Earticle

Machine reading comprehension (MRC) is a fundamental task in natural language processing (NLP), with existing models struggling to capture long-range dependencies and handle complex semantic nuances, particularly in Chinese. This paper proposes the Collaborative Semantic Reader (C-S Reader), a novel model that combines RoBERTa_wwm_ext pre-training and multi-level attention mechanisms to enhance semantic understanding. Experiments on the DuReader2 dataset show that C-S Reader significantly outperforms baseline models in both the Rouge-L and BLEU-4 scores, demonstrating its effectiveness in processing long documents and capturing complex semantic relationships. Our work provides a scalable solution for Chinese MRC tasks and highlights future challenges, including long-range dependency modeling and ambiguity in complex questions.

과거에는 동물들이 가족 구성원으로 받아들여지기보다는 가축으로 활용되었지만, 시간이 지나면서 동물들의 생활 환경이 집 안으로 바뀌면서 반려동물로서 인식되는 경향이 커지고, 반려동물 양육 인구도 증가하고 있다. 이러한 변 화로 반려동물은 가족 구성원으로 여겨지고, 그들의 건강과 복지에 관한 관심도 증가하고 있다. 반려동물의 건강은 적절한 식이 관리를 통해 유지될 수 있다. 이를 위해서는 사료의 다양성과 영양성을 고려해야 한다. 최근 연구에서 는 반려동물의 장내 미생물 환경과 면역 체계 조절, 소화 기능 등과의 관계가 밝혀지고 있어 사료 선택의 중요성이 부각 되고 있다. 반려동물의 장내 미생물 환경은 사료의 성분과 밀접한 연관이 있으며, 적절한 식이조절을 통해 장 내 미생물 환경을 조성한다. 이러한 사실은 반려동물의 건강을 유지하기 위한 적절한 식이 선택을 통해 장내 미생물 환경 조성의 중요성을 강조한다. 따라서 본 연구는 반려동물의 건강과 사료 선택 사이의 연관성을 고려하여 반려동 물의 장내 미생물 환경과 관련된 논문을 수집하고, 여러 언어 모델을 결합한 앙상블 모델을 사용한 기계 독해를 통 해 반려동물 상태에 맞춤형 사료를 추천하는 시스템을 개발한다. 이 시스템은 반려동물의 건강에 도움이 되는 사료 선택을 돕고, 반려인들에게 정확하고 유용한 사료를 추천한다. 이 연구 결과는 반려동물의 건강과 행복한 삶을 위한 실질적인 가치를 가지며, 반려동물 사료 시장에 유용한 지침을 제공할 것으로 기대한다.

In the past, animals were primarily considered livestock rather than family members. However, as time has passed, the living environment of animals has shifted indoors, leading to an increasing trend in perceiving them as pets. This transformation has led to a growing population of pet owners, who now view their animals as integral family members, and consequently, there is a heightened interest in the health and well-being of these beloved companions. The health of pets can be maintained through proper dietary management, and this necessitates careful consideration of the variety and nutritional content of pet food. Recent research has unveiled the intricate relationship between a pet's gut microbiome, immune system regulation, and digestive functions, shedding light on the pivotal role of dietary choices. The gut microbiome of pets is closely intertwined with the composition of pet food, emphasizing the importance of cultivating a suitable gut environment through appropriate dietary adjustments. These insights emphasize the significance of selecting the right diet for maintaining a pet's health and underline the importance of fostering a favorable gut microbiome. Consequently, this research takes into account the correlation between pet health and food selection. It seeks to collect papers related to pets' gut microbiome, and through machine comprehension employing an ensemble model that combines various language models, the study aims to develop a system for recommending tailored pet food based on the pet's condition. This system is designed to facilitate the selection of food that contributes to the well-being of pets and offers pet owners accurate and valuable food recommendations. The outcomes of this research are expected to hold substantial practical value, as they will contribute to the health and happiness of pets, while also providing useful guidelines for the pet food market.

