In the past decade, information extraction played an essential role in bioinformatics, medical analysis, and the healthcare system, and it’s still in the budding stage for nanomaterials. The crucial step in the information extraction process is extracting and analyzing unstructured text documents. The use of transformer models for this task has been rapidly growing in the field of nanomaterials. To facilitate our work, this paper proposes the fundamental task of relation extraction manually and outlines the step-by-step characteristics of a pre-trained Bert Model. BERT (Bi-Directional Encoder Representation from Transformers) symbolized the use of pre-train to comprehend the language and fine-tune it to learn a particular task. The training phase is usually composed of the next sentence prediction model and the MLM (Masked Language Model). The various relation extraction BERT algorithms that have been applied in recent articles are examined in this paper. This article analyzed the data for nanomaterial existing documents, for information extraction, by using BERT models like SciBert, MatSciBert, and BioBert. This paper identifies the relations present in the sentence and labels the entities, using the NER (named entity recognition) model. Finally, use the most recent and cutting-edge deep learning techniques as baseline models and carry out extensive experiments using MatSciBert, which yielded a good results accuracy of 72.79%. In essence, this paper provides detailed information on successfully customizing deep learning for healthcare applications. According to the comprehensive findings from the models, the article has shown that relation extraction in Material Bert gives better accuracy compared to other BERT models.

Nanoinformatics, MLM, Relation Extraction, BERT;