开放获取生物医学文献的实体关系抽取

海量开放获取生物医学文献中蕴含着大量的知识，对其进行深度挖掘意义重大。然而全文文献中的实体关系挖掘面临着高维稀疏矩阵计算、语义漂移和谓词发现问题。针对这些问题，本项目拟开展如下研究：1.提出基于文献空间结构信息和表示学习模型的深度学习实体关系抽取算法，有效处理长尾分布实体及其关系上面临的严重稀疏问题。2.针对实体关系抽取中的语义漂移问题，分析其形成原因和类别，提出度量方法，融合深度学习与半监督学习模型解决语义漂移问题。3.将生物医学实体和关系生成知识图谱，运用路径排序算法和本体扩展模型从两实体间的多步关系路径预测它们之间的关系，发现新的谓词。4.将新提出的模型和算法整理形成生物医学文献开放实体关系抽取平台，为生物学家和医学家提供服务。

Plenty of knowledge is included in the massive open access biomedical literatures, however, entity extraction on full texts are fall into high-dimensional matrix computing, semantic change and predicate discovery problems. To address these issues, our main researches are as follows: 1. With the structure information of literature and representation learning model, a novel biomedical entity relationship extraction method based on deep learning is going to proposed. It can solve the serious sparse problem of long-tailed entity and relationship distributions. 2. For the semantic change problems, we are going to analysis the cause and classification of them. Moreover, the quantity measurement of semantic change will be designed. Then, a hybrid model of deep learning and semi-supervised learning is going to be presented for semantic change solution. 3. Generate a biomedical knowledge graph based on biomedical entities and relationships. Using path ranking algorithm and ontology extension model, we will predict the relationships between entities and discover the predicates from the muti-step path. 4. Based on the new proposed models and algorithms, an open entity relationship extraction platform will be constructed. These researches will provide the service for biologists and medical doctors.