제안되어왔다. 특히 얼굴 표정에서 감정을 인식하기 위한 문제 역시 딥 컨볼루션 신경망을 사용한 연구가 활발히 진행되고 있다. 본 논문에서는 사람의 얼굴 표정에서 나타나는 감정을 인식하기 위한 효율적인 컨볼루션 신경망을 새롭게 제안한다. 본 논문에서는 경량화된 컨볼루션 신경망을 설계하기 위해 깊이별 분리 컨볼루션을 사용하여 파라미터 수와 연산량을 감소시키고 특징 연결과 채널 정보의 압축 및 복원을 통해 특징의 재사용성과 채널 정보를 강화하였다. 제안하는 모델의 학습 파라미터 개수는 0.48 M(Million)으로 기존 모델에 비해 적은 파라미터 개수로 FER2013, RAF-single 데이터셋에서 각각 70.69%, 85.4%의 정확도를 달성하였다.

기존의 음성 기반 감정 인식 연구들은 단일한 음성 특징값으로 모델을 훈련하고 감정을 예측하였다. 그러나 음성은 복합적인 요소로 이루어져 있기 때문에 다양한 특징값을 사용한 모델을 설계할 필요가 있다. 따라서 본 논문에서는 멜 스펙트로그램, MFCC, Centroid, Roll off를 이용한 합성곱 신경망(Convolutional neural network, CNN) 모델을 제안한다. 제안하는 모델은 각 특징값에 대해 개별적으로 합성곱 신경망을 구성하고, 합성곱 신경망의 출력값에 다항 로지스틱 회귀를 적용하여 감정을 예측한다. 실험 결과 본 논문에서 제안하는 모델이 61.5%의 감정 예측 정확도를 얻어 기존에 연구되었던 단일 특징값을 사용한 모델보다 최대 27.2%p의 성능 향상을 달성하여 음성을 이용한 감정 인식 모델 설계 시 음성의 다양한 측면을 고려할 수 있는 여러 가지 특징값을 사용하는 것이 적절하다는 결론을 도출하였다.

샷 경계 검출(Shot Boundary Detection)은 영상 콘텐츠 분석을 위한 필수적인 기술이며, 다양한 방식으로 편집된 영상의 샷 경계를 정확하게 검출하기 위한 연구가 지속되어 왔다. 그러나 기존에 연구들은 고정된 샷 경계 검출 알고리즘이나 매뉴얼한 작업과 같이 학습이 불가능한 과정이 포함되어 있어 성능 개선에 한계가 있었다. 본 논문에서는 이러한 과정을 제거한 End-to-End 모델을 제안한다. 제안하는 모델은 시공간 정보 추출성능을 높이기 위해 행동 인식 데이터셋을 이용한 전이학습을 사용하고, 샷 경계 검출 성능을 높이기 위해 개선된 지식의 증류기법(Knowledge Distillation)을 결합한다. 제안하는 모델은 ClipShots 데이터셋에서 DeepSBD에 비해 cut transition과 gradual transition이 각각 5.4%, 41.29% 높은 성능을 보였고, DSM과의 비교에서 cut transition의 정확도가 1.3% 더 높은 결과를 보였다.

이미지 기반의 증강현실 분야에서 가상의 객체를 증강하고자하는 물체의 위치와 방향을 추적하기 위한 방법으로 영상의 호모그래피(Homography)를 활용하는 방법이 있다. 호모그래피는 물체를 촬영하는 위치나 방향에 따른 영상의 변화를 나타내는 값으로, 호모그래피를 추정하기 위한 방법으로 영상의 특징점을 활용한 RANSAC 알고리즘이 주로 활용되고 있다. 그러나 호모그래피를 추정하고자 하는 물체 이외의 배경에 특징점이 많은 경우 호모그래피 추정에 긴 시간이 소요되고 정확한 호모그래피를 추정하기 어렵다. 이를 해결하기 위해 본 논문에서는 영상의 입체정보를 나타내는 깊이 정보를 활용하는 방법을 제안한다. 영상의 깊이 정보를 이용하여 거리가 먼 영역과 거리가 가까운 영역을 추정하고, 거리가 먼 영역의 특징점을 필터링함으로써 정확한 호모그래피를 추정하고자하였다. 그 결과 호모그래피의 참 정보 비율이 14% 증가하였고 수행 시간과 샘플링 횟수가 60% 감소하였다.

In this paper, a new network architecture for Korean fonts transfer procedures along with an adversarial network is proposed. The architecture we proposed consists of two sub-nets: (1)an unbalanced u-net is responsible for transferring specified fonts style to another while maintaining semantic and structure information; (2)an adversarial nets. Our model employs a compound loss function that includes a L1 loss, a constant loss and a binary GAN loss to help in generating desired target fonts. The experiments demonstrate that we proposed model can automatically generate convincing realistic-looking target fonts.

The advancement of deep learning leads to the trend of using very deep network. But, it is not practical in real project, especially for people who have limited resources or real-time requirement. In this paper, we propose a new backbone network for object detection combined with multiple dilation and feature summation. By using feature summation, we can prevent loss of spatial information that is caused by convolving. And we can widen the receptive field of individual neurons without adding more parameters by using multiple dilated convolution. And by using shallow neural network as backbone network, our network can be trained and used in environment with limited resources and without pretraining it in ImageNet dataset. Based on our experiment, our network got 71% accuracy.

This paper presents a multi-scale region proposal network (RPN) for visual object tracking, inspired by Faster R-CNN and Yolo detectors which adopt an RPN to significantly speed up the detection time and achieve state-of-the-art detection performance. We expand them to apply a multi-scale region proposal network for visual tracking. Our proposed network can utilize both fine-grained features from shallow convolutional layers and discriminative features from deep convolutional layers. The features of shallow layers are good at accurate objects localization, and the features of deep convolutional layers can efficiently distinguish between target objects and backgrounds. A multi-domain learning mechanism is applied to train our network in an end-to-end way. To predict a new target object and its location in a new frame, we propose an re-ranking algorithm to determine a true object by exploiting spatial modeling, scale variants and color attributes of object proposals. Our tracker is validated on the OTB-15 object tracking benchmark, and achieves 0.603 for the success rate and 0.760 for the precision rate of the one-pass evaluation. Additionally, our tracker can run at 22 frames per second, which is very close to real-time speed. Experiment results show its outstanding performance in both tracking accuracy and speed by comparing it with existing state-of-the-art methods.

360비디오는 프레임에 왜곡이 있고 해상도가 크기 때문에 프레임 전체를 썸네일로 사용하기에 어려움이 있다. 본 논문에서는 360비디오에 적합한 썸네일을 생성하기 위해 3차 투영법을 이용하여 왜곡을 해소하면서 프레임을 작은 단위로 나눈다. 나눠진 프레임에서 CNN으로 검출한 물체와 비디오의 부가정보 사이의 유사도를 계산하여 썸네일을 선택하는 방법을 제안한다. 또한 3차 투영법 적용 전과 후 프레임에서 CNN의 물체 검출률의 비교 실험을 진행하였고 3차 투영법을 이용한 방법이 더 높은 검출률을 보였다.

Interactive video is a type of video which provides interactions for obtaining video related information or participating in video content. However, authors of interactive video need to spend much time to create the interactive video content. Many researchers have presented methods and features to solve the time-consuming problem. However, the methods are still too complicated to use and need to be automated. In this paper, we suggest an automatic interactive video authoring method via object recognition. Our proposed method uses deep learning based object recognition and an NLP-based keyword extraction method to annotate objects. To evaluate the method, we manually annotated the objects in the selected video clips, and we compared proposed method and manual method. The method achieved an accuracy rate of 43.16% for the whole process. This method allows authors to create interactive videos easily.

No Data

증강현실에서 영상과 증강된 콘텐츠 간의 이질감을 줄이기 위해서 정확한 호모그래피 행렬을 추정해야 하며, 정확한 호모그래피 행렬을 추정할때 RANSAC 알고리즘이 널리 사용된다. 그러나 RANSAC 알고리즘은 랜덤 샘플링 과정을 반복적으로 거치기 때문에 불필요한 연산 과정이 발생하고 이로 인해 알고리즘의 효율이 저하된다. 이러한 단점을 극복하기 위해 DCS-RANSAC 알고리즘이 제안 되었다. 제안된 DCS-RANSAC 알고리즘은 이미지를 특징점 분포 패턴에 따라 그룹으로 분류하고 각 그룹에 제약조건 문제를 적용하여 불필요한 연산 과정을 줄이고 정확도를 향상시킨 알고리즘이다. 그러나DCS-RANSAC 알고리즘에서 사용된 이미지 그룹 데이터는 수동적인 방법을 통해 직관적으로 분류되어 있지만 특징점 분포 패턴이 다양하지 않아 분류시 정확도가 저하되는 경우가 있다. 위의 문제점을 해결하기 위해 본 논문에서는 머신러닝 기법을 통해 이미지들을 자동으로 분류하고 각 그룹마다 각기 다른 제약조건을 적용하는 MCS-RANSAC 알고리즘을 제안한다. 제안하는 알고리즘은 머신러닝 기법을 사용하여 전처리 단계에서 이미지를 분류하고 분류된 이미지에 제약조건을 적용시켜 알고리즘의 처리시간을 줄이고 정확도를 향상시켰다. 실험 결과 본 논문에서 제안하는 MCS-RANSAC이 DCS-RANSAC 알고리즘에 비해 수행시간이 약 6% 단축되었고 오차율은 약 15% 줄어들었으며 참정보 비율은 2.8% 증가한 것으로 확인되었다.

