• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.601-604
• /
• 2023

기술이 발전함에 따라 복합적인 모달리티 정보를 포함하는 멀티미디어 데이터의 수집이 용이해지면서, 사람의 성격 특성을 이해하고 이를 개인화된 에이전트에 적용하고자 하는 연구가 활발히 진행되고 있다. 본 논문에서는 비디오 스트림 구조를 활용하여 사용자 특성을 예측하기 위한 동적 키프레임 추출 방법을 제안한다. 비디오 데이터를 효과적으로 활용하기 위해서는 무작위로 선택한 프레임에서 특징을 추출하던 기존의 방법을 개선하여 영상 내 시간에 따른 정보와 변화량을 기반으로 중요한 프레임을 선택하는 방법이 필요하다. 본 논문에서는 제 3자가 평가한 Big-five 지표 값이 레이블링된 대표적인 데이터셋인 First Impressions V2 데이터셋을 사용하여 외면에서 발현되는 특징들을 기반으로 영상에서 등장하는 인물들의 성격 특성을 예측했다. 결론에서는 선택된 키프레임에서 멀티 모달리티 정보를 조합하여 성격 특성을 예측한 결과와 베이스라인 모델과의 성능을 비교한다.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.18 no.4
• /
• pp.71-79
• /
• 2019

By using convolutional neural networks (CNNs) based on semantic segmentation, road surface damage detection has being studied. In order to generate the CNN model, it is essential to collect the input and the corresponding labeled images. Unfortunately, such collecting pairs of the dataset requires a great deal of time and costs. In this paper, we proposed a road surface damage detection technique based on semi-supervised learning using pseudo labels to mitigate such problem. The model is updated by properly mixing labeled and unlabeled datasets, and compares the performance against existing model using only labeled dataset. As a subjective result, it was confirmed that the recall was slightly degraded, but the precision was considerably improved. In addition, the $F_1-score$ was also evaluated as a high value.

• The Journal of Korean Institute of Information Technology
• /
• v.17 no.9
• /
• pp.99-112
• /
• 2019

Cracks on pavement surfaces are critical signs and symptoms of the degradation of pavement structures. Image-based pavement crack detection is a challenging problem due to the intensity inhomogeneity, topology complexity, low contrast, and noisy texture background. In this paper, we address the problem of pavement crack detection and segmentation at pixel-level based on a Deep Neural Network (DNN) using gray-scale images. We propose a novel DNN architecture which contains a modified U-net network and a high-level features network. An important contribution of this work is the combination of these networks afforded through the fusion layer. To the best of our knowledge, this is the first paper introducing this combination for pavement crack segmentation and detection problem. The system performance of crack detection and segmentation is enhanced dramatically by using our novel architecture. We thoroughly implement and evaluate our proposed system on two open data sets: the Crack Forest Dataset (CFD) and the AigleRN dataset. Experimental results demonstrate that our system outperforms eight state-of-the-art methods on the same data sets.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.43-50
• /
• 2024

This paper evaluates the Korean sentiment analysis performance of large language models like GPT-3.5 and GPT-4 using a zero-shot approach facilitated by the ChatGPT API, comparing them to pre-trained Korean models such as KoBERT. Through experiments utilizing various Korean sentiment analysis datasets in fields like movies, gaming, and shopping, the efficiency of these models is validated. The results reveal that the LMKor-ELECTRA model displayed the highest performance based on F1-score, while GPT-4 particularly achieved high accuracy and F1-scores in movie and shopping datasets. This indicates that large language models can perform effectively in Korean sentiment analysis without prior training on specific datasets, suggesting their potential in zero-shot learning. However, relatively lower performance in some datasets highlights the limitations of the zero-shot based methodology. This study explores the feasibility of using large language models for Korean sentiment analysis, providing significant implications for future research in this area.

• Composites Research
• /
• v.37 no.1
• /
• pp.40-45
• /
• 2024

In this research, we introduce a novel approach that employs a 3D convolutional neural network (CNN) model to predict the permeability of Gas Diffusion Layers (GDLs). For training the model, we create an artificial dataset of GDL representative volume elements (RVEs) by extracting morphological characteristics from actual GDL images obtained through X-ray tomography. These morphological attributes involve statistical distributions of porosity, fiber orientation, and diameter. Subsequently, a permeability analysis using the Lattice Boltzmann Method (LBM) is conducted on a collection of 10,800 RVEs. The 3D CNN model, trained on this artificial dataset, well predicts the permeability of actual GDLs.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.5
• /
• pp.75-82
• /
• 2023

