• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.41-46
• /
• 2021

기계 독해는 문단과 질문이 주어질 때에 정답을 맞추는 자연어처리의 연구분야다. 최근 기계 독해 모델이 사람보다 높은 성능을 보여주고 있지만, 문단과 질의가 크게 변하지 않더라도 예상과 다른 결과를 만들어 성능에 영향을 주기도 한다. 본 논문에서는 문단과 질문 두 가지 관점에서 적대적 예시 데이터를 사용하여 보다 강건한 질의응답 모델을 훈련하는 방식을 제안한다. 트랜스포머 인코더 모델을 활용하였으며, 데이터를 생성하기 위해서 KorQuAD 1.0 데이터셋에 적대적 예시를 추가하여 실험을 진행하였다. 적대적 예시를 이용한 데이터로 실험한 결과, 기존 모델보다 1% 가량 높은 성능을 보였다. 또한 질의의 적대적 예시 데이터를 활용하였을 때, 기존 KorQuAD 1.0 데이터에 대한 성능 향상을 확인하였다.

• Journal of The Korean Association For Science Education
• /
• v.31 no.2
• /
• pp.225-238
• /
• 2011

The purpose of this study was to investigate the effects of the instruction with question-creation training, compared with traditional science instructions. The instruction with question-creation training is to give students chances to make questions by themselves based on what they learned before the end of the classes. The four effects of the instruction were studied: students' achievement, metacognition, self-efficacy, and the level of the questions created by the students according to different proficiency levels. Research data was gathered from 65 second grade students at a middle school in Busan. The comparative group was instructed in traditional lecture-type teaching method. The experimental group was instructed with questioncreation training. Students in the experimental group were asked to make 3 questions by themselves and then, to solve their peers' questions about 15 minutes before the end of the classes. Both groups were divided into 3 groups by proficiency level according to the results of last semester's science test. Before the research, a metacognition test and a self-efficacy test were conducted. After the research, an achievement test, a question level test, a metacognition test, and a self-efficacy test were conducted and analyzed by t-test. The research data for question level was analyzed by one-way ANCOVA. The results of this study revealed that question-creation training has a positive effect on student's achievement, metacognition, and self-efficiency. It also showed most of the students have gained an ability to make higher-level questions regardless of their proficiency level due to the increased number of students who made higher-level questions. It also showed that most of the students could gain an ability to make higher-level questions regardless of their proficiency level from the fact that the number of students who made higher-level questions increased in every proficiency level.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2024.01a
• /
• pp.185-186
• /
• 2024

본 논문에서는 CNN 알고리즘을 사용한 수학 문제 답지 추론 모델에 대한 소개를 다룬다. 현재의 학습 보조 서비스 중에서도 질문에 답하는 서비스들이 흔하지만, 수학 문제에 특화된 이미지 기반 답지 추론 서비스는 부족한 상황이다. 본 논문에서는 MathDataset 클래스를 활용하여 수학 문제 이미지와 정답을 연결하는 데이터셋을 생성하고, CNN 알고리즘을 사용하여 모델을 훈련하는 방법을 제시한다.

• The Korean Journal of Applied Statistics
• /
• v.34 no.2
• /
• pp.191-204
• /
• 2021

Visual Question Answering (VQA) and image captioning are tasks that require understanding of the features of images and linguistic features of text. Therefore, co-attention may be the key to both tasks, which can connect image and text. In this paper, we propose a model to achieve high performance for VQA by image caption generated using a pretrained standard transformer model based on MSCOCO dataset. Captions unrelated to the question can rather interfere with answering, so some captions similar to the question were selected to use based on a similarity to the question. In addition, stopwords in the caption could not affect or interfere with answering, so the experiment was conducted after removing stopwords. Experiments were conducted on VQA-v2 data to compare the proposed model with the deep modular co-attention network (MCAN) model, which showed good performance by using co-attention between images and text. As a result, the proposed model outperformed the MCAN model.