Earticle

This study focuses on the development of a learning management s ystem (LMS)-based automatic annotation authoring tool for computer -assisted interpreter training (CAIT). The primary goal of the research is to establish a specialized LMS that integrates AI-powered speech-to -text (STT) and automatic annotation functionalities to create an effici ent feedback system. Through the use of external APIs (Application P rogramming Interface), the tool implements automatic detection of di sfluency elements such as fillers, pauses, and repetitions in STT-gen erated interpreting outputs, which are visualized using a TextAE (Tex t Annotation Editor) program. Additionally, the system enhances obj ectivity in evaluation by providing visualized assessment results and supports autonomous practice for learners through a self-study featu re. The significance of this study lies in its effective application of AI technologies to interpreter training, enabling an integrated system th at encompasses task creation, execution, and evaluation in one platf orm. It is anticipated that this research will contribute to the digital tr ansition of interpreter training and improve educational efficiency.

This study aims to describe music-related request patterns observed in AI assistant platforms and formalize them by using the local grammar graph(LGG) methodology. These patterns, transformed into Finite-State Transducers in the Unitex platform, allow us to automatically classify and annotate various sentences used for request, which is crucial for diverse machine learning approaches. In this study, we examined a corpus of tweet texts crawled by Deco-T-Crawler as seed data and classified music-related requests into 3 categories: music on/off, volume up/down and music-play control. Then each category was divided into sub-categories which were represented by several LGGs. By applying these LGGs and the DECO Korean dictionary on test data, we obtained the information retrieval performance. In this regard, the linguistic resources proposed in this study are proved concrete and meaningful.

Automatic image annotation (AIA) is an active topic of research in computer vision and pattern recognition. In the last two decades, large amount of researches on AIA have been proposed, mainly including classification-based methods and probabilistic modeling methods. As one of the most common methods for AIA, support vector machine (SVM) has been widely applied in the multimedia research community, especially for image classification, image annotation and retrieval. However, compared with various SVM methods and their corresponding applications in the literature, there is almost no review research and analysis about SVM related studies. So the current paper, to start with, elaborates the basic principles of SVM. Followed by it summarizes SVM with applications to image annotation from three aspects of SVM ensemble for AIA, SVM with mixture of kernels for AIA and hybrid SVM for AIA respectively. In addition, SVM exploited in several other applications are also briefly reviewed. Finally, we end this paper with a summary of some important conclusions and highlight the potential research directions of SVM in automatic image annotation for the future.

In this paper, we present a new method for automatic image annotation by applying semi- supervised learning based on the Bayesian framework. On the one hand, we employ the semi- supervised learning, i.e., transductive support vector machine (TSVM) to enhance the quality of training image data, which is a promising way to find out the underlying relevant data from the unlabeled ones. On the other hand, a simple yet very efficient Bayesian model is built to implement image annotation by the maximum a posteriori (MAP) criterion. The novelty of our method mainly lies in two aspects: exploiting TSVM to improve the quality of training image dataset and utilizing the Bayesian model to predict the candidate annotations for the unseen images. Experimental results on the standard Corel dataset demonstrate that the proposed method is superior or highly competitive to several state-of-the-art approaches.

In recent years, automatic image annotation (AIA) has been applied to cross-media retrieval usually due to its advantage of mining correlations of images and annotation texts efficiently. However, some AIA methods just annotate images as a unit and the accuracy of annotation may not be acceptable. In this paper, we propose a kind of probabilistic model which may assign keywords to an un-annotated image automatically based on a training dataset of images. Images in the training dataset are segmented into regions and a kind of vocabulary called blob is used to represent these image regions. Blobs are generated by using K-Means algorithm to cluster these image regions. Through this model, we can predict the probability of assigning a keyword into a blob. After the accomplishment of annotation, a keyword corresponds to one image region. Furthermore, the feature vectors of text documents are generated by TF.IDF method and images’ automatic annotation information is used to retrieve relevant text documents. Experiments on the IAPR TC-12 dataset and 500 Wikipedia webpages about landscape show the usefulness of applying probabilistic model of AIA to the cross-media retrieval.