In collaborative filtering systems, the item rating prediction values calculated by the systems are very important for customer satisfaction with the recommendation list. In the time-aware system, predictions are calculated by reflecting the rating time of users, and in general, exponentially lower weights are assigned to past rating values. In this study, to find out whether the influence of rating time on the rating value varies according to various factors, the correlation between user rating value and rating time is investigated by the degree of user rating activity, the popularity of items, and item genres. As a result, using two types of public datasets, especially in the sparse dataset, significantly different correlation index values were obtained for each factor. Therefore, it is confirmed that the influence weight of the rating time on the rating prediction value should be set differently in consideration of the above-mentioned various factors as well as the density of the dataset.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.5
• /
• pp.221-226
• /
• 2021

Collaborative filtering based recommender systems recommend user-preferrable items based on rating history and are essential function for the current various commercial purposes. In order to determine items to recommend, prediction of preference score for unrated items is estimated based on similar rating history. Previous studies usually employ two methods individually, i.e., similar user based or similar item based ones. These methods have drawbacks of degrading prediction accuracy in case of sparse user ratings data or when having difficulty with finding similar users or items. This study suggests a new rating prediction method by integrating the two previous methods. The proposed method has the advantage of consulting more similar ratings, thus improving the recommendation quality. The experimental results reveal that our method significantly improve the performance of previous methods, in terms of prediction accuracy, relevance level of recommended items, and that of recommended item ranks with a sparse dataset. With a rather dense dataset, it outperforms the previous methods in terms of prediction accuracy and shows comparable results in other metrics.

• Korean Journal of Remote Sensing
• /
• v.35 no.2
• /
• pp.279-288
• /
• 2019

We evaluated the land cover classification performance of SegNet, which features semantic segmentation of aerial imagery. We selected four semantic classes, i.e., urban, farmland, forest, and water areas, and created 2,000 datasets using aerial images and land cover maps. The datasets were divided at a 8:2 ratio into training (1,600) and validation datasets (400); we evaluated validation accuracy after tuning the hyperparameters. SegNet performance was optimal at a batch size of five with 100,000 iterations. When 200 test datasets were subjected to semantic segmentation using the trained SegNet model, the accuracies were farmland 87.89%, forest 87.18%, water 83.66%, and urban regions 82.67%; the overall accuracy was 85.48%. Thus, deep learning-based semantic segmentation can be used to classify land cover.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.11a
• /
• pp.1215-1218
• /
• 2021

한국을 방문하는 외국인들은 매년 증가하고 있고 방한 목적 중 식도락관광이 3위에 오를 만큼 세계에서 한국 음식은 위상이 높아지고 있다. 하지만, 한국에서의 알레르기 성분 표시는 법적 의무가 아니기 때문에 대부분의 한식당에서는 이를 표시하지 않고 있고 알레르기가 있는 외국인 관광객들은 한국 음식 섭취에 있어서 상당한 위험과 불편함을 부담하고 있다. 이에 본 논문에서는 머신러닝을 활용하여 사진 촬영만으로 쉽고 정확하게 알레르기 성분을 제공하고자 사물 이미지 데이터 전처리를 위한 HSV(Hue, Saturation, Value) 데이터 전처리 기법을 제안한다. 제안하는 기법은 이미지의 HSV의 평균 및 분산, 표준편차를 통해 불필요한 데이터를 제거한다. 성능평가에서는 비빔밥, 불고기, 제육볶음 등 사진 약 500장의 데이터 셋을 구성하여 HSV의 평균 및 분산을 통해 이미지를 제거하는 방식으로 구축한 데이터 셋을 TensorFlow를 통해 정확도와 학습시간을 측정한다. 측정결과, 제안하는 기법으로 구축한 데이터 셋은 최소 15%에서 최대 25% 높은 정확도와 최소 37.96%에서 최대 42.85% 높은 정도 낮은 학습시간을 보여주었다. 향후 HSV를 활용한 데이터 전처리 기법은 더 많은 데이터를 통해 더욱 구체적인 성능 분석이 필요하다. 또한, 실질적인 개발 및 구현을 통해 제안하는 데이터 전처리 기법의 더욱 현실적인 검증이 필요하다.

• The Journal of the Korea Contents Association
• /
• v.9 no.11
• /
• pp.455-464
• /
• 2009

To overcome article-oriented search functions and provide author-oriented ones, a namesake problem for author names should be solved. Author disambiguation, proposed as its solution, assigns identifiers of real individuals to author name entities. Although recent state-of-the-art approaches to author disambiguation have reported above 90% performance, there are few academic information services which adopt author-resolving functions. This paper describes a large-scale test set for author disambiguation which was created by KISTI to foster author resolution researches. The result of these researches can be applied to academic information systems and make better service. The test set was constructed from DBLP data through web searches and manual inspection, Currently it consists of 881 author names, 41,673 author name entities, and 6,921 person identifiers.