• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.3
• /
• pp.91-100
• /
• 2020

Unlike the existing Visual Question Answering(VQA) problems, the new Visual Commonsense Reasoning(VCR) problems require deep common sense reasoning for answering questions: recognizing specific relationship between two objects in the image, presenting the rationale of the answer. In this paper, we propose a novel deep neural network model, KG_VCR, for VCR problems. In addition to make use of visual relations and contextual information between objects extracted from input data (images, natural language questions, and response lists), the KG_VCR also utilizes commonsense knowledge embedding extracted from an external knowledge base called ConceptNet. Specifically the proposed model employs a Graph Convolutional Neural Network(GCN) module to obtain commonsense knowledge embedding from the retrieved ConceptNet knowledge graph. By conducting a series of experiments with the VCR benchmark dataset, we show that the proposed KG_VCR model outperforms both the state of the art(SOTA) VQA model and the R2C VCR model.

• School Mathematics
• /
• v.19 no.2
• /
• pp.345-367
• /
• 2017

The purpose of this study is to investigate the proportional reasoning of middle school students during the process of expressing and interpreting the graphs. We collected data from a teaching experiment with four 7th grade students who participated in 23 teaching episodes. For this study, the differences between student A and student B-who joined theteaching experiment from the $1^{st}$ teaching episode through the $8^{th}$ -in understanding graphs are compared and the reason for their differences are discussed. The results showed different proportional solving strategies between the two students, which revealed in the course of adjusting values of two given variables to seek new values; student B, due to a limited ability for proportional reasoning, had difficulty in constructing graphs for given situations and interpreting given graphs.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.71-75
• /
• 2020

오픈북 질의응답 문제는 올바른 정답을 고르기 위해 사람들끼리 공유하고 있는 상식정보가 필요한 질의로 이루어져있다. 기계가 사람과 달리 상식 정보를 이용하여 결론을 도출하는 상식 추론을 하기 위해서는 적절한 상식 정보를 논리적으로 사용하여야 한다. 본 연구에서는 적절한 상식정보의 선택과 논리적 추론을 위하여, 질의에 대한 Abstract Meaning Representation (AMR) 그래프를 이용하여 적절한 상식 정보를 선택하고 그의 해석을 용이하게 만들었다. 본 연구에서 제안한 상식 그래프 경로 학습 모델은 오픈북 질의응답 문제에서 대표적 언어모델인 BERT의 성능보다 약 7%p 높은 55.02%의 정확도를 달성하였다.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.9
• /
• pp.39-48
• /
• 2024

In this study, we propose a method to augment the provided reasoning paths to improve the answer performance and explanatory power of KGQA. In the proposed method, we utilize LLMs and GNNs to retrieve reasoning paths related to the question from the knowledge graph and evaluate reasoning paths. Then, we retrieve the external information related to the question and then converted into triples to answer the question and explain the reason. Our method evaluates the reasoning path by checking inference results and semantically by itself. In addition, we find related texts to the question based on their similarity and converting them into triples of knowledge graph. We evaluated the performance of the proposed method using the WebQuestion Semantic Parsing dataset, and found that it provides correct answers with higher accuracy and more questions with explanations than the reasoning paths by the previous research.

• Journal of KIISE:Software and Applications
• /
• v.36 no.11
• /
• pp.928-935
• /
• 2009

Ontology schema reasoning is used to maintain consistency of concepts and build concept hierarchy automatically. For the purpose, the search of concepts must be inevitably performed. Ontology schema reasoning performs the test of subsumption relationships of all the concepts delivered in the test set. The result of subsumption tests is determined based on the creation of complete graphs, which seriously weighs with the performance of reasoning. In general, the process of creating complete graph has been known as expressive procedure. This process is essential in improving the leading performance. In this paper, we propose a method enhancing the classification performance by identifying unnecessary subsumption test supported by optimized searching method on subsumption relationship test among concepts. It is achieved by propagating subsumption tests results into other concept.