Corpora annotated with lots of linguistic information are required to develop robust and statistical natural language processing systems. Building such corpora, however, is an expensive, labor-intensive, and time-consuming work. To help the work, we design and implement an annotation tool for establishing a Korean dependency tree-tagged corpus. Compared with other annotation tools, our tool is characterized by the following features: independence of applications, localization of errors, powerful error checking, instant annotated information sharing, user-friendly. Using our tool, we have annotated 100,904 Korean sentences with dependency structures. The number of annotators is 33, the average annotation time is about 4 minutes per sentence, and the total period of the annotation is 5 months. We are confident that we can have accurate and consistent annotations as well as reduced labor and time.

This paper presents the development of specialized software for annotating volume-of-interest on ¹⁸F-FDG PET/CT images with the goal of facilitating the studies and diagnosis of head and neck cancer (HNC). To achieve an efficient annotation process, we employed the SE-Norm-Residual Layer-based U-Net model. This model exhibited outstanding proficiency to segment cancerous regions within ¹⁸F-FDG PET/CT scans of HNC cases. Manual annotation function was also integrated, allowing researchers and clinicians to validate and refine annotations based on dataset characteristics. Workspace has a display with fusion of both PET and CT images, providing enhance user convenience through simultaneous visualization. The performance of deeplearning model was validated using a Hecktor 2021 dataset, and subsequently developed semi-automatic annotation functionalities. We began by performing image preprocessing including resampling, normalization, and co-registration, followed by an evaluation of the deep learning model performance. This model was integrated into the software, serving as an initial automatic segmentation step. Users can manually refine pre-segmented regions to correct false positives and false negatives. Annotation images are subsequently saved along with their corresponding ¹⁸F-FDG PET/CT fusion images, enabling their application across various domains. In this study, we developed a semi-automatic annotation software designed for efficiently generating annotated lesion images, with applications in HNC research and diagnosis. The findings indicated that this software surpasses conventional tools, particularly in the context of HNC-specific annotation with ¹⁸F-FDG PET/CT data. Consequently, developed software offers a robust solution for producing annotated datasets, driving advances in the studies and diagnosis of HNC.

Single-cell RNA sequencing (scRNA-seq) has been widely applied to provide insights into the cell-by-cell expression difference in a given bulk sample. Accordingly, numerous analysis methods have been developed. As it involves simultaneous analyses of many cell and genes, efficiency of the methods is crucial. The conventional cell type annotation method is laborious and subjective. Here we propose a semi-automatic method that calculates a normalized score for each cell type based on user-supplied cell type-specific marker gene list. The method was applied to a publicly available scRNA-seq data of mouse cardiac non-myocyte cell pool. Annotating the 35 t-stochastic neighbor embedding clusters into 12 cell types was straightforward, and its accuracy was evaluated by constructing co-expression network for each cell type. Gene Ontology analysis was congruent with the annotated cell type and the corollary regulatory network analysis showed upstream transcription factors that have well supported literature evidences. The source code is available as an R script upon request.

Summarization is a complex cognitive task and its simulation is very difficult for machines. This paper presents an automatic summarization strategy that is based on a discourse categorization of the textual information. This categorization is carried out by the automatic identification of discourse markers in texts. We defend here the use of discourse methods in automatic summarization. Two evaluations of the summarization strategy are presented. The summaries produced by our strategy are evaluated with summaries produced by humans and other applications. These two evaluations display well the capacity of our application, based on EXCOM, to produce summaries comparable to the summaries of other applications.

Emotion recognition is very important for the extraction of expressive information. In this paper, we provide a robust and versatile emotion annotation scheme that can not only annotate explicit and implicit expressions of emotion,but also can encode different levels of information for a given emotion content. In addition, taking cognitive psychologists’ theories into account, large and comparatively high-quality emotion corpora are automatically created to allow for emotion recognition in Chinese and English. Our annotation scheme can easily be adapted for different kinds of applications dealing with emotion, and,being generic, can also be applied to other languages. We also discuss the two kinds of emotion representations used in our corpus, namely, holistic and componential representations. We find that the two representations have their own qualities and shortcomings and that choosing between them ultimately depends mainly on the application type.

이미지 인식과 내용분석은 이미지 검색과 멀티미디어 데이터 활용 분야에서 핵심기술이라 할 수 있다. 특히 최근 스마트폰, 디지털 카메라, 블랙박스 등에서 수집되는 영상 데이터 양이 급격히 증가하고 있다. 이에 따라 이미지를 인식하고 내용을 분석하여 활용할 수 있는 기술에 대한 요구가 점차 증대되고 있다. 본 논문에서는 이미지 내용정보를 이용하여 자몽으로 이미지로부터 태그정보를 추출하는 방법을 제안한다. 이 방법은 기계학습 기법인 CNN(Convolutional Neural Network)에 ImageNet의 이미지 데이터와 라벨을 학습시킨 후, 새로운 이미지로부터 라벨정보를 추출하는 것이다. 추출된 라벨을 태그로 간주하고 검색에 활용한다면 기존 검색시스템의 정확도를 향상시킬 수 있다는 것을 실험을 통하여 확인하였다.

All these images consist a considerable database. What's more, the semantic meanings of images are well presented by the surrounding text and links. But only a small minority of these images have precise assigned keyphrases, and manually assigning keyphrases to existing images is very laborious. Therefore it is highly desirable to automate the keyphrases extraction process. In this paper, we first introduce WWW image annotation methods, based on low level features, page tags, overall word frequency and local word frequency. Then we put forward our method of multi-cues integration image annotation. Also, show multi-cue image annotation method is more superior than other method through an experiment.

인터넷의 빠른 발전으로 현재 HTML 웹 페이지에 내장된 영상들은 눈에 띄게 두드러졌다. 내용을 묘사하고 주의를 끄는 놀랄만한 함수 때문에 영상들은 웹 페이지에서 사실상 중요하게 되었다. 모든 영상들은 가공할 만한 데이터베이스로 구성되어있다. 게다가. 영상들의 의미론적인 의미도 주변의 텍스트나 링크에 의해 잘 표현된다. 하지만 이들 영상의 소수들이 주요 구에 정확히 할당되고 주요 구들을 현재의 영상에 수작업으로 할당하는 것은 매우 어렵다. 따라서 주요 구들을 추출하는 절차의 자동화는 매우 바람직하다. 본 논문에서는 먼저 저수준 특징, 페이지 태그, 전체적인 단어 빈도수와 지역적 단어 빈도수를 기반으로 한 WWW 영상 주석 방법을 소개한다. 그리고 멀티-큐 통합영상 주석 방법을 전개해 나간다. 또한 실험을 통하여 멀티-큐 영상 주석 방법이 다른 방법보다 우수함을 보여준다.

As the rapid development of the Internet, the embedded images in HTML web pages nowadays become predominant. For its amazing function in describing the content and attracting attention, images become substantially important in web pages. All these images consist a considerable database. What's more, the semantic meanings of images are well presented by the surrounding text and links. But only a small minority of these images have precise assigned keyphrases. and manually assigning keyphrases to existing images is very laborious. Therefore it is highly desirable to automate the keyphrases extraction process. In this paper, we first introduce WWW image annotation methods, based on low level features, page tags, overall word frequency and local word frequency. Then we put forward our method of multi-cues integration image annotation. Also, show multi-cue image annotation method is more superior than other method through an experiment.

