基于关联文本的知识图谱表示学习研究

Research on Knowledge Graph Representation Learning Based on Associated Text

近年来的研究表示,知识图谱嵌入对于学习多关系数据的表示是有效的。但是,大多数方法都局限于知识图谱中的结构化数据,这阻碍了实体语义信息的全面表达。因此,要优化嵌入,重要的是考虑更广泛的信息来源,例如文本、图像等。通过获取每个实体的相关文本文档,根据关系生成文本表示,与此同时生成结构表示,将两种表示联合学习。通过链接预测任务评估模型效果,与单独的平移距离模型相比,这种方法有更好的性能。

Recent studies show that knowledge graph embedding is effective for learning multi relational data representation.However,most methods are limited to structured data in knowledge graph,which hinders the full expression of entity semantic information.Therefore,to optimize the embedding,it is important to consider a wider range of information sources,such as text,images and so on.By obtaining the relevant text documents of each entity,the text representation is generated according to the relationship,and the structure representation is generated at the same time,and the two representations are learned jointly.Compared with the single translation distance model,this method has better performance.