• KIISE Transactions on Computing Practices
• /
• v.24 no.3
• /
• pp.138-144
• /
• 2018

A video is a vivid medium similar to human's visual-linguistic experiences, since it can inculcate a sequence of situations, actions or dialogues that can be told as a story. In this study, we propose story learning/regeneration frameworks from videos with successive event order supervision for contextual coherence. The supervision induces each episode to have a form of trajectory in the latent space, which constructs a composite representation of ordering and semantics. In this study, we incorporated the use of kids videos as a training data. Some of the advantages associated with the kids videos include omnibus style, simple/explicit storyline in short, chronological narrative order, and relatively limited number of characters and spatial environments. We build the encoder-decoder structure with successive event order embedding, and train bi-directional LSTMs as sequence models considering multi-step sequence prediction. Using a series of approximately 200 episodes of kids videos named 'Pororo the Little Penguin', we give empirical results for story regeneration tasks and SEOE. In addition, each episode shows a trajectory-like shape on the latent space of the model, which gives the geometric information for the sequence models.

• Korean Journal of Cognitive Science
• /
• v.8 no.4
• /
• pp.19-33
• /
• 1997

Icon is a small graphic which represents a particular function in computer interface. Comprehension and learning of icons are needed for using them because they represent some meanings. This study was conducted to examine which were more important factors in using icons. First, representational characteristics of icons were examined. Second, using a eyetracker, the way of coding of redundant display was examined which removed obscurity of display and gave more meaning to a user. Icons using attribute of object and direct discription of action were more efficient than icons of other representational form in comprehension. Redundant display was proved to have some advantages, especially when the icons had low familarity. Visual information processing of icons were discussed according to familarity and learning.

• Korean Journal of Ecology and Environment
• /
• v.43 no.1
• /
• pp.161-167
• /
• 2010

In this study, we implemented an experimental approach of ecological model development in order to emphasize the importance of input variable selection with respect to time-delayed arrangement between input and output variables. Time-series modeling requires relevant input variable selection for the prediction of a specific output variable (e.g. density of a species). Inadequate variable utility for input often causes increase of model construction time and low efficiency of developed model when applied to real world representation. Therefore, for future prediction, researchers have to decide number of time-delay (e.g. months, weeks or days; t-n) to predict a certain phenomenon at current time t. We prepared a total of 3,900 equation models produced by Time-Series Optimized Genetic Programming (TSOGP) algorithm, for the prediction of monthly averaged density of a potamic phytoplankton species Stephanodiscus hantzschii, considering future prediction from 0- (no future prediction) to 12-months ahead (interval by 1 month; 300 equations per each month-delay). From the investigation of model structure, input variable selectivity was obviously affected by the time-delay arrangement, and the model predictability was related with the type of input variables. From the results, we can conclude that, although Machine Learning (ML) algorithms which have popularly been used in Ecological Informatics (EI) provide high performance in future prediction of ecological entities, the efficiency of models would be lowered unless relevant input variables are selectively used.

• Journal of KIISE:Software and Applications
• /
• v.31 no.12
• /
• pp.1649-1659
• /
• 2004

Text Generation is a process of generating comprehensible texts in human languages from some underlying non-linguistic representation of information. Among several sub-processes for text generation to generate coherent texts, this paper concerns referring expression generation which produces different types of expressions to refer to previously-mentioned things in a discourse. Specifically, we focus on pronominalization by zero pronouns which frequently occur in Korean. To build a generation model of referring expressions for Korean, several features are identified based on grammatical information and cost-based centering model, which are applied to various machine learning techniques. We demonstrate that our proposed features are well defined to explain pronominalization, especially pronominalization by zero pronouns in Korean, through 95 texts from three genres - Descriptive texts, News, and Short Aesop's Fables. We also show that our model significantly outperforms previous ones with a 99.9% confidence level by a T-test.

• Journal of Information Science Theory and Practice
• /
• v.9 no.1
• /
• pp.24-34
• /
• 2021

The popularity of social networking sites (SNS) has facilitated communication between users. The usage of SNS helps users in their daily life in various ways such as sharing of opinions, keeping in touch with old friends, making new friends, and getting information. However, some users misuse SNS to belittle or hurt others using profanities, which is typical in cyberbullying incidents. Thus, in this study, we aim to identify profane words from the ASKfm corpus to analyze the profane word distribution across four different roles involved in cyberbullying based on lexicon dictionary. These four roles are: harasser, victim, bystander that assists the bully, and bystander that defends the victim. Evaluation in this study focused on occurrences of the profane word for each role from the corpus. The top 10 common words used in the corpus are also identified and represented in a graph. Results from the analysis show that these four roles used profane words in their conversation with different weightage and distribution, even though the profane words used are mostly similar. The harasser is the first ranked that used profane words in the conversation compared to other roles. The results can be further explored and considered as a potential feature in a cyberbullying detection model using a machine learning approach. Results in this work will contribute to formulate the suitable representation. It is also useful in modeling a cyberbullying detection model based on the identification of profane word distribution across different cyberbullying roles in social networks for future works.

