JMIR Publications

ABSTRACT

Background:

The most current methods applied for the intra-sentence relation extraction in biomedical texts are inadequate for the document-level relation extraction, in which the relationship may cross sentence boundary. Hence, some approaches have been proposed to extract relations by splitting the document-level datasets through heuristic rules and learning methods. However, these approaches may introduce additional noise and do not really solve the problem of inter-sentence relation extraction. It is challenging to avoid noise and extract cross-sentence relation.

Objective:

This study is aimed to avoid errors by dividing the document-level dataset, to verify that self-attention structure can extract biomedical relation in a document with long distance dependencies and complex semantic, and to discuss which entity pretreatment method is better for biomedical relation extraction.

Methods:

The paper proposes a new data preprocessing and attempt to apply the pretrained self-attention structure for the document biomedical relation extraction with an entity replacement method to capture very long-distance dependencies and complex semantics.

Results:

Compared with the state-of-art approaches, our method greatly improved the value of precision. The result shows that our approach increases F1 value compared with the state-of-art methods. Through experiments of biomedical entity pretreatments, we find that model using an entity replacement method can improve performance.

Conclusions:

When considering all target entity pair as a whole in document-level dataset, pretrained self-attention structure is suitable to capture very long-distance dependencies and learn textual context and complicate semantics. Replacement method for biomedical entities is conductive to biomedical relation extraction, especially to document-level relation extraction.