Recommendation Systems apply social relationship to enhance recommendation performance and allow users to get reliable information. This paper propose an enhanced user modeling based on link attributes of the social network. For this, similarity values for users in terms of their item preferences are calculated. A relationship-centric neighbor group is identified using social relationships explicitly presented by users, and these social relationships are represented by links considering information on both direct and indirect relationships. These link attributes consist of the link degree and strength. The link degree is identified by using six degrees of Kevin Bacon and is represented through numeric values for relationships between users. The link strength is calculated by considering the number of directly and indirectly connected relationships with the target user and is applied to the proposed user-modeling method. The experimental results demonstrate that the proposed recommender system using the user-modeling method based on link attributes shows better prediction accuracy and recommendation quality than benchmark systems, suggesting that user modeling based on link attributes in a recommender system can enhance the usefulness of recommendations.

In computer vision especially in image processing, it has become popular to apply deep convolutional networks for supervised learning. Convolutional networks have shown a state of the art results in classification, object recognition, detection as well as semantic segmentation. However, supervised learning has two major disadvantages. One is it requires huge amount of labeled data to get high accuracy, the second one is to train so much data takes quite a bit long time. On the other hand, unsupervised learning can handle these problems more cheaper way. In this paper we show efficient way to learn features for classification in an unsupervised way. The network trained layer-wise, used backpropagation and our network learns features from unlabeled data. Our approach shows better results on Caltech-256 and STL-10 dataset.

Abstract—We describe a new representation for multiple articulated sport players for court player detection. Instead modeling object into deformable parts templates or mixtures of small parts which just capture local appearance of parts and spatial relations between parts, our proposed model trained with local part information with global constraint by a structured SVM can capture not only such local appearance and spatial relations above, but also the semantic relations between body parts which are the critical factors for precisely detecting objects and pose estimation. Our approach has several novel properties: (1) we adopt typical articulated partbased model with global appearance constraint to control trade-off between recall and precision for detection (2) we incorporate semantic knowledge about various articulation of court players into our mixture-of-parts model, these semantic knowledge are popularly used for pose estimation (3) after the root (global) and part (local) bounding boxes are predicted by our system, we train a linear least-squares regression model to output the final detection results which yields considerable improvements in performance. In our experiments with very challenging APIDIS basketball dataset and standard INRIA person dataset, it indicates that our detection system achieves state-of-art performance compared to previous works.

Convolutional Networks (ConvNets) are powerful models that learn hierarchies of visual features, which could also be used to obtain image representations for transfer learning. The basic pipeline for transfer learning is to first train a ConvNet on a large dataset (source task) and then use feed-forward units activation of the trained ConvNet as image representation for smaller datasets (target task). Our key contribution is to demonstrate superior performance of multiple ConvNet layer features over single ConvNet layer features. Combining multiple ConvNet layer features will result in more complex feature space with some features being repetitive. This requires some form of feature selection. We use AdaBoost with single stumps to implicitly select only distinct features that are useful towards classification from concatenated ConvNet features. Experimental results show that using multiple ConvNet layer activation features instead of single ConvNet layer features consistently will produce superior performance. Improvements becomes significant as we increase the distance between source task and the target task.

In question-and-answer (QA) systems, various queries need to be processed. In particular, those queries such as temporal information require complex query generation processes. This paper proposes a temporal representation model that can support qualitative and quantitative temporal information on historical ontology by applying the concept of ordinals and sets and introduces operators that allow a QA system to easily handle complex temporal queries. To verify the effectiveness of the proposed model and operators, historical scenarios are presented to show that they can effectively handle complex temporal queries.

No Data

RANSAC 알고리즘은 컴퓨터 비전 분야에서 호모그래피 행렬을 추정하는데 많이 사용되고 있다. CS-RANSAC 알고리즘은 RANSAC 알고리즘에 제약조건을 설정하여 정확도를 높인 알고리즘이지만 샘플링 단계에서 정확한 호모그래피를 추정하는데 불필요한 특징점을 선택하여 알고리즘의 효율성을 저하시키는 경우가 있다. 따라서 본 논문에서는 Symmetric Transfer Error로 특징점이 참정보인지 평가하고 불필요한 특징점을 필터링하여 CS-RANSAC 알고리즘의 속도와 정확도를 증가시키는 방법을 제안한다. 실험은 제안하는 알고리즘의 수행시간과 오차율을 비교하였고, 실험 결과 본 논문에서 제안한 방법이 기존 CS-RANSAC 알고리즘보다 수행시간이 평균적으로 약 5% 단축되었고 정확도는 약 14% 향상 되었다.

Multi-day trip itinerary planning is complex and time consuming task, from selecting a list of worth visiting POIs to arranging them into an itinerary with various constraints and requirements. In this paper, we present CYTRIP, a multi-day trip itinerary planning system that engages human computation (i.e. crowd recommendation) to collaboratively recommend POIs by providing a shared workspace. CYTRIP takes input the collective intelligence of crowd (i.e. recommended POIs) to build a multi-day trip itinerary taking into account user’s preferences, various time constraints and locations. Furthermore, we explain how we engage crowd in our system. The planning problem and domain are formulated as AI planning using PDDL3. The preliminary empirical experiments show that our domain formulation is applicable to both single-day and multi-day trip planning.

No Data

No Data

No Data

A homography matrix is used in computer vision field to solve the correspondence problem between a pair of stereo images. RANSAC algorithm is often used to calculate the homography matrix by randomly selecting a set of features iteratively. CSRANSAC algorithm in this paper converts RANSAC algorithm into two-layers. The first layer is addressing sampling problem which we can describe our knowledge about degenerate features by mean of Constraint Satisfaction Problems (CSP). By dividing the input image into a grid and making feature points into discrete domains, we can model the image into the CSP model to efficiently filter out degenerate features. By expressing the knowledge about degenerate feature samples using CSP in the first layer, so that computer has knowledge about how to skip computing the homography matrix in the model estimation step for the second layer. The experimental results show that the proposed CS-RANSAC algorithm can outperform the most of variants of RANSAC without sacrificing its execution time.

Well-known media applications of Chromecast such as YouTube, Viki usually use a time slider on the second screen for video navigation. However, the time slider does not offer easy way to access particular contents. Here we propose an aligned thumbnails-based video browsing system with Chromecast using content-based video browsing approach by automatically detecting scenes. It uses hierarchical sliding interface to browse the playing video on TV easily. Through the experimental results, our proposed system helps search specific contents quickly.

This research work designs an interactive and contextual visualization tool called ConTag (Contextual Tag) for video-based learning scenarios. The purpose of this study is to better understand the variation of learners’ interactions and collaborative learning processes supported by ConTag. The system consists of i) contextual tags, ii) word clouds, and iii) transcript pointers. The paper examines how learners participate in and interact with ConTag and evaluates how they perceive their experience. The results indicate that learners actively participated in discussions and took good advantage of ConTag. Learners' interactions included self-reflection, elaboration, internalization, and support, and they showed moderately positive attitudes toward ConTag participation. The implications of using ConTag and its social collaborative tool (Twitter API) are discussed in the video-based learning context.

In recent years, the internet use has led to rapid increases in multimedia resources. In this regard, handling multimedia data has become an important issue in information retrieval fields, and multimedia indexing plays an essential role in handling multimedia data. Textual metadata plays an important role in describing and analyzing multimedia resources. However, because only few multimedia resources provide textual metadata, there is a need for methods to generate the metadata. This paper proposes an automatic method for indexing cooking videos using textual data. The proposed method consists of web crawling for cooking videos and food recipes, the extraction of time-stamped text captions, and caption-recipe alignment using a dynamic programming technique. With information on captionrecipe alignment, automatic indexing is performed for cooking videos. The experimental results show that the proposed approach achieves highly accurate indexing. The user interface allows users to easily move to specific pinpoints with indexed videos.

Aircraft maintenance and training play one of the most important roles in ensuring flight safety. The maintenance process usually involves massive numbers of components and substantial procedural knowledge of maintenance procedures. Maintenance tasks require technicians to follow rigorous procedures to prevent operational errors in the maintenance process. In addition, the maintenance time is a cost-sensitive issue for airlines. This paper proposes intelligent augmented reality (IAR) system to minimize operation errors and time-related costs and help aircraft technicians cope with complex tasks by using an intuitive UI/UX interface for their maintenance tasks. The IAR system is composed mainly of three major modules: 1) the AR module 2) the knowledge-based system (KBS) module 3) a unified platform with an integrated UI/UX module between the AR and KBS modules. The AR module addresses vision-based tracking, annotation, and recognition. The KBS module deals with ontology-based resources and context management. Overall testing of the IAR system is conducted at Korea Air Lines (KAL) hangars. Tasks involving the removal and installation of pitch trimmers in landing gear are selected for benchmarking purposes, and according to the results, the proposed IAR system can help technicians to be more effective and accurate in performing their maintenance tasks.