The smart tourism chatbot service provide smart tourism services to users easily and conveniently along with the smart tourism app. In this paper, the tourism information QA (Question Answering) service is proposed based on the task-oriented smart tourism chatbot system [13]. The tourism information QA service is an MRC (Machine reading comprehension)-based QA system that finds answers in context and provides them to users. The tourism information QA system consists of NER (Named Entity Recognition), DST (Dialogue State Tracking), Neo4J graph DB, and QA servers. We propose tourism information QA service uses the tourism information NER model and DST model to identify the intent of the user's question and retrieves appropriate context for the answer from the Neo4J tourism knowledgebase. The QA model finds answers from the context and provides them to users through the smart tourism app. We develop the tourism information QA model by transfer learning the bigBird model, which can process the context of 4,096 tokens, using the tourism information QA dataset.

Machine reading comprehension is the task of understanding a given context and finding the correct response in that context. A simple recurrent unit (SRU) is a model that solves the vanishing gradient problem in a recurrent neural network (RNN) using a neural gate, such as a gated recurrent unit (GRU) and long short-term memory (LSTM); moreover, it removes the previous hidden state from the input gate to improve the speed compared to GRU and LSTM. A self-matching network, used in R-Net, can have a similar effect to coreference resolution because the self-matching network can obtain context information of a similar meaning by calculating the attention weight for its own RNN sequence. In this paper, we construct a dataset for Korean machine reading comprehension and propose an $S^2-Net$ model that adds a self-matching layer to an encoder RNN using multilayer SRU. The experimental results show that the proposed $S^2-Net$ model has performance of single 68.82% EM and 81.25% F1, and ensemble 70.81% EM, 82.48% F1 in the Korean machine reading comprehension test dataset, and has single 71.30% EM and 80.37% F1 and ensemble 73.29% EM and 81.54% F1 performance in the SQuAD dev dataset.

We propose the VS³-NET model to solve the task of question answering questions with machine-reading comprehension that searches for an appropriate answer in a given context. VS³-NET is a model that trains latent variables for each question using variational inferences based on a model of a simple recurrent unit-based sentences and self-matching networks. The types of questions vary, and the answers depend on the type of question. To perform efficient inference and learning, we introduce neural question-type models to approximate the prior and posterior distributions of the latent variables, and we use these approximated distributions to optimize a reparameterized variational lower bound. The context given in machine-reading comprehension usually comprises several sentences, leading to performance degradation caused by context length. Therefore, we model a hierarchical structure using sentence encoding, in which as the context becomes longer, the performance degrades. Experimental results show that the proposed VS³-NET model has an exact-match score of 76.8% and an F1 score of 84.5% on the SQuAD test set.

The study aims to examine whether a machine translator can be effective in promoting learners’ performance in reading comprehension. The main reason for using a machine translator, in this case Papago, was to increase students’ engagement in order to better comprehend the content of selected reading materials. Twenty-seven students were assigned to a wordlist group while thirty-four were assigned to a Papago group. The wordlist group was provided with a wordlist before reading the main texts while the Papago group used Papago. Both groups took a reading comprehension test before and after the experiment. The test comprised 20 reading comprehension and vocabulary questions. A significant difference in reading performance was found between the wordlist group and the Papago group. Furthermore, the results of the survey revealed that the participants’ perspectives in both groups showed significant differences in all items. Open-ended responses demonstrated the benefits and drawbacks of utilizing the wordlists and Papago. Consequently, the findings of this study indicate that students preferred the wordlists to Papago for understanding the reading texts. That is, using machine translators may not be useful in L2 reading. In this light, several pedagogical implications for designing and using machine translators are proposed.

오픈 도메인 기계독해는 질문과 연관된 단락이 존재하지 않아 단락을 검색하는 검색 기능을 추가한 모델이다. 문서 검색은 단어 빈도 기반인 TF-IDF로 많은 연구가 진행됐으나 문서의 양이 많아지면 낮은 성능을 보이는 문제가 있다. 아울러 단락 선별은 단어 기반 임베딩으로 많은 연구가 진행됐으나 문장의 특징을 가지는 단락의 문맥을 정확히 추출하지 못하는 문제가 있다. 그리고 문서 독해는 BERT로 많은 연구가 진행됐으나 방대한 파라미터로 느린 학습 문제를 보였다. 본 논문에서는 언급한 3가지 문제를 해결하기 위해 문서의 길이까지 고려한 BM25를 이용하며 문장 문맥을 얻기 위해 InferSent를 사용하고, 파라미터 수를 줄이기 위해 ALBERT를 이용한 오픈 도메인 기계독해를 제안한다. SQuAD1.1 데이터셋으로 실험을 진행했다. 문서 검색은 BM25의 성능이 TF-IDF보다 3.2% 높았다. 단락 선별은 InferSent가 Transformer보다 0.9% 높았다. 마지막으로 문서 독해에서 단락의 수가 증가하면 ALBERT가 EM에서 0.4%, F1에서 0.2% 더 높았다.