이 연구는 한국어 문법 항목 자동 주석의 성능을 향상시키고자 언어 모델과 규칙 기반 검색의 하이브리드 주석기를 구축하였다. 주석기가 비연속 다단어 문법 항목을 포함하여 문법 항목의 후보들을 높은 확률로 검색할 수 있도록 규칙 기반 파이프라인에 문법 항목별로 규칙을 설정하였으며 검색 재현율에서 1에 가까운 점수를 얻었다. 인코더 모델로 DeBERTa-v3-base-korean을 사용하여 도메인 적응 비지도 사전 학습, 다중 과제 학습, 골드 주석 자료 미세 조정을 실시하였으며 디코더 모델로는 GPT4.1, Qwen 3 8b, Llama 3.1 8b를 미세 조정하여 사용하였다. 다의 의미가 없는 문법 항목에 대해서 인코더 모델을 사용하여 문법 항목의 오탐 여부를 판정하여 주석하게 였으며 1에 가까운 f1점수로 달성하였다. 다의 의미가 있는 문법 항목의 경우에 다의 의미 혹은 오탐 여부를 판정하여 주석하게 하였고 인코더, 디코더 모델을 모두 실험하여 0.83의 f1점수를 기록하였다. 이 결과들은 말뭉치에서 문법 항목을 자동으로 추출하는 데 이 주석기가 실제로 활용될 수 있음을 시사한다.

This study developed a hybrid annotator that combines language models with rule-based retrieval to improve the performance of automatic annotation of Korean grammatical items. The annotation task targeted five grammatical items, including items involving polysemy and discontinuous constructions. In the rule-based pipeline, item-specific rules were designed for each grammatical item, achieving a retrieval recall close to 1.0. The encoder model was trained through domain-adaptive unsupervised pretraining, multi-task learning, and fine-tuning on gold-annotated data. The decoder model was trained through fine-tuning on gold-annotated data. For non-polysemous grammatical items, the encoder model achieved an F1 score close to 1.0. For polysemous grammatical items, both encoder and decoder models were evaluated, achieving an F1 score of 0.83. These results suggest that the proposed annotator can be practically used for automatic extraction of grammatical items.

This study investigates the automatic annotation of functional elements in Korean thanking speech-act utterances using GPT and explores its potential for Korean corpus annotation. The annotation criteria in this study are grounded in local grammar theory. Six functional elements of Korean thanking speech acts (Beneficiary, Qualification, Thanking, Benefactor, Intensifier, and Specifier) were adopted as annotation criteria, and a Korean spoken dialogue corpus was annotated. To examine the possibility of automating linguistic annotation, this study used manual annotations produced by human annotators as the gold standard and evaluated the automatic annotation results generated by the GPT-5.4 Thinking model against them. The results showed that GPT-5.4 Thinking achieved over 70% accuracy in the sentence-level evaluation. In the tag-level evaluation, the model achieved precision, recall, and F1 scores of over 80%. This study is significant in that it demonstrates the potential of LLMs as assistive tools for Korean semantic annotation.

기존의 내용기반 비디오 검색 시스템들을 주석기반 검색 또는 특징기반 검색과 같은 단일 방식으로만 검색을 하므로 검색 효율이 낮을 뿐 아니라 완전한 자동 처리가 되지 않아 시스템 관리자나 주석자의 많은 노력을 요구한다. 본 논문에서는 주석기반 검색을 이용하여 대용량의 비디오 데이터에 대한 사용자의 다양한 의미검색을 지원하는 에이전트 기반에서의 자동화되고 통합된 비디오 의미기반 검색 시스템을 제안한다. 사용자의 기본적인 질의를 분석하고 질의에 의해 추출된 키 프레임의 이미지를 사용자가 선택함으로써 인덱싱 에이전트는 추출된 키 프레임의 주석에 대한 의미를 더욱 구체화시킨다.