• Annual Conference of KIPS
• /
• 2022.05a
• /
• pp.557-559
• /
• 2022

최근 딥러닝의 기술발전으로 자연어 처리 분야에서 Q&A, 문장추천, 개체명 인식 등 다양한 연구가 진행 되고 있다. 딥러닝 기반 자연어 처리에서 좋은 성능을 보이는 트랜스포머 기반 BERT 모델의 성능향상에 대한 다양한 연구도 함께 진행되고 있다. 본 논문에서는 토픽모델인 잠재 디리클레 할당을 이용한 토픽별 지식그래프 분류와 입력문장의 토픽을 추론하는 방법으로 K-BERT 모델을 학습한다. 분류된 토픽 지식그래프와 추론된 토픽을 이용해 K-BERT 모델에서 대용량 지식그래프 사용의 효율적 방법을 제안한다.

• Journal of the Korean Society for information Management
• /
• v.34 no.1
• /
• pp.177-195
• /
• 2017

This study attempts to infer research fronts using factor graph model based on heterogeneous features. The model suggested by this study infers research fronts having documents with the potential to be cited multiple times in the future. To this end, the documents are represented by bibliographic, network, and content features. Bibliographic features contain bibliographic information such as the number of authors, the number of institutions to which the authors belong, proceedings, the number of keywords the authors provide, funds, the number of references, the number of pages, and the journal impact factor. Network features include degree centrality, betweenness, and closeness among the document network. Content features include keywords from the title and abstract using keyphrase extraction techniques. The model learns these features of a publication and infers whether the document would be an RF using sum-product algorithm and junction tree algorithm on a factor graph. We experimentally demonstrate that when predicting RFs, the FG predicted more densely connected documents than those predicted by RFs constructed using a traditional bibliometric approach. Our results also indicate that FG-predicted documents exhibit stronger degrees of centrality and betweenness among RFs.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.8
• /
• pp.319-330
• /
• 2021

Recently, in the field of open domain natural language question answering, multi-task, multi-hop question answering has been studied extensively. In this paper, we propose a novel deep neural network model using hierarchical graphs to answer effectively such multi-task, multi-hop questions. The proposed model extracts different levels of contextual information from multiple paragraphs using hierarchical graphs and graph neural networks, and then utilize them to predict answer type, supporting sentences and answer spans simultaneously. Conducting experiments with the HotpotQA benchmark dataset, we show high performance and positive effects of the proposed model.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• v.9 no.2
• /
• pp.566-570
• /
• 2005

생물학적 유전자 배열에서 다양한 레벨로 분자 세포 간 네트워크를 입증하여 고 처리를 응용하여 수치학적인 표현 모델 분석으로 정보공학 네트워크를 연구한다. 확률적 그래프 모델을 사용하여 네트워크의 계층적 구성 특성을 이용하여 생물학적 통찰력을 확률함수를 응용해 복잡한 세포 간 네트워크에 대한 고 대역 처리 데이터의 근원인 DNA 마이크로 배열을 응용하여 유전자 베이스네트워크 논리를 유전자 표현 레벨로 나타낸다. 유전자 데이터로부터 확률적 그래프 모델들을 추정 및 분석하고 논리적으로 예측하여 확률적 그래프 모델이 정보공학 네트워크로 확장 추론 한다.

• Annual Conference of KIPS
• /
• 2019.10a
• /
• pp.994-997
• /
• 2019

영상 기반 상식 추론(VCR) 문제는 기존의 영상 기반 질문-응답(VQA) 문제들과는 달리, 영상에 포함된 사물들 간의 관계 파악과 답변 근거 제시 등 별도의 상식 추론이 요구되는 새로운 지능 문제이다. 본 논문에서는 입력 데이터(영상, 자연어 질문, 응답 리스트)에서 사물들 간의 관계와 맥락 정보를 추출해내는 모듈들 외에, 별도로 ConceptNet과 같은 외부 지식 베이스로부터 관련 상식들을 직접 가져다 GCN 기반의 지식 그래프 임베딩 과정을 거쳐 추가적으로 활용할 수 있는 모듈들을 포함한 새로운 심층 신경망 모델인 KG_VCR을 제안한다. 제안 모델인 KG_VCR의 세부 설계사항들을 소개하고, VCR 벤치마크 데이터 집합을 이용한 다양한 실험들을 통해 제안 모델의 성능을 입증한다.