An open domain machine reading comprehension is a model that adds a function to search paragraphs as there are no paragraphs related to a given question. Document searches have an issue of lower performance with a lot of documents despite abundant research with word frequency based TF-IDF. Paragraph selections also have an issue of not extracting paragraph contexts, including sentence characteristics accurately despite a lot of research with word-based embedding. Document reading comprehension has an issue of slow learning due to the growing number of parameters despite a lot of research on BERT. Trying to solve these three issues, this study used BM25 which considered even sentence length and InferSent to get sentence contexts, and proposed an open domain machine reading comprehension with ALBERT to reduce the number of parameters. An experiment was conducted with SQuAD1.1 datasets. BM25 recorded a higher performance of document research than TF-IDF by 3.2%. InferSent showed a higher performance in paragraph selection than Transformer by 0.9%. Finally, as the number of paragraphs increased in document comprehension, ALBERT was 0.4% higher in EM and 0.2% higher in F1.

Most of the existing machine reading research has used Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) algorithms as networks. Among them, RNN was slow in training, and Question Answering Network (QANet) was announced to improve training speed. QANet is a model composed of CNN and self-attention. CNN extracts semantic and syntactic information well from the local corpus, but there is a limit to extracting the corresponding information from the global corpus. Graph Convolutional Networks (GCN) extracts semantic and syntactic information relatively well from the global corpus. In this paper, to take advantage of this strength of GCN, we propose I-QANet, which changed the CNN of QANet to GCN. The proposed model performed 1.2 times faster than the baseline in the Stanford Question Answering Dataset (SQuAD) dataset and showed 0.2% higher performance in Exact Match (EM) and 0.7% higher in F1. Furthermore, in the Korean Question Answering Dataset (KorQuAD) dataset consisting only of Korean, the learning time was 1.1 times faster than the baseline, and the EM and F1 performance were also 0.9% and 0.7% higher, respectively.

거대 언어 모델에서 일부 특정 도메인에 대한 적응력 부족을 향상시킬 수 있는 기법인 파인튜닝(Fine-tuning)과 검색 증강 생성(RAG)의 장단점을 비교하는 연구가 활발히 진행되고 있다. 그러나 이러한 기존 연구는 질문과 정답의 출처 문서 쌍이 이미 입력된 환경에서도 여전히 RAG가 파인튜닝에 비해 특정 도메인의 성능을 손쉽게 향상시킬 수 있는 잠재력이 있는지에 대한 비교가 없었다. 이에 본 연구에서는 출처 문서가 이미 입력된 환경에서 소형 언어 모델(sLLM)을 활용한 한국어 기계독해(KorQuAD)와 감성 분석 작업(NSMC)에서 파인튜닝과 RAG의 성능을 비교하였다. 실험 결과, RAG는 기계독해에서 10.2%, 감성분석에서 32.3%의 성능 향상을 보였으며, 파인튜닝과 병행할 경우 각각 11.5%와 41.9%의 성능이 향상되었다. 이를 바탕으로, 컴퓨팅 자원이 부족하면 RAG를, 충분하다면 두 기법의 상호 보완 방식으로 성능을 향상시키는 방안을 제시하였다.

Studies have actively compared fine-tuning and retrieval-augmented generation (RAG) to enhance the adaptability of large language models (LLMs) to specific domains. Although question and source document pairs are already available in certain environments, no comparative study has assessed whether RAG has the potential to improve domain-specific performance more easily than fine-tuning in such settings. In this study, we compare the performance of fine-tuning and RAG for Korean machine reading comprehension (KorQuAD) and sentiment analysis (NSMC) tasks using small language models (sLLM). Our results show that RAG improved performance by 10.2% on KorQuAD and 32.3% on NSMC. when RAG and fine-tuning were combined, performance improved by 11.5% and 41.9%, respectively. Our results indicate that RAG is advantageous when resources are limited, while a complementary use of both methods can outperform fine-tuning when sufficient resources are available.