Enterprise Organization of National Program of Excellence in Software

The 45th International ACM SIGIR Conference Research and Development in Information Retention (SIGIR) will be published in July 2022 with two papers from Heo Tae-hoon and Park Chung-won's master's degree in artificial intelligence and information search.

1. Choongwon Park, Youngjoong Ko, “QSG Transformer: Transformer with Query-Attentive Semantic Graph for Query-Focused Summarization”, Proceedings of the 45^th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2022), July 2022.

In this study, we propose a new technique to improve the performance of 'query-based document summaries' that generate query-friendly summaries in documents. The suggestion technique uses multiple natural language processing techniques to connect the words of a query and document to form a single graph, which is used to generate a summary. To efficiently utilize the constructed graph for query-based document summarization, we propose a new graph artificial neural network and paste it into a transformer model. Experiments using two datasets showed that the proposed technique outperformed both previous studies.

2. Taehun Huh and Youngjoong Ko, "Lightweight Meta-Learning for Low-Resource Abstractive Summarization", Proceedings of the 45^th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2022), July 2022

In this study, we propose a new technique to improve the performance of 'low resource generation summaries' with fewer labeled learning data. The proposed model uses meta-learning to quickly adapt to its domain using less data. It also addresses the problem of overfitting to less data by enabling only lightweight modules added to existing language models during learning. Experiments with a total of 11 summary datasets resulted in higher Rouge score performance than previous studies.

Ko Young-joong | yjko@skku.edu | Natural Language Processing Lab | http://nlp.skku.edu/