인터랙티브 비디오는 일반적인 비디오와 클릭 가능한 객체를 동시에 제공하여 사용자와 상호작용이 가능한 비디오를 말한다. 이러한 인터랙티브 비디오를 제작하기 위해서는 클릭 가능한 객체의 정보(이미지, 하이퍼링크, 위치 등)를 일반적인 비디오 위에 어노테이션 하는 과정을 수행하게 된다. 일반적으로 어노테이터는 전체 어노테이션 과정 중에서도 특정 프레임에 어노테이션 될 클릭 가능한 객체를 특정 위치에 어노테이션 하는데 많은 시간을 소비하고 있다. 본 논문에서는 이러한 문제를 해결하기 위해 어노테이터가 최소한의 시간을 소비해서 효과적으로 어노테이션 할 수 있는 얼굴 영역기반의 인터랙티브 비디오 어노테이션 시스템을 제안한다.

No Data

No Data

No Data

스마트폰, 스마트TV, 구글 크롬캐스트 등 스마트 미디어의 발달로 인해 정보와 사용자간의 상호작용의 중요성이 강조되고 있고 사용자들은 자신이 원하는 정보를 다른 단계를 거치지 않고 쉽고 빠르게 얻기 위한 요구사항이 증가하고 있다. 그러나 아직 여러 분야에서 사용자의 요구사항에 대한 해결책이 매우 부족하다. 본 논문에서는 요리분야의 요리비디오에서 이러한 문제점을 발견하여 이를 개선하기 위한 방안을 살펴보았다. 요리비디오는 단방향으로 정보를 제공하기 때문에 사용자가 요구하는 정보를 얻기 위해서는 웹 검색과 같은 별도의 과정이 필요하고, 또한 요리를 하는데 있어 불편함을 겪어야 한다. 이러한 불편함을 해소하기 위하여 요리 비디오에서 사용자가 원하는 장면에 정확한 정보를 제공하고 요리를 하면서 효율적으로 인터랙션을 하기 위한 시스템을 구현하였다. 시스템 평가를 위해 HCI분야의 연구에서 제안되어 많은 논문들이 평가 척도로 사용하는 사용자 설문방식을 기반으로 시스템을 평가하기 위한 실험을 통해 시스템 인터페이스, 전체 시스템의 만족도, 정보의 정확성에서 좋은 결과를 얻었다.

In this paper, we proposed a method to enable the use of shared interactive video in order to promote collaborative environment in authoring non-linear video process. Our proposed method addressing a problem that occurs when applying nested non-linear flow in the shared interactive video. The system enables the collaboration of authors to use existing interactive videos and gives them full control to apply the non-linear flow on top of each video. The system can employ collaborative manner in authoring non-linear interactive video.

뉴스 콘텐츠의 많은 이용자들은 단순히 뉴스를 이용만 하는 것이 아니라 SNS를 통해 해당 뉴스를 다시 다른 이용자에게 전달한다. 뉴스 서비스들은 이런 이용자들을 위해 뉴스를 카테고리, 태그 등으로 분류하고 분류된 정보를 제공한다. 하지만 이런 분류 방식은 뉴스의 세부적인 내용을 나타내지 못해 뉴스 전달시 이용자 간의 컨텍스트 공유에 어려움이 있다. 이를 보안하기 위해 본 논문에서는 세부내용들의 키워드와 비디오에서의 스토리가 시작되는 부분을 함께 제공하는 프레임워크를 제안한다. 뉴스 대본을 스토리에 따라 분할한 뒤 세부 내용의 키워드를 추출하고 분할된 뉴스 대본과 비디오를 동기화 시켜 비디오 시작지점을 찾는다. 이 세부 내용의 키워드와 시작지점을 태그와 함께 사용자에게 제공한다면 이용자가 SNS에서 뉴스 전파 및 의견 공유할 때 컨텍스트 공유가 보다 원활해 질 것이다. 그리고 정확한 키워드, 시작지점 추출을 위해서는 정확한 스토리 분할이 요구되는데 이를 위해 이를 위해 의미적 유사도 기반의 스토리 분할 방법과 벤치마크 알고리즘을 이용한 방법의 성능 비교 실험을 진행하였고 의미적 유사도 기반의 스토리 분할 방법이 더 높은 성능을 보였다.

No Data

This paper presents our proposed method to automatically generate topical video content. We exploiting video transcripts to generate Conceptual graph and Tag clouds visualization, known as “ConTag” or Contextual Tag. The system is able to provide users with contextual information according to the video content. To achieve this we employed popular text mining algorithm, TF-IDF model and phrase association algorithm, Apriori algorithm. In this work, we also solved passive learning problem by facilitating users with interactive conceptual visualization system and communication mechanism over Social Networking Service (SNS) which accompanied by contextual and temporal information correspond to the particular time and scene from the video.

Social network is a huge communication platform for providing people to connect with one another and to bring users together to share common interests, experiences, and their daily activities. Relationship of each user in social network known as social relationship has been used for modeling to help improving effectiveness of recommendation in existing collaborative filtering (CF) method. In this paper, we propose a model that integrates social relationship discovered from user’s behavior with traditional CF in order to increase performance and prediction accuracy of recommender system. We identify user’s relationship by using user’s behavior with their friends such as posts and comments in Facebook. The experimental results have shown that our proposed model improves the accuracy of prediction in terms of MAE and performance in terms of Precision and Recall compared with traditional CF.

Recommender systems play an important role in online business ecosystem, especially to recommend users’ new items. The most critical problem in the recommender systems is providing high accuracy of recommendation to new users which lack of preference to compute similarity between users. In this paper, we propose a recommender system to solve the cold start problem by combining traditional collaborative filtering of users’ rating preference and the users’ genres interest that derived from SNS. First we compute users’ similarity according to their rating on movies. Second we also compute the users’ similarity from genre interest extracted from SNS. We combine these both similarities information in order to produce new user’s similarity. Our experiment results show that our approach is outperform in cold start problem compared to traditional collaborative filtering.

Cloud computing is an emerging technology that allows users to utilize on-demand computation, storage, data and services from around the world. The main contribution of this work is to propose a new multi-objective genetic algorithm coupled with the viral infection capable of handling hard constraints, such as restrictions on task scheduling, which are not handled by current algorithms. Furthermore, our algorithm can optimize any number of parameters such as execution time, cost, reliability, and availability; In addition, it can handle restrictions such as deadlines and requirements on the different variables. Using data of the Amazon EC2 cloud resources and workflows from London e-Science Center, we have been investigating the problem of scheduling workflow applications and have compared our algorithm with other scheduling algorithms. Our experimentations have shown the efficiency of our algorithm and have confirmed that the viral infection operator is a powerful tool when solving hard constraints.

Collaborative filtering (CF) is one of the most widespread methods for recommendation approach. It has been used efficiently to help users to find items that they should appreciate, by identifying users that can be characterized as “similar”, using a similarity measure based on items rating data. However, traditional CF approaches have shown limitation occasioned by problems such as sparsity and cold start. This situation can cause a user to stop using a system due to lack of accuracy in produced recommendations. This paper suggests another source of information, item taxonomies, which have become extremely common among recommender systems for products classification in diverse domains. We combine such knowledge with an inference process on the taxonomy hierarchy to propose a new similarity measure, which focuses on improving recommendations under cold-start conditions. Experiments made on the MovieLens dataset indicate that our method improves results when applied to cold start situations.

TV-commerce is a new form of shopping that allows consumer to view, select and buy products from Smart TV. To do so, sellers annotate videos and associate it with information from online e-commerce systems in a semantic manner. In this work, we propose an e-commerce information derivation mechanism for video annotation using Linked Open Data (LOD) with faceted search. Annotation information is derived from e-commerce LODs, which linked distributed data across e-commerce web. We incorporated faceted search to allow consumer to easily make a information derivation query defined by GoodRelations ontology. The derived information is displayed as a faceted graph facilitating information choice.

User profile is an essential component for accessing the personalized information from the web. Efficiency of personalized accessed information highly depends on how to model the user details to construct user profile. Previously, user profile was constructed by collecting list of keywords to inferring user interests. These kinds of approaches are not sufficient for many applications. In this paper, we have proposed a new method for constructing semantic user profile for personalized information access. User’s query is extended using ontological profile for generation personalized search context. Experimental results show that our method of constructing semantic profile is effective for searching information with individual needs

Traditional recommendation systems provide appropriate information to a target user after analyzing user preferences based on user profiles and rating histories. However, most of people also consider the friend’s opinions when they purchase some products or watch the movies. As social network services have been recently popularized, many users obtain and exchange their opinions on social networks. This information is reliable because they have close relationships and trust each other. Most of the users are satisfied with the information. In this paper, we propose a recommendation system based on advanced user modeling using social relationship of users. For the user modeling, both direct and indirect relations are considered and the relation weight between users is calculated by using six degrees of Kevin Bacon. From the experimental results, our proposed social filtering method can achieve better performance than a traditional user-based collaborative filtering method.

This paper proposes an interactive semantic video creation based on an augmented reality environment. This approach relies on Intelligent Augmented Reality (IAR) [5], which was developed for handling complex tasks in aircraft maintenance. The approach consists of two main modules, Augmented Reality (AR) Module and Knowledge Base System (KBS). The system was extended using an interactive semantic video ontology to express a moving object and semantic information. This system is called an iSEE (Interactive SEmantic vidEo) Player. In addition, it showed better cost performance with Cubic Spline Interpolation [15], which generates an interactive semantic video system from an IAR context manager.

