电子学报 ›› 2019, Vol. 47 ›› Issue (9): 1841-1847.DOI: 10.3969/j.issn.0372-2112.2019.09.004

• 学术论文 • 上一篇    下一篇

一种基于实体描述和知识向量相似度的跨语言实体对齐模型

康世泽, 吉立新, 刘树新, 丁悦航   

  1. 战略支援部队信息工程大学, 河南郑州 450002
  • 收稿日期:2018-10-11 修回日期:2019-04-09 出版日期:2019-09-25
    • 通讯作者:
    • 康世泽
    • 作者简介:
    • 吉立新 男,1970生,江苏淮安人.战略支援部队信息工程大学研究员、博士生导师,研究方向为电信网信息关防;刘树新 男,1987生,山东潍坊人.战略支援部队信息工程大学助理研究员,研究方向为复杂网络、移动通信网安全;丁悦航 女,1995年生,山东济南人.战略支援部队信息工程大学硕士研究生,研究方向为知识图谱.
    • 基金资助:
    • 国家自然科学基金 (No.61521003,No.61601513)

Cross-Lingual Entity Alignment Model Based on the Similarities of Entity Descriptions and Knowledge Embeddings

KANG Shi-ze, JI Li-xin, LIU Shu-xin, DING Yue-hang   

  1. Information Engineering University, Zhengzhou, Henan 450002, China
  • Received:2018-10-11 Revised:2019-04-09 Online:2019-09-25 Published:2019-09-25
    • Corresponding author:
    • KANG Shi-ze
    • Supported by:
    • National Natural Science Foundation of China (No.61521003, No.61601513)

摘要: 跨语言实体对齐旨在找到不同语言知识图谱中指向现实世界同一事物的实体.传统的跨语言实体对齐方法通常仅依靠知识图谱内部的结构信息,但实际上一些知识图谱提供的实体描述信息也可以被利用.本文提出了一种结合知识图谱的内部结构和实体描述信息共同进行跨语言实体对齐的模型.该模型首先通过训练基于知识图谱结构信息的知识向量找到可能被对齐的实体对,再结合实体描述信息利用改进后的共享参数模型选出最终的对齐实体,最后通过迭代对齐的方法重复前两个步骤找到更多的对齐实体直到训练结束.实验结果表明,与基准算法相比,本文所提模型在跨语言实体对齐任务上可以取得相对不错的结果.

关键词: 跨语言实体对齐, 知识向量, 跨语言实体描述相似度

Abstract: Cross-lingual entity alignment aims to find entities in knowledge graphs of different languages that point to the same objects in the real world. Traditional cross-lingual entity alignment methods usually rely solely on the internal structure information of the knowledge graph, but in fact entity description information provided by some knowledge graphs can also be utilized. This paper proposes an entity alignment model that combines the internal structure information of the knowledge graph with the entity description information for cross-lingual entity alignment. The model first finds the entity pairs that may be aligned by training the knowledge embeddings based on the structure information of the knowledge graph,and then uses entity descriptions to select the final aligned entity pairs based on the improved optimal alignment similarity model. Finally, the model iteratively align the first two steps to find more aligned entity pairs until the end of the training. The experimental results show that compared with the benchmark algorithms, the proposed model can achieve relatively good results in cross-lingual entity alignment task.

Key words: cross-lingual entity alignment, knowledge embeddings, cross-lingual description similarity

中图分类号: