面向链接预测的知识图谱嵌入研究综述

A Survey of Knowledge Graph Embedding for Link Prediction

摘要:

知识图谱在人工智能领域有着广泛的应用,如信息检索、自然语言处理、推荐系统等。然而,知识图谱的开放性往往意味着它们是不完备的,具有自身的缺陷。鉴于此,需建立更完整的知识图谱,以提高知识图谱的实际利用率。利用链接预测通过已有关系来推测新的关系,从而实现大规模知识库的补全。通过比较基于翻译模型的知识图谱链接预测模型,从常用数据集与评价指标、翻译模型、采样方法等方面分析知识图谱链接预测模型的框架,并对基于知识图谱的链接预测模型进行了综述。

Abstract:

Knowledge graphs have a wide range of applications in the field of artificial intelligence, such as information retrieval, natural language processing, recommender systems, etc. However, the openness of knowledge graphs often means that they are incomplete and have their own flaws. In view of this, it is necessary to establish a more complete knowledge graph to improve the actual utilization of the knowledge graph. Using link prediction to infer new relations through existing relations, so as to realize the completion of large-scale knowledge base. By comparing the knowledge graph link prediction models based on translation models, the framework of the knowledge graph link prediction model is analyzed from the aspects of commonly used datasets and evaluation indicators, translation models, and sampling methods, and the link prediction models based on knowledge graphs are reviewed.