증강현실 콘텐츠를 2D 이미지기반으로 저작할 때, 작성된 증강현실 콘텐츠를 카메라 시점과 일치시켜 합 성하기 위해 호모그래피를 이용한다. 이때 증강현실 콘텐츠를 이질감 없이 합성하기위해 정확한 호모그래 피 행렬을 추정해야 한다. 그러나 호모그래피 행렬 추정 시 사용되는 특징점들이 선형을 이루거나, 특정 영역에 군집을 이루는 경우 정확한 호모그래피 행렬을 추정하지 못하는 문제가 발생한다. 본 논문에서는 이러한 문제를 해결하기 위해 선형제약, 거리제약을 적용한 CSP 기반의 RANSAC 알고리즘을 제안한다. 실험결과 호모그래피 행렬 추정 시 CSP를 적용한 RANSAC 알고리즘이 기존의 랜덤샘플링 또는 삼각형의 넓이를 이용한 샘플링을 적용한 RANSAC 알고리즘보다 정확도가 향상됨을 보였다.

With the emergence of the Smartphone, people can use Online Social Network Services ubiquitously, leading to a significant increase of the number of participants in online social networks. Under these circumstances, online users will require an intelligent and intuitive social relationship management system such as the ontology-driven browsing method. In this paper, to build a user-centered semantic social network and to represent entities and relationships with ontology to improve retrieval performance of the semantic social network, we will design our ontology extended from FOAF, RELATIONSHIP and propose a new method to compute closeness among friends using resources on social networks. Furthermore, we evaluate our ontology-driven browsing on via implementing a prototype system. Our proposed method concerning closeness can achieve 85% precision.

In this paper, we present a mobile collaborative outdoor augmented reality system. The system will provide a user who captures image of a building facade with related introduction. The proposed system mainly consists of two major parts - annotation module and AR browsing module. Annotation module works with existing standards and linked data to allow users create annotations of the building. Annotation data in our system which only contains URI and metadata, such cluster ID, cluster centroid and the image feature vectors. When user using the system browsing the landmark in outdoor, the system first recognize the landmark and calculate the camera pose from the matching, and then the related introduction is retrieved from linked data through its URI. Experiments show that the proposed augmented reality system is efficient and robust in many cases.

Many online shoppers tend to seek for the opinions of early adopters before making a purchase decision to reduce the risk of buying a new product. Social network provides a very suitable environment for user to share as well as to absorb the most relevant information among friends. However, within a very huge graph of friends, a task of determine who are the most appropriate persons for user to seek opinions on a specific product is yet a limitation within the current approach. We propose a method of defining friends who are the most appropriate persons for user to discuss with. Through Facebook data, user profile is generated in order to determine user’s interest and knowledge about a specific product. Our method has been empirically tested by a case study in order to evaluate the results generated by our system. From the results of user case study, our system can effectively filter the most appropriate friends that user can discuss with on a specific product.

지난 11월 20일부터 시행 중인 청소년 보호법이 요즘 이슈가 되고 있다. 이는 그것은 바로 여성 가족부에서 추진 중인 ‘게임 셧 다운제’이다. 청소년들의 수면권과 학습권을 보호하자는 것인데, 학교에 다니는 청소년들이 방과 후 학원을 거쳐 늦은 시간에 귀가하면, 곧바로 수면이 아닌 온라인 게임을 하느라 제대로 된 휴식을 취하지 못한다는 것이다. 게임 셧 다운제는 만 9세이상 16세 미만의 청소년을 자정부터 다음날 오전 6시까지 온라인 게임에 접속을 차단하는 것인데, 기존에 있는 정책들과 대립되는 문제가 있다. 본 논문에서는 게임 셧 다운제의 문제점에 대해 연구하는 것을 목적으로 한다.

네트워크 기술의 발전으로 모바일 기기 콘텐츠 서비스의 일종인 N-Screen이 활발해 지고 있다. 이러한 N-Screen 서비스는 TV, PC 그리고 모바일 기기(스마트폰, 태블릿)등의 다양한 스크린에 소비되는 콘텐츠를 제공하는 것이다. 그러나 현재 대부분의 서비스들은 콘텐츠를 소비하는 기기에 맞게 인코딩 된 다수의 콘텐츠를 보유하고 있어야한다. 하지만 현존하는 다양한 디스플레이 규격과 코덱에 적합하도록 콘텐츠를 변환하여 보관하는 것은 불가능한 상황이다. 따라서 현재 대부분의 서비스들은 대표적인 일부의 장비에 대한 콘텐츠만을 제공하고 있다. 또한 다양한 규격의 기기에 맞게 콘텐츠를 변환하여 제공하기 위해서는 고도의 컴퓨팅 파워가 요구된다. 따라서 본 논문은 스마트 TV에 그리드 컴퓨팅 기술을 적용하여 다양한 규격의 기기에 N-Screen 서비스를 제공하는 기술을 제안하고 설명한다.

In this paper, we propose an effective learning method using virtual reality and ontology technologies for appreciation activity in the art education. Appreciation activity, one of the learning methods, indicates sensory, aesthetic understanding through watching and enjoying art. However, current art education methods cannot support a variety of learners’ needs and interests. For watching the art, the learner should visit a museum or an art gallery. And some famous art works may not be exhibited publicly. To solve these spatial and temporal issues, therefore, we propose a new learning method in the art using virtual reality and ontology. In virtual Reality, learners can enjoy the art in their computer. And ontology provides useful information to learners. In this paper, we model the ontology and use Google Art Project for virtual reality.

Most previous studies of ontology integration have simply involved blind or exhaustive matching among all concepts across ontologies. Therefore, the computational complexity of integrating two ontologies is O(n2). In addition, semantic mismatches, logical inconsistencies and conceptual conflicts in ontology integration have not yet become avoidable. The main contribution of the approach presented here is to reduce the computational complexity and to enhance the accuracy of ontology integration. The key idea of this approach is to start from an Anchor (two matched concepts) to work towards a collection of matched pairs among its neighboring concepts by computing similarities between the ``priorly

협업적 여과 시스템은 사용자에 대한 클러스터링을 구축한 후, 구축된 클러스터를 기반으로 사용자에게 영화를 추천한다. 하지만 사용자 클러스터링 구축에 많은 시간이 소요되고, 사용자가 평가한 영화가 피드백이 되었을 경우 재구축이 쉽지 않다. 본 논문에서는 사용자 클러스터링의 재구축을 용이하게 하기 위해 빈발패턴 네트워크를 이용하여 클러스터링을 구축하고, 이를 협업적 여과 시스템에 적용하여 영화를 추천한다. 구축된 클러스터를 통해 사용자 클러스터를 재구축시 소요되는 시간 비용을 줄이면서, 전통적인 협업적 여과 시스템과 유사한 성능의 추천이 가능하게 되었다.

본 논문은 빠르게 변화하는 시대 흐름에 따라 더욱 다양해진 학습자의 요구와 흥미를 충족하기 위한 미술교육프로그램으로 가상현실을 이용한 구글 아트프로젝트와 온톨로지를 이용하여 보다 효과적인 미술교육 방법을 제시한다. 가상현실을 통해 학습자는 예술작품을 실사로 본 것과 같이 체험을 할 수 있으며, 온톨로지를 통해 해당 작품에 관련된 정보을 바탕으로 학습자에게 구성주의적 학습을 유도할 수 있다. 이를 통해 제한적인 미술 교육에 대한 대안을 마련할 수 있으며, 최첨단의 멀티미디어를 활 용한 시각문화 학습에도 도움이 될 것으로 기대한다.

최근 이러닝과 같은 서비스로 인해 동영상 형태의 멀티미디어 소비가 증가함에 따라 사용자들은 자신이 원하는 멀티미디어를 찾거나 멀티미디어에 등장하는 객체의 정보를 브라우징 하고자 하는 요구사항이 증대되고 있다. 이러한 사용자들의 요구사항을 충족시키기 위해서는 노동집약적인 어노테이션 작업이 불가피하다. 특히 객체 기반 어노테이션에서는 수많은 연구자들이 객체 검출 및 인식, 트래킹 등의 컴퓨터 비전 기술을 적용한 자동 어노테이션을 시도하고 있으나 동영상 내의 다양한 객체를 다루기에 어려움이 있다. 또한 기존 키워드 기반 어노테이션은 객체의 모든 관련 정보를 개별적으로 관리해야 하므로 대용량 데이터 관리와 브라우징에 대한 어려움이 따른다. 이러한 문제를 극복하고자 본 논문에서는 시맨틱 데이터에 접근하기 위해 링크드 데이터를 이용하여 다수의 어노테이션 수행하는 사용자들이 협업적으로 객체에 대한 어노테이션을 하고, 이 후 서비스 이용자들이 시맨틱 어노테이션을 통해 효율적인 브라우징을 할 수 있도록 하는 시스템을 제안한다.

