A5033984182- Topic Modeling
- Natural Language Processing Techniques
- Speech and dialogue systems
- Multimodal Machine Learning Applications
- Assistive Technology in Communication and Mobility
- Advanced Text Analysis Techniques
- Advanced Graph Neural Networks
- Text and Document Classification Technologies
- Technology and Data Analysis
- Semantic Web and Ontologies
- Digital Accessibility for Disabilities
- Computational and Text Analysis Methods
- Robotics and Automated Systems
- Recommender Systems and Techniques
- Teaching and Learning Programming
- Neurobiology of Language and Bilingualism
- Speech Recognition and Synthesis
- Sentiment Analysis and Opinion Mining
- Language Development and Disorders
- Educational Games and Gamification
- Innovation in Digital Healthcare Systems
- Artificial Intelligence in Games
Korea University
2019-2023
We focus on multi-turn response selection in a retrieval-based dialog system.In this paper, we utilize the powerful pre-trained language model Bi-directional Encoder Representations from Transformer (BERT) for system and propose highly effective post-training method domain-specific corpus.Although BERT is easily adopted to various NLP tasks outperforms previous baselines of each task, it still has limitations if task corpus too focused certain domain.Posttraining (e.g., Ubuntu Corpus) helps...
CommonsenseQA is a task in which correct answer predicted through commonsense reasoning with pre-defined knowledge. Most previous works have aimed to improve the performance distributed representation without considering process of predicting from semantic question. To shed light upon interpretation question, we propose an AMR-ConceptNet-Pruned (ACP) graph. The ACP graph pruned full integrated encompassing Abstract Meaning Representation (AMR) generated input questions and external knowledge...
We propose a contextual emotion classifier based on transferable language model and dynamic max pooling, which predicts the of each utterance in dialogue. A representative analysis task, EmotionX, requires to consider information from colloquial dialogues deal with class imbalance problem. To alleviate these problems, our leverages self-attention weighted cross entropy loss. Furthermore, we apply post-training fine-tuning mechanisms enhance domain adaptability utilize several machine...
Language model pretraining is an effective method for improving the performance of downstream natural language processing tasks. Even though modeling unsupervised and thus collecting data it relatively less expensive, still a challenging process languages with limited resources. This results in great technological disparity between high- low-resource numerous In this paper, we aim to make technology more accessible by enabling efficient training pretrained models. It achieved formulating as...
Abstractive dialogue summarization is a challenging task for several reasons. First, most of the important pieces information in conversation are scattered across utterances through multi-party interactions with different textual styles. Second, dialogues often informal structures, wherein individuals express personal perspectives, unlike text summarization, tasks that usually target formal documents such as news articles. To address these issues, we focused on association between from...
Despite the striking advances in recent language generation performance, model-generated responses have suffered from chronic problem of hallucinations that are either untrue or unfaithful to a given source. Especially task knowledge grounded conversation, models required generate informative responses, but hallucinated utterances lead miscommunication. In particular, entity-level hallucination causes critical misinformation and undesirable conversation is one major concerns. To address this...
앙상블 기법은 여러 모델을 종합하여 최종 판단을 산출하는 기계 학습 기법으로서 딥러닝 모델의 성능 향상을 보장한다. 하지만 대부분의 앙상블만을 위한 추가적인 모델 또는 별도의 연산을 요구한다. 이에 우리는 기법을 교차 검증 방법과 결합하여 비용을 줄이며 일반화 성능을 높이는 제안한다. 본 기법의 효과를 입증하기 위해 MRPC, RTE 데이터셋과 BiLSTM, CNN, ELMo, BERT 이용하여 기존 기법보다 향상된 보인다. 추가로 검증에서 비롯한 원리와 변수에 따른 변화에 대하여 논의한다.
This work presents KoBigBird-large, a large size of Korean BigBird that achieves state-of-the-art performance and allows long sequence processing for language understanding. Without further pretraining, we only transform the architecture extend positional encoding with our proposed Tapered Absolute Positional Encoding Representations (TAPER). In experiments, KoBigBird-large shows overall on understanding benchmarks best document classification question answering tasks longer sequences...
Yoonna Jang, Suhyune Son, Jeongwoo Lee, Junyoung Yuna Hur, Jungwoo Lim, Hyeonseok Moon, Kisu Yang, Heuiseok Lim. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.
Kisu Yang. Proceedings of the 13th International Joint Conference on Natural Language Processing and 3rd Asia-Pacific Chapter Association for Computational Linguistics (Volume 1: Long Papers). 2023.
We focus on multi-turn response selection in a retrieval-based dialog system. In this paper, we utilize the powerful pre-trained language model Bi-directional Encoder Representations from Transformer (BERT) for system and propose highly effective post-training method domain-specific corpus. Although BERT is easily adopted to various NLP tasks outperforms previous baselines of each task, it still has limitations if task corpus too focused certain domain. Post-training (e.g., Ubuntu Corpus)...
CommonsenseQA is a task in which correct answer predicted through commonsense reasoning with pre-defined knowledge. Most previous works have aimed to improve the performance distributed representation without considering process of predicting from semantic question. To shed light upon interpretation question, we propose an AMR-ConceptNet-Pruned (ACP) graph. The ACP graph pruned full integrated encompassing Abstract Meaning Representation (AMR) generated input questions and external knowledge...
Abstractive dialogue summarization is a challenging task for several reasons. First, most of the important pieces information in conversation are scattered across utterances through multi-party interactions with different textual styles. Second, dialogues often informal structures, wherein individuals express personal perspectives, unlike text summarization, tasks that usually target formal documents such as news articles. To address these issues, we focused on association between from...