본 연구는 전문기관에서 생산되는 검증된 문서의 정보를 웹상의 수많은 검증되지 않은 문서에 자동 주석하여 신뢰도를 향상하고 심화 정보를 추가하는 시스템을 제안한다. 제안하는 시스템은 국가암정보센터의 검증된 문서들에서 추출한 19,304개 명제를 위키피디아 암 관련 문서에서 추출한 1,486개 명제에 주석하는 과제를 수행하기 위해, 기존 인공 신경 정리 증명계의 순환 모듈을 단어 임베딩 모듈로 교체하였다. 이를 통해 기존의 근본적인 문제점이었던 학습 시간 문제를 해결하였고, 동일한 환경에서 기존 시스템의 학습 시간이 233.9일로 추정된 것에 비해 재구축한 시스템은 102.1분 내로 학습이 완료되었다. 제안하는 시스템의 장점은 명제를 텐서로 인코딩하여 미분 가능하게 전체적인 연산을 진행하는 인공 신경 정리 증명계가 단어의 정확한 일치를 파악하는 전통적인 정리 증명계를 포함하며 동시에 유사어 관계로부터의 논리 전개 역시 가능하게 한다는 점을 실제 문서 데이터에서 입증했다는 것이다.

We present a system that automatically annotates unverified Web sentences with information from credible sources. The system turns to neural theorem proving for an annotating task for cancer related Wikipedia data (1,486 propositions) with Korean National Cancer Center data (19,304 propositions). By switching the recursive module in a neural theorem prover to a word embedding module, we overcome the fundamental problem of tremendous learning time. Within the identical environment, the original neural theorem prover was estimated to spend 233.9 days of learning time. In contrast, the revised neural theorem prover took only 102.1 minutes of learning time. We demonstrated that a neural theorem prover, which encodes a proposition in a tensor, includes a classic theorem prover for exact match and enables end-to-end differentiable logic for analogous words.

비디오 데이터를 효율적으로 처리하기 위해서는 비디오 데이터가 가지고 있는 내용에 대한 정보를 데이터 베이스에 저장하고 사용자들의 다양한 질의를 처리할 수 있는 의미기반 검색 기법이 요구된다. 기존의 내용기반 비디오 검색 시스템들은 주석기반 검색 또는 특징기반 검색과 같은 단일 방식으로만 검색을 하므로 검색 효율이 낮을 뿐 아니라 완전한 자동 처리가 되지 않아 시스템 관리자나 주석자의 많은 노력을 요구한다. 본 논문에서는 주석기반 검색과 특징기반 검색을 이용하여 대용량의 비디오 데이터에 대한 사용자의 다양한 의미검색을 지원하는 에이전트 기반에서의 자동화되고 통합된 비디오 의미기반 검색 시스템을 제안한다. 사용자의 기본적인 질의와 질의에 의해 추출된 키 프레임의 이미지를 선택함으로써 에이전트는 추출된 키 프레임의 주석에 대한 의미를 더욱 구체화시킨다 또한. 사용자에 의해 선택된 키 프레임은 질의 이미지가 되어 제안하는 특징기반 검색기법을 통해 가장 유사한 키 프레임을 검색한다. 따라서 의미기반 검색을 통해 비디오 데이터의 검색의 효율을 높일 수 있도록 시스템을 설계한다.

In order to process video data effectively, it is required that the content information of video data is loaded in database and semantic- based retrieval method can be available for various query of users. Currently existent contents-based video retrieval systems search by single method such as annotation-based or feature-based retrieval, and show low search efficiency and requires many efforts of system administrator or annotator form less perfect automatic processing. In this paper, we propose semantic-based video retrieval system which support semantic retrieval of various users by feature-based retrieval and annotation-based retrieval of massive video data. By user's fundamental query and selection of image for key frame that extracted from query, the agent gives the detail shape for annotation of extracted key frame. Also, key frame selected by user become query image and searches the most similar key frame through feature based retrieval method that propose. Therefore, we design the system that can heighten retrieval efficiency of video data through semantic-based retrieval.