Technical manuals are very diverse, ranging from manuals on software to manuals on commodities, general instructions and technical manuals that deal with specific domains such as mechanical maintenance. Due to the vast amount of manual information finding the necessary information is quite difficult. In case of electronic maintenance manuals currently used by companies, mechanics should search for the related information to accomplish their tasks. And it is difficult to grasp relationships among contents in manuals. Search process is time-consuming and laborious for mechanics. Many researchers have adopted ontology to solve these problems and semantically represent contents of manuals. However if ontology becomes very large and complex, it is not easy to work with ontology. Visualization has been an effective way to grasp and manipulate ontology. In this research, we model new ontology to represent and retrieve contents of manuals and design the visualization system based on proposed ontology. To model ontology, we analyzed aircraft maintenance process, extracted concepts and defined relationships between concepts. After modeling, we created instances of each class using technical manuals. Our system visualizes related information so that mechanics can intuitively grasp the information. This allows workers to easily get information for given tasks and to reduce their time to search related information. Also, related information can be understood at a time through visualization.

We propose a novel T-commerce model based on semantic interaction, as the TV broadcasting environment is moving to interactive digital TV such as IPTV, Hybrid TV, and Smart TV. The purpose of our work is to design and implement the system of our new T-commerce model. Thus, our system applies spatial relation rules and ontology for object recognition if a viewer is interested in one on the screen of TV program such as drama, movie, and news. All object information needs to be annotated in a TV program to identify the user selection. However, the effort must be reduced, as it costs too much. The program can be identified as an annotated item, a combination of known base area and spatial relations to the user selection, even though it does not have the exact position. The TV system can semantically interact with the user via this combination. A prototype system, semantic interaction TV, has been implemented to achieve this semantic interaction. Our case studies show that the proposed system is useful to T-commerce.

Ongoing development of information technology provides a fundamental environment of maintaining vast manual information in digital form. Demand of digital information increases because of easy handling and presentation of documents on various structure manner. As a result, technical documents like mechanical device maintenance manuals have numerous advantages to move from page version into electronic edition. However, it is still difficult for context-aware systems to provide information relevant to specific context because the information scatters in multiple parts of the document. The object detection techniques of context-aware systems could not recognize all objects completely. In this paper, we propose a context-aware system which can recommends relevant information through analyzing context delivered by a vedio device located in workplace around. To compare the current context with documents, three metrics are used for computing similarities between contextual information and keywords extracted from documents. For our experiments, we preprocessed hundreads of TASKs in the aircraft’s maintenance manual and made several cases for context. Our experiments showed that our proposing system could provide information related with context.

No Data

Experts finding is an important issue for finding potential contributors or expertise in a specific field. In scientific research, researchers often try to find an experts list related to their interest areas to acquire the knowledge about state arts of current research and novices can get benefit to find new ideas for research. In this paper, we proposed an ontological model to find and rank the experts in a particular domain. First, an Academic Knowledge Base(AKB) is built for a particular domain and then an academic social network (ASN) is constructed based on the information provided by the knowledge base for a given topic. In our approach, we proposed a cohesive modeling approach to investigate academic information considering heterogeneous relationship. Our proposed model provides a novel approach to organize and manage the real world academic information in a structural way which can share and reuse by others. Based on this structured academic information an academic social network is built to find the experts for a particular topic. Moreover, the academic social network ranks the experts with a ranking scores depending upon relationships among expert candidates. Finally, we verify the experimental evaluations of our model which improve precision of finding experts compare to baseline methods.

No Data

With the increasing interest in multimedia annotation, emotion annotation is being recognized as an essential resource which can be applied for a variety of purposes, including video information retrieval and dialogue systems. Here we introduce an automatic emotion annotation schema for dialogues, which could be used for the retrieval of specific scenes in film. Distinguished from previous works, we apply a new approach using the hypernym/hyponym relations and synonyms of WordNet, which enables us to organize a novel set of emotional concepts and to automatically detect whether a specific emotional word is associated with a specified emotional concept through measuring the conceptual distance between them.

In digital era, people tend to maintain personal photos in digital spaces not only to share their experiences with social friends but also to jog their own memory. Therefore, an effective solution is crucial to the growth of the needs for recording one’s daily life. In this study, we have developed a complete system for personal photo diary system, namely MePTroy. With a friendly user interface, users can easily maintain personal episodes and memories with photos. In addition, MePTroy also support a flexible method for photo search based on the position of facial appearance that enables users to access episodes quickly. By integrating face detection and recognition technologies, as well as a friendly UI, MePTory offers diverse functionalities to annotate and search photos.

A movie, a typical multimedia data, includes several textual information such as subtitles and annotation texts as well as audio-visual information. In general, a story of a movie is progressed by dialogues of characters and the dialogues are produced based on a pre-written script that contains detailed descriptions of all the information related to the movie. The script consists of scene headings, character names, dialogues, actions, and so on. Rich information related to the movie can be extracted easily and accurately from each component of the script. However, since the script does not contain time information, additional processes are required to synchronize the script with the movie. The subtitle is usually made from the script’s dialogue and has time information to display text on screen when the characters speak. Therefore, synchronization between the script and the movie can be accomplished through an alignment process between the script’s dialogue and its’ subtitle. In this paper, we propose a unique method of scene boundary detection exploited script-subtitles alignment. By synchronizing the script to movie, we identify scene boundaries with semantic level. In addition, we build a thesaurus to enhance alignment performance. Experimental results with real movies show that the proposed method can successfully, not only align scripts with those subtitles, but also detect scene boundaries.

장편 드라마, 영화 등 스토리의 호흡이 긴 콘텐츠를 제작함에 있어 스토리의 인과율 문제는 큰 이슈가 되고 있다. 특히 인터랙티브 스토리텔링이 요구하는 비선형적, 동시적 서사는 작가가 기억해야 하는 사실 정보를 폭증시키고 그에 따라 논리적 모순이 생길 가능성이 높아지고 있다. 이러한 문제를 극복하기 위하여 본 논문에서는 이미 만들어진 콘텐츠로부터 사실 정보를 추출, 분석하여 작가의 기억을 보조하고 작품의 구조적 오류를 방지해주는 캐릭터 이력 추적 시스템을 제안한다.

No Data

No Data

No Data

생산자로부터 고객에 이르는 상품 공급 활동을 관리하는 경영기법인 공급사슬관리(SCM)는 기존의 폐쇄적이고 정적인 환경에서 개방적이고 동적인 환경으로 변화하고 있다. 이러한 변화에 대응하기 위한 실험이나 연구는 부족한 상황이다. TAC SCM은 동적인 SCM 환경에서 다른 경쟁자들과의 경쟁을 통하여 다양한 SCM 전략을 시험할 수 있다. 본 연구에서는 실시간 SCM 환경에서 능동적 할인 전략을 적용한 에이전트를 설계 및 구현하였다. 능동적 할인은 입찰 성공률에 따라 할인율을 변경하여 적용하여 항상 일정한 생산량을 보유할 수 있도록 한다. 본 연구의 실험 결과에서 제안한 에이전트가 서버에서 제시한 Dummy 에이전트보다도 높은 수익과 안정된 생산을 할 수 있음을 보였다.

디지털 기술의 발달로 기존의 콘텐츠를 소비하는 역할을 했던 대중들이 손쉽게 콘텐츠를 제작하고 향유할 수 있게 되었다. 이러한 현상으로 최근 디지털스토리텔링이 큰 화제가 되어 많은 곳에서 공모전 등을 통해 웹 환경에서 콘텐츠를 생산하고 있다. 디지털 스토리텔링은 디지털 기술을 활용하여 만들어진 스토리텔링 콘텐츠로 웹 환경에서 다양한 형태로 제작되고 있다. 하지만 현실은 단순한 저작 도구의 환경이 아날로그에서 디지털로의 변화일 뿐 집단지성을 표방한 웹 2.0 환경에는 동 떨어져 있다. 따라서 본 논문에서는 웹 환경에서 집단지성을 적용하여 협업적으로 제작되는 디지털 스토리텔링 콘텐츠에서 나타나는 절차 및 필요한 요소를 분석하고 이에 맞는 플랫폼을 제안한다.

본 연구는 차량 경로 스케줄링 문제(VRSPTW, the Vehicle Routing and Scheduling Problem with Time Window)를 해결하기 위하여, 멀티 비용 함수(Multi Cost Function)를 갖는 개미 군집 최적화(Ant Colony Optimization)을 이용한 휴리스틱을 제안하였다. 멀티 비용 함수는 각 개미가 다음 고객 노드로 이동하기 위해 비용을 평가할 때 거리, 요구량, 각도, 시간제약에 대해 서로 다른 가중치를 반영하여 우수한 초기 경로를 구할 수 있도록 한다. 본 연구의 실험결과에서 제안된 휴리스틱이 Solomon I1 휴리스틱과 기회시간이 반영된 하이브리드 휴리스틱보다 효율적으로 최근사 해를 얻을 수 있음을 보였다.

Collaborative tagging is a popular method in social resource sharing system. The reason why collaborative tagging system becomes popular is users can freely tags on the resources, like Flickr1 and Delicious2. Any relationship between tags and resources are not made explicitly when tagging in folksonomy. Due to lack of semantic meaning, it is difficult for users to find related resources. Also, short of sufficient number of relevant tags increase the data sparsity which decreases the rate of information extraction against the user queries. Defining semantic relationship between tags, resources and users is an important research issue to retrieval of related information from folksonomies. In this paper, we propose a WordNet based method to find the semantic relationship between tags in folksonomy. Finally, the experiment results using Flickr data show our proposed method works well finding semantic relationship between tags in folksonomy.

