RGB images provide a practical and cost-effective method of caries detection. However, the ambiguity of RGB caries images may lead to labeling errors during annotation, which can reduce the robustness of caries classification models. To address this, we propose Core Inter-Category Contrastive Learning (CICC) to improve the robustness of caries classification models. Rather than relying on traditional network fine-tuning, CICC focuses on improving the robustness of the model to label errors from a novel perspective by identifying core data that are highly relevant to the caries category. CICC utilizes the Jensen-Shannon Divergence to select core data, mitigating the impact of label errors on model performance. Inter-Category Contrastive Learning enhances feature representations of samples from different categories to improve the model's discrimination between caries categories. We validated the effectiveness of CICC in improving model robustness from model optimization and experimental results. Extensive experiments demonstrate that CICC significantly outperforms other comparative methods in caries classification performance and robustness. Our code is available at: https://github.com/papercode-for-cheung/CICC.

Abstract The automatic summarization of scientific articles differs from other text genres because of the structured format and longer text length. Previous approaches have focused on tackling the lengthy nature of scientific articles, aiming to improve the computational efficiency of summarizing long text using a flat, unstructured abstract. However, the structured format of scientific articles and characteristics of each section have not been fully explored, despite their importance. The lack of a sufficient investigation and discussion of various characteristics for each section and their influence on summarization results has hindered the practical use of automatic summarization for scientific articles. To provide a balanced abstract proportionally emphasizing each section of a scientific article, the community introduced the structured abstract, an abstract with distinct, labeled sections. Using this information, in this study, we aim to understand tasks ranging from data preparation to model evaluation from diverse viewpoints. Specifically, we provide a preprocessed large‐scale dataset and propose a summarization method applying the introduction, methods, results, and discussion (IMRaD) format reflecting the characteristics of each section. We also discuss the objective benchmarks and perspectives of state‐of‐the‐art algorithms and present the challenges and research directions in this area.

Abstract Coauthorship prediction applies predictive analytics to bibliographic data to predict authors who are highly likely to be coauthors. In this study, we propose an approach for coauthorship prediction based on bibliographic network embedding through a graph‐based bibliographic data model that can be used to model common bibliographic data, including papers, terms, sources, authors, departments, research interests, universities, and countries. A real‐world dataset released by AMiner that includes more than 2 million papers, 8 million citations, and 1.7 million authors were integrated into a large bibliographic network using the proposed bibliographic data model. Translation‐based methods were applied to the entities and relationships to generate their low‐dimensional embeddings while preserving their connectivity information in the original bibliographic network. We applied machine learning algorithms to embeddings that represent the coauthorship relationships of the two authors and achieved high prediction results. The reference model, which is the combination of a network embedding size of 100, the most basic translation‐based method, and a gradient boosting method achieved an F1 score of 0.9 and even higher scores are obtainable with different embedding sizes and more advanced embedding methods. Thus, the strengths of the proposed approach lie in its customizable components under a unified framework.

Abstract This study examines the relationship between team's gender composition and outputs of funded projects using a large data set of National Institutes of Health (NIH) R01 grants and their associated publications between 1990 and 2017. This study finds that while the women investigators' presence in NIH grants is generally low, higher women investigator presence is on average related to slightly lower number of publications. This study finds empirically that women investigators elect to work in fields in which fewer publications per million‐dollar funding is the norm. For fields where women investigators are relatively well represented, they are as productive as men. The overall lower productivity of women investigators may be attributed to the low representation of women in high productivity fields dominated by men investigators. The findings shed light on possible reasons for gender disparity in grant productivity.

The objective of this paper is to systematically assess sources' (e.g., journals and proceedings) impact in knowledge trading. While there have been efforts at evaluating different aspects of journal impact, the dimension of knowledge trading is largely absent. To fill the gap, this study employed a set of trading‐based indicators, including weighted degree centrality, Shannon entropy, and weighted betweenness centrality, to assess sources' trading impact. These indicators were applied to several time‐sliced source‐to‐source citation networks that comprise 33,634 sources indexed in the Scopus database. The results show that several interdisciplinary sources, such as Nature, PLoS One, Proceedings of the National Academy of Sciences, and Science, and several specialty sources, such as Lancet, Lecture Notes in Computer Science, Journal of the American Chemical Society, Journal of Biological Chemistry, and New England Journal of Medicine, have demonstrated their marked importance in knowledge trading. Furthermore, this study also reveals that, overall, sources have established more trading partners, increased their trading volumes, broadened their trading areas, and diversified their trading contents over the past 15 years from 1997 to 2011. These results inform the understanding of source‐level impact assessment and knowledge diffusion.

In this article, we propose a graph‐based interactive bibliographic information retrieval system—GIBIR. GIBIR provides an effective way to retrieve bibliographic information. The system represents bibliographic information as networks and provides a form‐based query interface. Users can develop their queries interactively by referencing the system‐generated graph queries. Complex queries such as “papers on information retrieval, which were cited by John's papers that had been presented in SIGIR” can be effectively answered by the system. We evaluate the proposed system by developing another relational database‐based bibliographic information retrieval system with the same interface and functions. Experiment results show that the proposed system executes the same queries much faster than the relational database‐based system, and on average, our system reduced the execution time by 72% (for 3‐node query), 89% (for 4‐node query), and 99% (for 5‐node query).