Research On Biological Relationship Extraction Based On Multi-granularity Semantic Fusio

With the rapid development of information technology,a large amount of unstructured text data is generated,which hides a lot of valuable information.Information extraction aims to help people automatically extract required information from massive data,so this technology can be applied to fields with large amounts of data such as economics,news,and biomedicine.In recent years,the published biological literature has increased exponentially,and the research and application of information extraction in the biological field has become more and more urgent.However,there are obvious differences in the characteristics of texts in different fields.For example,there are a large number of proper nouns such as proteins and drugs in the biological literature.Accurately extracting the interaction between these biological entities is of great significance for the study of cell metabolism and pathological processes.In relation extraction for the biological domain,a parser is usually used to obtain shallow semantics such as part of speech and grammar.However,there are usually many duplicate entity pairs in biological literature,and they have different relationships in different semantics,and it is difficult to integrate shallow semantics These same entities are distinguished.Therefore,this paper uses deep learning technology to obtain deeper semantic relationships,and studies a biological relationship extraction model based on sentence semantics.The main work is as follows:(1)Aiming at the problems that a sentence in biological literature contains multiple duplicate biological entities,and these entities are easy to confuse the extracted relation instances and generate redundant information,this paper proposes a biological relation extraction model based on the local semantics of sentences.The model divides the sentence structure through a multi-channel network structure,effectively distinguishing relation instances from non-target entities.Each channel updates the parameters independently,so that the same word in the sentence can learn different representations,focusing on the semantics of the sentence at a finer granularity,thereby reducing the interference of redundant information and enhancing the discriminability of entities.It shows good generalization ability.(2)Aiming at the problems that the same entity pairs show different interaction relationships in different sentences,and the densely distributed biological entities result in limited information available between entity pairs,this paper proposes a biological entity based on global and local semantic fusion.Relation extraction model.The model firstly encodes the global semantics of sentences based on Bio BERT,and uses convolutional neural networks to extract global semantic features to effectively distinguish the relationship between the same entity pair in different sentences.At the same time,a biological relation extraction model based on the local semantics of sentences is used to obtain the local semantic features of sentences to supplement the global semantic features.By fusing global and local semantic features,the semantic information learned by the model is more comprehensive,which greatly improves the extraction performance.The semantic-based biological relationship extraction model proposed in this paper has excellent performance in entity relationship extraction.The performance of the relationship extraction model based on global and local semantic fusion has reached the current advanced level based on deep learning,and has performed generalization ability on biological standard datasets such as AImed.