Vehicle Routing Problems (VRP) arise in many real-life applications within transportation and logistics. This paper describes a novel tabu search heuristic methodology for the Muilt-Trip Vehicle Routing Problem with Time Windows (MTVPTW). Unlike the classical vehicle routing problem, the MTVPTW permits vehicles to make multiple trips and from a central depot within a given time duration to serve (pick up/ delivery) a number of customers, each customer must be done within a time window; the drivers’ schedules must respect the maximum legal travelling time and minimum legal waiting time; the unloading time is considered. We present results of our algorithm on standard data sets such as solomon data and outline plans to improve upon these solutions.

현대 사회에서 가정의 난방과 산업의 연료 등으로 사용되는 가스 시설은 우리에게 편한 생활환경을 제공하지만, 사용자의 취급부주의로 인한 가스 폭발 사고는 막대한 피해를 초래한다. 현재 가정 등에서 가스 사고에 대처하기 위한 유일한 수단으로 가스 안전 기기들을 사용하고 있지만, 사용자 부재 시에는 대처 방안이 전무하다. 본 논문은 실내의 상황 정보를 지속적으로 수집 및 관리하여 일정 수준 이상의 가스 농도가 검출되면 규칙 기반의 추론을 통해 상황인식 서비스를 제공하는 홈 네트워크에서 가스 누출 상황인식 시스템을 제안한다.

Most previous research on ontology integration has focused on similarity measurements between ontological entities, e.g., lexicons, instances, schemas and taxonomies, resulting in high computational costs due to the need to consider all possible pairs between two given ontologies. In this chapter, we present a novel approach to reduce computational complexity in ontology integration. Thereby, we present an enriched ontology model for a formal ontological analysis that enables us to build ontologies with a clean and untangled taxonomic structure. The importance and types of concepts, for priority matching and direct matching between concepts, respectively are proposed. Identity-based similarity is computed, which enables us to avoid comparisons of all properties related to each concept, while matching between concepts. The problem of conflict in ontology integration has initially been explored on the instance-level and concept-level. This is useful to avoid many cases of mismatching.

There is a huge amount of indexed web resources are used by millions of users in every country the Earth. It is this huge amount of the content and the integration of information across space and time that leads to the Web containing itself a social network. In this pape, we proposed a frame- work for construction and visualization of a social network by collaborative web approach, in which a consensus method is proposed to identify personal information from different web search spaces (include personal/ organizational sites, blogs, publications) of user. A personal ontology is automatically generated from the personal information. Construction of social network is carried out by analyzing relationship among people, contents and organizations found from the different search spaces of each target user.

In this paper, we introduce our proposed framework for visualization of the blog’s content using ontology method to let blogger know the existing relationship between the blogs. The proposed framework is to explain the details on how the important information from the blog such as category, friend and content will be extracted, how the ontology will be implemented in this system and how visualization framework will be done. From this proposed visualization, we highly expect to provide an interactive tool for information retrieval where blogger can have multiple ways to catch the information from the favorite blogs.

The main goal of this research is to investigate the techniques that implicitly build ontology-based user profiles. In particular, automatically building profiles based on user’s information (blogs, publications, home pages,...) generated from the internet. We proposed a framework to search the user details to build profile automatically. An initial profile is constructed with user interest in hierarchical manner and this profile is learned by assigning user details collected by our search method. Main focused on how quickly we can collect user’s information and achieve a profile stability, and how effectively improve profiles. Along with the framework we also consider a new method to extract feature from document. A Wordnet and Lexico-syntactic pattern for hyponyms approach is applied to export the important feature of document to represent the profile ontology. The user profile further improved by learning interesting knowledge from similar profiles.

Managing the video content for searching and summarizing has become a challenging task. Extracting semantics from video scenes enables information to be presented in a more understandable manner. Finding the semantics between video contexts is a difficult task; much recent research has focused on this issue. Most videos, such as TV serials and commercial movies, are character- centric. Therefore, the context and relationship between characters needs to be organized systematically to analyze the video. So, it is necessary to identify the contextual relationships between characters in the scene and the video. We propose Character-Net, a network structure. It finds characters in a group of shots, extracts the speaker and listeners in the scene, represents it with character-based graphs and draws the relationship between all characters by accumulating the character-based graphs at video. In this paper, we describe how to build Character-Net. Experimental results show Character-Net is an effective methodology to extract the major characters in videos.

본 논문은 동영상 검색 시에 사용되는 동영상 어노테이션 기술에 대하여 대화 참여자 판단을 위한 영상적 요소를 제안한다. 이러한 영상적 요소들은 동영상 내의 대화상황을 구성하며 대화 참여자를 판단하는데 이용될 수 있다. 대화 참여자는 대화상황 파악의 극히 중요한 부분으로써 동영상 검색 시에 기준이 될 수 있으며 동영상의 이야기 전개를 이끌어 나간다. 대화의 영상적 요소에는 자막의 유무, 장면, 인물 등장순서, 시선방향, 패턴 등이 있다. 본 논문에서는 이러한 요소들이 동영상 내의 역할을 기술하였고 영상적 요소에 근거하여 대화 참여자를 판단하려 한다. 각각의 요소들만으로는 대화 참여자를 정확하게 판단하는 근거를 제공할 수 없으므로 여러 요소들이 결합되고 일정한 제약조건이 만족되었을 때 정확한 결과를 얻을 수 있다. 따라서 본 논문에서는 실험을 통하여 기존의 관련연구와 비교하였고 대화 참여자 판단의 정확성이 제고되었음을 증명하였다.

본 논문은 영화나 드라마 등에서 등장하는 인물의 배역명을 웹 페이지로부터 추출하여 자동적으로 어노테이션(Annotation)하는 방법을 제안한다. 영상 내에서 각각의 배우들이 등장함에 따라 발생되는 다양한 시각정보와 배역명을 일치시키는 일은 상당히 어려운 일이다. 기존의 등장인물에 대한 자동 어노테이션 방법들은 영상과 관련된 자막(Subtitle)을 활용하거나 자막과 대본(Script)을 활용하여 등장인물과 배역명을 연결시킨다. 하지만 자막이나 대본만으로 등장인물을 매칭할 경우, 영상의 이미지와 배역명의 직접적인 연관성을 찾을 수 없기 때문에 부적절한 어노테이션의 배제가 어렵다. 그러므로 영상으로부터 등장인물의 검출과 함께 관련 웹 페이지로부터 등장인물의 정보를 추출하여 이들간의 유사도를 측정하여 자동으로 어노테이션이 가능하도록 하는 것이 본 연구의 목적이다. 이를 위한 본 연구의 프로세스는 다음과 같다. (ⅰ)영상으로부터 등장인물을 자동으로 분석하고 검출한 후, (ⅱ)기존의 웹 정보 추출 방법을 이용하여 웹 페이지로부터 등장인물의 이미지와 등장인물을 기술할 수 있는 키워드인 배역명을 추출한다. (ⅲ)영상으로부터 검출된 등장인물의 이미지를 학습시키고, 영상 속 등장인물과 웹으로부터 추출된 이미지와의 비교를 통하여, 등장인물에 대한 배역명을 자동으로 어노테이션하는 시스템을 구현한다.

차량경로문제는 지리적으로 산재해 있는 고객들에게 서비스를 수행하고 다시 차고지로 복귀하는데 소 요되는 비용을 최소화하도록 차량의 경로를 결정하는 문제이다. 본 논문에서는 NP-hard 문제로 잘 알려 진 차량경로문제를 풀기 위하여 빈발 경로 패턴을 이용한 메타 휴리스틱 해법을 제안한다. 이는 좋은 해 의 집합에서 빈번히 발생하는 경로의 패턴을 이용하여 더욱 합리적인 해의 공간을 탐색함으로써 개선된 해를 발견할 수 있게 해준다. 본 메타 휴리스틱의 성능 비교를 위하여 기존의 알려진 다양한 휴리스틱들 과 비교 실험을 하였으며 기존의 결과보다 향상되었음을 보였다.

Extracting local ontology from domain-specific documents for the purpose of acquiring knowledge or semantic information to extend their ontologies is considered very important. Main components of ontology are concepts and relations between concepts. In this paper, we focus on extracting triples, in which verbs are relations and subjects/objects are concepts, from documents based on natural language. Further, we show that term frequency is the most reliable measure among tf-idf and entropy on evaluating relations extracted from documents, particularly the aircraft maintenance manual.

Vehicle Routing Problem(VRP) is a well known NP-hard problem, where an optimal solution to the problem cannot be achieved in reasonable time, as the problem size increases. Due to this problem, many researchers have proposed a heuristic using a local search. In this paper, we propose the Constraint Satisfaction Problem(CSP) model in Large Neighborhood Search. This model enables us to reduce the size of local search space. In addition, it enables the easy handling of many constraints in the real-world.

Some data structures had been proposed for storing meaningful information compressed from an original database to find frequent itemsets since FP-tree introduced by Han, et al. In this context, we propose a new data structure, called a frequent pattern network, which represents items as vertices and 2-itemsets as edges of the network. In addition, our focus is how to use this framework efficiently to discover associative relationships among items on the network. To support the goal, we use a clustering method to group the vertices on the network into clusters, so that items within a cluster have high similarity in comparison to one another, but are very dissimilar to items in other clusters. Our experiments show the accuracy of clustering items on the network using three similarity methods. From the results, the confidence similarity had a stronger influence than correlation and edge weight similarity during the task of clustering items.