• The Mathematical Education
• /
• v.58 no.3
• /
• pp.459-481
• /
• 2019

This study aims to investigate students' statistical thinking through statistical processes in different instructional settings: Teacher-centered instruction vs. student-centered learning. We first developed instructional materials that allowed students to experience all the processes of statistics, including data collection, data analysis, data representation, and interpretation of the results. Using the instructional materials for four classes, we collected and analyzed the data from 57 seventh graders' discourse and artifacts from two different instructional settings using the analytic framework generated on the basis of literature review. The results showed that students felt difficulty particularly in the process of data collection and graph representations. In addition, even though data description has been heavily emphasized for data analysis in statistics education, it is surprisingly discovered that students had a hard time to understand the relationship between data and representations. Also, there were relationships between students' statistical thinking and instructional settings. Even though both groups of students showed difficulty in data collection and graph representations of the data, there were significant differences between the groups in terms of their performance. Whereas students from student-centered learning class outperformed in making decisions considering verification and justification, students from teacher-centered lecture class did better in problems requiring accuracy than the counterpart. The results from the study provide meaningful implications on developing curriculum and instructional methods for statistics education.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.8
• /
• pp.301-310
• /
• 2021

The TATI model is a Traffic Accident Text to RGB Image model, which is a methodology proposed in this paper for predicting the severity of traffic accidents. Traffic fatalities are decreasing every year, but they are among the low in the OECD members. Many studies have been conducted to reduce the death rate of traffic accidents, and among them, studies have been steadily conducted to reduce the incidence and mortality rate by predicting the severity of traffic accidents. In this regard, research has recently been active to predict the severity of traffic accidents by utilizing statistical models and deep learning models. In this paper, traffic accident dataset is converted to color images to predict the severity of traffic accidents, and this is done via CNN models. For performance comparison, we experiment that train the same data and compare the prediction results with the proposed model and other models. Through 10 experiments, we compare the accuracy and error range of four deep learning models. Experimental results show that the accuracy of the proposed model was the highest at 0.85, and the second lowest error range at 0.03 was shown to confirm the superiority of the performance.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.5
• /
• pp.296-302
• /
• 2021

This paper proposes an approach to the classification of respiratory states of humans based on visual information. An ultra-wide-band radar sensor acquired respiration signals, and the respiratory states were classified based on two-dimensional (2D) images instead of one-dimensional (1D) vectors. The 1D vector-based classification of respiratory states has limitations in cases of various types of normal respiration. The deep neural network model was employed for the classification, and the model learned the 2D images of respiration signals. Conventional classification methods use the value of the quantified respiration values or a variation of them based on regression or deep learning techniques. This paper used 2D images of the respiration signals, and the accuracy of the classification showed a 10% improvement compared to the method based on a 1D vector representation of the respiration signals. In the classification experiment, the respiration states were categorized into three classes, normal-1, normal-2, and abnormal respiration.

• Education of Primary School Mathematics
• /
• v.25 no.3
• /
• pp.233-249
• /
• 2022

Measurement plays an important role in both school mathematics and real life. Among the measurement areas, length is the first to learn and is the basis for measurement. Time is measured in its own way and is characterized by being the most abstract. This study attempted to analyze elementary mathematics textbooks and representations in real life to examine how the length and time of learning in school mathematics differ from those represented in real life. Based on this, we tried to derive implications for the direction of measurement education and elementary math textbooks. As a result of the analysis, the concept of length was used the same in real life and school mathematics. However, terms such as distance, depth, and height were not defined, and the representation of the approximate value was presented in a fragmentary form. In addition, there were parts where students were likely to feel confused in school mathematics and real life, such as the same units such as 'minutes and seconds' were used in time. Therefore, considering these differences, it is necessary to consider the direction of composition of math textbooks and teaching and learning so that students can connect school mathematics and real life and understand widely about measurement concepts.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.360-365
• /
• 2022

문장 표현(sentence representation)은 자연어처리 분야 내의 다양한 문제 해결 및 응용 개발에 있어 유용하게 활용될 수 있는 주요한 도구 중 하나이다. 하지만 최근 널리 도입되고 있는 사전 학습 언어 모델(pre-trained language model)로부터 도출한 문장 표현은 이방성(anisotropy)이 뚜렷한 등 그 고유의 특성으로 인해 문장 유사도(Semantic Textual Similarity; STS) 측정과 같은 태스크에서 기대 이하의 성능을 보이는 것으로 알려져 있다. 이러한 문제를 해결하기 위해 대조 학습(contrastive learning)을 사전 학습 언어 모델에 적용하는 연구가 문헌에서 활발히 진행되어 왔으며, 그중에서도 레이블이 없는 데이터를 활용하는 비지도 대조 학습 방법이 주목을 받고 있다. 하지만 대다수의 기존 연구들은 주로 영어 문장 표현 개선에 집중하였으며, 이에 대응되는 한국어 문장 표현에 관한 연구는 상대적으로 부족한 실정이다. 이에 본 논문에서는 대표적인 비지도 대조 학습 방법(ConSERT, SimCSE)을 다양한 한국어 사전 학습 언어 모델(KoBERT, KR-BERT, KLUE-BERT)에 적용하여 문장 유사도 태스크(KorSTS, KLUE-STS)에 대해 평가하였다. 그 결과, 한국어의 경우에도 일반적으로 영어의 경우와 유사한 경향성을 보이는 것을 확인하였으며, 이에 더하여 다음과 같은 새로운 사실을 관측하였다. 첫째, 사용한 비지도 대조 학습 방법 모두에서 KLUE-BERT가 KoBERT, KR-BERT보다 더 안정적이고 나은 성능을 보였다. 둘째, ConSERT에서 소개하는 여러 데이터 증강 방법 중 token shuffling 방법이 전반적으로 높은 성능을 보였다. 셋째, 두 가지 비지도 대조 학습 방법 모두 검증 데이터로 활용한 KLUE-STS 학습 데이터에 대해 성능이 과적합되는 현상을 발견하였다. 결론적으로, 본 연구에서는 한국어 문장 표현 또한 영어의 경우와 마찬가지로 비지도 대조 학습의 적용을 통해 그 성능을 개선할 수 있음을 검증하였으며, 이와 같은 결과가 향후 한국어 문장 표현 연구 발전에 초석이 되기를 기대한다.