본 논문에서는 연주자 자세의 그라운드 트루스 획득을 위한 반자동 주석 방법인 SAAnnot-C3Pap를 제안한다. 기존 음악 도메인에서 2차원 관절 위치에 대한 그라운드 트루스를 획득하기 위하여 2차원 자세 추정 방법인 오픈포즈를 활용하거나 수작업으로 라벨링 하였다. 하지만 기존의 오픈포즈와 같은 자동 주석 방법은 빠르지만 부정확한 결과를 보인다는 단점이 있고, 사용자가 직접 주석을 생성하는 수작업 주석화의 경우 많은 노동력이 필요하다는 한계점이 있다. 따라서 본 논문에서는 그 둘의 절충 방안인 반자동 주석화 방법인 SAAnnot-C3Pap을 제안한다. 제안하는 SAAnnot-C3Pap은 크게 3가지 과정으로 오픈포즈를 사용하여 자세를 추출하고, 추출된 부분 중 오류가 있는 부분을 슈퍼바이즐리를 사용하여 수정한 뒤, 오픈포즈와 슈퍼바이즐리의 결과값을 동기화하는 과정을 수행한다. 제안하는 방법을 통하여 오픈포즈에서 발생하는 잘못된 2차원 관절 위치 검출 결과를 교정할 수 있었고, 2명 이상의 사람을 검출하는 문제를 해결하였으며, 연주 자세 그라운드 트루스 획득이 가능하였다. 실험에서는 반자동 주석 방법인 오픈포즈와 본 논문에서 제안하는 SAAnnot-C3Pap의 결과를 비교·분석한다. 비교 결과, 제안하는 SAAnnot-C3Pap는 오픈포즈로 잘못 수집된 자세 정보를 개선한 결과를 보였다.

In this paper, we propose SAAnnot-C3Pap, a semi-automatic annotation method for obtaining ground truth of a player's posture. In order to obtain ground truth about the two-dimensional joint position in the existing music domain, openpose, a two-dimensional posture estimation method, was used or manually labeled. However, automatic annotation methods such as the existing openpose have the disadvantages of showing inaccurate results even though they are fast. Therefore, this paper proposes SAAnnot-C3Pap, a semi-automated annotation method that is a compromise between the two. The proposed approach consists of three main steps: extracting postures using openpose, correcting the parts with errors among the extracted parts using supervisely, and then analyzing the results of openpose and supervisely. Perform the synchronization process. Through the proposed method, it was possible to correct the incorrect 2D joint position detection result that occurred in the openpose, solve the problem of detecting two or more people, and obtain the ground truth in the playing posture. In the experiment, we compare and analyze the results of the semi-automated annotation method openpose and the SAAnnot-C3Pap proposed in this paper. As a result of comparison, the proposed method showed improvement of posture information incorrectly collected through openpose.

본 논문은 시맨틱 웹의 실현을 위해서는 필수적인 작업인 웹문서의 의미를 자동으로 태깅할 수 있는 시스템에 관한 것이다. 웹상의 방대한 자원을 일일이 사람이 수작업으로 의미를 태깅한다는 것은 사실상 불가능하기 때문에 한국어 웹문서를 대상으로 대량의 학습 데이터를 수집하고 자연어처리 기법과 시소러스를 이용하여 특징을 추출한 후 SVM 기계학습을 통하여 개념분류기를 구축하였다. 한국어의 특징을 파악하여 의미 태깅에 필요한 특징 정보를 추출하기 위해서 형태소 분석과 구문 분석을 하였다. 추출된 특징정보는 가도카와 시소러스의 의미코드를 이용하여 학습벡터로 구성되는데, 이는 유사한 단어나 구를 하나의 개념코드로 매핑하여 시스템의 재현율을 높이는 역할을 하게 된다. 실험결과 자동 의미 태깅 분야에서 본 접근방법의 가능성을 확인할 수 있었다.

This paper is about an system which can perform automatic semantic annotation to actualize "Semantic Web." Since it is impossible to tag numerous documents manually in the web, it is necessary to gather large Korean web documents as training data, and extract features by using natural language techniques and a thesaurus. After doing these, we constructed concept classifiers through the SVM (support vector machine) teaming algorithm. According to the characteristics of Korean language, morphological analysis and syntax analysis were used in this system to extract feature information. Based on these analyses, the concept code is mapped with Kadokawa thesaurus, which made it possible to map similar words and phrase to one concept code, to make training vectors. This contributed to rise the recall of our system. Results of the experiment show the system has a some possibility of semantic annotation.