CMOI: A Collaborative Method for Multiple Ontology Integration

A Method for Integration across Text Corpus and WordNet-based Ontologies

Recommender systems, which have emerged in response to the problem of information overload, provide users with recommendations of contents that are likely to fit their needs. One notable challenge in a recommender system is the cold start problem. To address this issue, we propose a collaborative approach to user modeling for generating personalized recommendations for users. Our approach first discovers useful and meaningful patterns of users, and then enriches a personal model with collaboration from other similar users. In order to evaluate the performance of our approach, we compare experimental results with those of a probabilistic learning model, a user-based collaborative filtering, and vector space model. We present experimental results that show how our model performs better than existing work.

한정 용량 차량경로문제는 용량의 제약이 있는 차량을 사용하여 지리적으로 산재해 있는 고객들에게 서비스를 수행하고 다시 돌아오는 최소 이동시간의 차량경로를 결정하는 문제이다. 본 논문에서는 NP-hard 문제로 잘 알려진 한정 용량 차량경로문제를 풀기 위한 무게중심 기반의 삽입 클러스터링 휴리스틱 알고리즘을 제안한다. 이 알고리즘은 3단계로 구성된다. 첫 번째 단계는 초기 중심 설정단계로써 삽입 기반의 휴리스틱을 수행하여 초기 경로의 중심 값을 찾는 단계이다. 두 번째 단계는 클러스터 생성 단계로써 각 고객의 좌표만을 사용한 Kmeans 방법론과 좌표와 고객의 주문량을 함께 고려한 무게중심과의 경쟁을 통해 클러스터를 생성하게 된다. 세 번째 단계에서는 생성된 클러스터내의 경로 설정을 통하여 알고리즘을 종료하게 된다. 본 논문에서는 고객의 주문량을 휴리스틱 생성하는 요소로 활용함으로써 제약조건 활용의 새로운 가능성을 확인할 수 있었다. 제안한 휴리스틱의 성능 비교를 위한 두 가지의 Sweep 휴리스틱과의 비교실험을 통하여 더 적은 거리의 값을 산출해 낼 수 있었다.

인터넷과 디지털 멀티미디어의 발달은 인터넷상에 개인이 만든 UCC 동영상과 다양한 매체로부터 생성된 동영상의 기하급수적인 증가를 이끌어 왔다. 또한, 지속적으로 증가되는 동영상 속에서 원하는 동영상을 찾으려는 필요성과 동영상의 축약에 대한 필요성이 지속적으로 증대되어 왔다. 검색과 축약을 하기 위해 선행되어야 하는 것이 동영상 정보에서 의미를 추출하여 영상 정보를 어노테이션 해두는 작업이다. 동영상의 어노테이션 방식은 텍스트 기반의 방식과 시각정보 방식으로 분류할 수 있다. 텍스트 기반의 방식은 이미지의 주변의 텍스트 정보로부터 이미지의 내용을 추출 하는 방식이고, 시각정보 방식은 이미지 자체에서 시각정보를 추출하여 어노테이션 하는 방식이다. 하지만 동영상 주변에 동영상에 대해 설명하거나 관련된 글이 없는 경우도 있어 텍스트 기반의 방식은 한계점을 가질 수 밖에 없다. 따라서 동영상 자체에서 정보를 추출하는 방식으로 어노테이션을 진행하여야 한다. 본 논문은 동영상 어노테이션을 할 때 영상으로부터 얼굴 인식정보를 추출하고 자막정보와 연합하여 해당하는 장면의 상황정보를 추출하여 해당 장면에 어노테이션을 작성하는 것에 대한 연구이다. 얼굴인식 정보로부터 현재 장면에 누가 있는지 알아내고, 입의 움직임으로부터 화자가 누구인지 알아내고, 자막정보로부터 장면의 화자와 다른 사람과의 장면상황을 추출하여 XML로 변환하는 방법을 본 연구에서 제안한다.

A Method for Integration of WordNet-based Ontologies Using Distance Measures

With the increasing popularity of online video, efficient captioning and displaying the captioned text (subtitles) have also been issued with the accessibility. However, in most cases, subtitles are shown on a separate display below the screen. As a result, some viewers lose condensed information about the contents of the video. To elevate readability and visibility of viewers, in this paper, we present a framework for displaying synchronized text around a speaker in video. The proposed approach first identifies speakers using face detection technologies and subsequently detects a subtitles region. In addition, we adapt DFXP, which is interoperable timed text format of W3C, to support interchanging with existing legacy system. In order to achieve smooth playback of multimedia presentation, such as SMIL and DFXP, a prototype system, namely MoNaPlayer, has been implemented. Our case studies show that the proposed system is feasible to several multimedia applications.

According to continuous increasing of spam email, 92.6% of recent total email is known spam email. In this research, we will show an adaptive learning system that filter spam emails based on user’s action pattern as time goes by. In this paper, we consider relationship between user‘s actions such as what action is took after one action and how long does it take. They analyze that each action has how much meaning, and that it has an effect on filtering spam emails. And that in turn determines weight for each email. In experimentation, we will compare results of system of this research and weighted Bayesian classifier using real email dataset. Also, we will show how to handle personalization for concept drift and adaptive learning.

No Data

Table based Single Pass Algorithm for Clustering Electronic Documents in 20NewsGroups

Catch Crawler: Automatic Web Information Extractor using Style sheet

육상 화물과 같이 항공 화물도 화물 운송에 대한 운송 업체의 신뢰도가 중요하다. 이러한 배경으로 항공사나 포워더 같은 운송업체는 자신의 서비스 품질을 지속적으로 측정하고 평가하기 위해서 항공 화물 추적 관리 시스템을 사용한다. 그러나 각 항공 화물 추적 관리 시스템은 내부적으로 데이터의 형식 및 처리 방법 등을 서로 다르게 사용하기 때문에 타 시스템의 관리하에 있는 화물의 정보를 파악하는데 문제가 있다. 따라서 본 논문에서는 웹 서비스 기술을 사용하여 이종 시스템 간에서 발생하는 이러한 문제점을 해결하여 효율적으로 화물 정보를 관리하는 시스템을 제안하고 구현하였다. 또한 타 시스템에 대한 접근을 용이하게 하였으며, 화물에 대한 다양한 표현을 할 수 있으므로 보다 향상된 고객 서비스를 제공할 수 있었다.

No Data

With the spread of the digital library and the web, users can obtain a wide variety of information, and also can access novel content. In this environment, finding useful information from a huge amount of available content becomes a time consuming process. In this paper, we focus on user modeling for personalization to recommend content relevant to user interests. We exploit the data mining techniques for identifying useful and meaningful patterns of users. Each user model, collectively called PTP (Personalized Term Pattern), is represented as both interest patterns and disinterest patterns. We present empirical experiments using NSF research award datasets to demonstrate our approach and evaluate performance compared with existing methods.

This paper proposes a collaborative filtering method with user-created tags focusing on changes of web content and internet services. Collaborative tagging is employed as an approach in order to grasp and filter users’ preferences for items. In addition, we explore several advantages of collaborative tagging for future searching and information sharing which is used for automatic analysis of user preference and recommendation. We present empirical experiments using real dataset from del.icio.us to demonstrate our algorithm and evaluate performance compared with existing works.

No Data

No Data

A Divide-and-Conquer Approach to Detecting Latent Community of Practice from Virtual Organizations

With the spread of the Web, users can obtain a wide variety of in-formation, and also can access novel content in real time. In this environment, finding useful information from a huge amount of available content becomes a time consuming process. In this paper, we focus on user modeling for personalization to recommend content relevant to user interests. Techniques used for association rules in deriving user profiles are exploited for discovering useful and meaningful patterns of users. Each user preference is presented the frequent term patterns, collectively called PTP (Personalized Term Pattern) and the preference terms, called PT (Personalized Term). In addition, a content-based filtering approach is employed to recommend content corresponding with user preferences. In order to evaluate the performance of the proposed method, we compare experimental results with those of a probabilistic learning model and vector space model. The experimental evaluation on NSF research award datasets demonstrates that the proposed method brings significant advantages in terms of improving the recommendation quality in comparison with the other methods.

Social networks have been working as a medium to provide cooperative interactions between people. However, as some of users take malicious actions, the social network potentially contains some risks (e.g., information distortion). In this paper, we propose a robust information diffusion (or propagation) model to detect malicious peers on social network. Especially, we apply statistical sequence analysis to discover a peculiar patterns on recommendation flows. Through two experimentation, we evaluated the performance of risk discovery on social network.

No Data

No Data

Collaborative Filtering recommender systems, one of the most representative systems for personalized recommendations in E-commerce, enable users to find the useful information easily. But traditional CF suffers from some weaknesses: scalability and real-time performance. To address these issues, we present a novel model-based CF approach to provide efficient recommendations. In addition, we propose a new method of building a model with dynamic updates, when users present explicit feedback. The experimental evaluation on MovieLens datasets shows that our method offers reasonable prediction quality as good as the best of user-based Pearson correlation coefficient algorithm.

