Drug-Drug Interaction Extraction from Biomedical Text using Relation BioBERT with BLSTM

Abstract

Drug-drug interactions (DDIs) happen when two or more drugs interact. DDIs may change the effect of drugs in the body which can induce adverse effects and severe diseases for patients. As a result, detecting the interaction between drugs is essential. In the last years, many new DDIs have been found and added to the medical datasets with the rise in the number of discovered drugs. On the other hand, since a lot of this information is still in Biomedical articles and sources, there is a need for a method to extract DDIs information. However, despite the development of many techniques, attaining good prediction accuracy is the main issue. This paper proposes a deep learning approach that: 1) uses the power of Relation BioBERT (R-BioBERT) to detect and classify the DDIs and 2) employs the Bidirectional Long-Short Term Memory (BLSTM) to further increase the prediction quality. Not only does this paper study whether the two drugs have an interaction or not, but it also studies specific types of interactions between drugs. The paper also provides that using BLSTM can significantly increase the F-scores compared to the baseline model on the famous SemEval 2013, TAC 2018, and TAC 2019 DDI Extraction datasets.

Research on Automatic Terminological Dictionary Extraction Based on Bilingual Parallel Corpus

“…At the end of the network, there is an information fusion layer, which is reflected in the mathematical level as vectors and merges, so that the vectors fused by the information fusion layer contain the past and generalized information. Compared with multi-layer LSTM neural network, it has the advantages of comprehensive information, strong robustness, and the ability to take into account both front and back data in natural language processing, which enables the trained machine to learn more abstract samples [7].…”