This paper presents an overview of a quality scoring system that utilizes pre-trained deep neural network models. Two types of DL models, a classification and extractive question answering (EQA) models are used to implement components of the system. The abstracts of the scientific literature are classified into two groups, in-vivo and in-vitro, and a question and answering model architecture is constructed for extracting the following types of information (animal type, the number of animals, exposure dose, and signal frequency). The Bidirectional Encoder Representations of Transformers (BERT) model pre-trained with a large text corpus is used as our baseline model for classification and EQA tasks. The models are fine-tuned with 455 EMF-related research papers. In our experiments, the fine-tuned model showed improved performance on EQA tasks for the four-categories of questions compared to the baseline, and it also showed improvements on similar questions that were not used in training. This suggests the importance of retraining of deep learning model specifically in some areas requiring domain expertise such as scientific papers. However, additional research is needed on some implementation issues, in such cases where there are still multiple answers, or where there is no answer given in a context.

This study compares the performance of graph convolutional neural network (GCN) models with conventional natural language processing (NLP) models for classifying scientific literature related to radio frequency electromagnetic field (RF-EMF). Specifically, the study examines two GCN models: BertGCN and the citation-based GCN. The study concludes that the model achieves consistently good performance when the input text is long enough, based on the attention mechanism of BERT. When the input sequence is short, the composition parameter $\lambda$, which combines output values of the two subnetworks of BertGCN, plays a crucial role in achieving high classification accuracy. As the value of $\lambda$ increases, the classification accuracy also increases. The study also proposes and tests a simplified variant of BertGCN, revealing performance differences among the models under two different data conditions by the existence of keywords. This study has two main contributions: (1) the implementation and testing of a variant of BertGCN and citation-based GCN for document classification tasks related to radio frequency electromagnetic fields publications, and (2) the confirmation of the impact of model conditions, such as the existence of keywords and input sequence length, in the original BertGCN. Although this study focused on a specific domain, our approaches have broader implications that extend beyond scientific publications to general text classification.

This study introduces a multimodal approach designed to detect anomaly in satellite imagery. The approach is comprised of three main components: the image processing models (ResNet and Regulated Network), the text processing models (BERT and a graph model), and a combination model that acts as a classifier, seamlessly integrating features extracted from both image and text models. Our work utilizes the SpaceNet8 Challenge dataset, focusing on regions in Louisiana USA, damaged by Hurricane Ida in 2021, and the Ahrweiler district in Germany, impacted by flooding in Western Europe during the same year. The experimental results demonstrate significant performance improvements when integrating image processing models and text processing models, surpassing the baseline CNN models. This study offers insights for future work in anomaly detection and semantic segmentation, emphasizing the effectiveness of a multimodal approach with heterogeneous data.

This paper ¹ presents an overview of a data augmented classification and multi-span (multiple) answer system for extracting key information from academic publications. This study consists of two sections: (a) implementing a new fine-tuned model to solve the multiple answer extraction issue (b) reporting results of sub-classification in various RF-EMF topics. In our previous study, it has been found that the essential cause of the low performance of the extractive question answering (EQA) system for certain types of questions was the multiple answer issue. To solve this problem, this study applies the TASE (TAg-based Span Extraction) technique and introduces the results. Our approach can retrieve multiple answers spreading over a given text by referring to the pre-trained TASE model with fine accuracy. In addition, this work adopts 'PEO (Population, Exposure, Outcome)' from the 'PECO' of the WHO-funded study on RF-EMF safety, as our holistic research framework. Based on the PEO perspective, the results of three sub-topic (RF, SAR, Causal Relationship) classifications are presented. For both models of multi-span answer and classification tasks, the data-augmenting method plays an important role. In particular, it is found that our proposed system outperforms the pre-trained BERT model in multi-span answer tasks with our RF-EMF dataset.