Collaborative Filtering recommender system, one of the most representative systems for personalized recommendations in E-commerce, is a system assisting users in easily finding useful information. However, traditional collaborative filtering systems are typically unable to make good quality recommendations in the situation where users have presented few opinions; this is known as the cold start problem. In addition, the existing systems suffer some weaknesses with regard to quality evaluation: the sparsity of the data and scalability problem. To address these issues, we present a novel approach to provide enhanced recommendation quality supporting incremental updating of a model through the use of explicit user feedback. A model-based approach is employed to overcome the sparsity and scalability problems. The proposed approach first identifies errors of prior predictions and subsequently constructs a model, namely the user-item error matrix, for recommendations. An experimental evaluation on MovieLens datasets shows that the proposed method offers significant advantages both in terms of improving the recommendation quality and in dealing with cold start users.

Multi-agent systems have been attacking the challenges of distributed information retrieval. In this paper, we propose a consensus method-based framework to evaluate the performance of cooperative information retrieval tasks of the agents. Two well-known measurements, precision and recall, are extended to handle consensual closeness (i.e., local and global consensus) between the retrieved results. We show in a motivating example that the proposed criteria are prone to solve the problem of rigidity of classical precision and recall. More importantly, the retrieved results can be ranked with respect to the consensual score.

Collective intelligence is a key issue for efficient collaborations on semantic grid environment. Through interactions among agents on the semantic grid, we want to discover social centralities underlying on agent communities. In this paper, we propose an ontology mapping approach to build collective intelligence based on social mediation. As a case study, we demonstrate AnnotGrid platform to support manual annotation tasks of image dataset.

No Data

No Data

No Data

Recommender systems enable a user to decide which information is interesting and valuable in our world of information overload. Collaborative Filtering (CF), one of the most successful technologies in recommender systems suffers from improper use of personal information and the incredibility of recommendations. To deal with these issues, we have been focusing on the trust relationships between individuals, i.e. web of trust, especially for protecting the recommender system against profile injection attack. Based on trust propagation scheme, we proposed TCFMA architecture which is added agent-based scheme obtaining attack resistance property as well as improving the efficiency of distributed computing. In web of trust, users’ personal agents find a unique migration path made up of latent neighborhoods and reduce search scope to a reasonable level for mobile agents by using the Advogato algorithm. The experimental evaluation on Epinions.com datasets shows that the proposed method brings significant advantages in terms of dealing with profile injection attack without any loss of prediction quality.

Existing item bank systems present a variety of assessments for data management and integration with individual learning evaluation systems. However, as the data in these established item bank systems do not include seman-tics, the bank systems cannot analyze implications and perform accurate searches such as for synonymous words. Therefore, both learners and teachers can access only simple text data using the item bank system and often waste time checking unnecessary search results and extracting information from data repeatedly. Moreover, since there is no clear definition of the relationship between items and teachers or learners, or between data and units, it is difficult to use and share extra-item information. In order to solve these problems, this research describes the definition of conception and organization by constructing the Ontology of an E-Item Bank system using OWL, which is a form of semantic web technology. Furthermore, on the basis of Ontology, OWL metadata, individuals, are built, semantic factors are then extracted using OWLJessKB. We do not only use this extracted semantic factors as a fact, but also domain rules using SWRL for the JESS inferencing engine. As such, it is possible to make inferences and provide reasoning in web data structures. As a result, it is possible to combine all the data of the E-Item bank system and facilitate understanding of the meanings of concepts by the computer. In addition, a search service with inference can be applied to education and can lead to cooperative study between teachers and students.

A design approach for Contextualized collaborative recommender system by nurturing the user interactive feedback

No Data

No Data

RFID (Radio Frequency Identification Device) connection technology is acknowledges as an enabling technology impacting many areas of human endeavor. However, technical aspects of RFID technology based on RFID mid-dleware in embedded device is still sort of reality. The reason is that RFID middleware should be integrated into the Embedded system with USN (ubiqui-tous sensor network) platform in native access mechanisms that will take RFID information from diverse heterogeneous sensors in scattered, distributed busi-ness environment in real time. In this sense, application program that uses the RFID middleware should collect, smooth and filter RFID tag data from several various RFID readers in general purpose sever PC and provide integration func-tions with application programs such as existing middleware component and communication component to control the RFID readers. Consequently, in this research , we implement a middleware with DBMS and data filtering gateway component for the embedded system that serves as a data filtering to smooth and filter data and thus reduce the volume of data. With the synchronized data-base, further, our proposed middleware architecture can be incorporated into the solution of multiplatform application program that may take advantage of a web-service technology to be run with OS in a general server PC. Our experi-mental run justified the implementation structure RFID middleware for the em-bedded system.

No Data

With the continuous growth of e-commerce there is a huge amount of products information available online. To choose products is time-consuming. On the other hand, online shop managers expect to apply information techniques with the least investment to get the highest efficiency to increase profit and perfect service. Hence many e-commerce systems use association rule mining to further refine their management. However previous association rule algorithms have two limitations. First, they only use the number to weight item’s essentiality and ignore essentiality of item profit. Second, they did not consider the relationship between number and profit of item when they do mining. We address a novel algorithm, profit-based association rule algorithm, to mine rules. In this algorithm we use profit-based technique to generate 1-itemsets and the multiple minimum supports mining technique to generate N-items large itemsets. Through several experiments, we have shown that these optimization techniques can yield significant performance benefit.

인터넷이 생활의 일부로 자리 잡은 최근 개인의 개성을 표출할 수 있는 블로그가 각광받고 있다. 본 논문에서는 이러한 블로그 환경에 교육계 분야에서 최근 화두가 되는 E-Learning을 접목시켜 각 개인의 블로그를 조직화하여 협업 학습을 할 수 있는 E-Learning 시스템을 제안한다. 현재 E-Learning 시스템들이 다양한 시도에도 불구하고, 아직은 학교 교육에 대한 과외 대체 교육에 머물고 있고 학습자 개개인에게 개별적인 학습 피드백을 제공하기 위해 많은 시간이 소요되며, 전통적인 교실 수업에 존재하는 사회적 교류를 제공하지 못하고 있다. 따라서 이 논문에서는 블로그 환경에서 학습자끼리의 코멘트에 의한 상호작용을 통해 자발적인 협업 학습 서비스를 제공하고자 한다.

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

No Data

CRMaps: Managing Customer Relationships Based on Ontological Fuzzy Cognitive Maps

No Data

No Data

MoCAAS : Auction Agent System Using a Collaborative Mobile Agent in Electronic Commerce

Service delivery agent system for mobile device

Intelligent XML Content Retrieval Manager for the e-learning Environment

Dialogue-based Cyber Salesperson for Music Recommendation

Collaborative Information Filtering by Using Categorized Bookmarks on the Web

Intelligent Gathering of Contents on Distance Education using Mobile Agent

Intelligent Cyber Logistics using Reverse Auction in Electronic Commerce

Knowledge sharing system using Mobile Agent: A case study of the Q&A board

Recommender System with Configuration in Electronic Commerce

BISAgent: Collaborative Web Browsing through Sharing of Bookmark Information

No Data

No Data

No Data

No Data

No Data

Web-based Intelligent Tutoring System for Collaborative Learning

Interactive Simulation Program for Train scheduling Using Advanced Expert System

No Data

Cooperation in Multi-Agent System for Meeting Scheduling

Ramp Activity Expert System for Scheduling and Co-ordination at an Airport

No Data

No Data

No Data

No Data

No Data

Agents-Based Framework for Brokerage Between Buyers and Sellers on Electronic Commerce

No Data

No Data

RACES: Ramp Activity Coordination Expert System

No Data

The Scheduling Strategies on Ramp Activity Coordination Expert System

No Data

Modern Modeling of A Classic Multiperiod Problem

Interactive Expert Model in Ramp Coordination System

No Data

No Data

No Data

No Data

Reactive Scheduling in Meta-Programming : A Case Study of Course Scheduling

No Data

No Data

No Data

No Data

Constraint-Directed Course Scheduler

Constraint Logic Programming:An Integration of Artificial Intelligence with Linear Programming

No Data

No Data

No Data

Anticipatory Pruning Networks in CLP languages

Tutorial:Constraint Logic Programming

Constraint Directed reasoning in Prolog

Constraint Solving in CLP language

샷 경계 검출(Shot Boundary Detection)은 영상 콘텐츠 분석을 위한 필수적인 기술이며, 다양한 방식으로 편집된 영상의 샷 경계를 정확하게 검출하기 위한 연구가 지속되어 왔다. 그러나 기존에 연구들은 고정된 샷 경계 검출 알고리즘이나 매뉴얼한 작업과 같이 학습이 불가능한 과정이 포함되어 있어 성능 개선에 한계가 있었다. 본 논문에서는 이러한 과정을 제거한 End-to-End 모델을 제안한다. 제안하는 모델은 시공간 정보 추출성능을 높이기 위해 행동 인식 데이터셋을 이용한 전이학습을 사용하고, 샷 경계 검출 성능을 높이기 위해 개선된 지식의 증류기법(Knowledge Distillation)을 결합한다. 제안하는 모델은 ClipShots 데이터셋에서 DeepSBD에 비해 cut transition과 gradual transition이 각각 5.4%, 41.29% 높은 성능을 보였고, DSM과의 비교에서 cut transition의 정확도가 1.3% 더 높은 결과를 보였다.