双语影视知识图谱的构建研究

发布时间：2018-03-21 17:49

本文选题：影视本体　切入点：双语　出处：《北京大学学报(自然科学版)》2016年01期 　论文类型：期刊论文

【摘要】：提出一种双语影视知识图谱(BMKG)的构建流程。通过半自动化的方法构建了双语影视本体(BMO),将各个影视数据源对齐到BMO,以保持异构数据源的语义描述一致性。在知识链接方面,在充分挖掘和利用领域特征的基础上,采用基于Word2Vec和TFIDF两种向量模型的实体相似度计算方法,使相似度特征增加一倍,大大提升了模型的链接效果。在实体匹配方面,提出基于相似度传播算法的实体匹配算法,并利用影视数据源之间的内在联系,克服了跨语言实体之间计算相似度的语言障碍。实验结果表明,当阈值取到0.75以上时,实体匹配的准确率都能达到90%左右。此外,还建立了影视知识图谱共享平台,并提供开放性的数据访问和查询接口。
[Abstract]:A bilingual film knowledge mapping (BMKG) of the construction process. Through the method of semi automated construction of bilingual Film Ontology (BMO), the video data source is aligned to the BMO, in order to maintain the semantic heterogeneous data source description consistency. In the knowledge link, based on fully utilizing the field characteristics. The methods of calculating the similarity of Word2Vec and TFIDF two based on vector model, the similarity is doubled, greatly enhance the model link effect. The matching entity similarity propagation algorithm is proposed based on entity matching algorithm, and the intrinsic link between the video source data, to overcome the cross language between entity computing language the obstacle of similarity. The experimental results show that when the threshold is more than 0.75, the accuracy rate of matching entity could reach about 90%. In addition, the film also established a knowledge map sharing platform, It also provides open data access and query interfaces.

【作者单位】：清华大学计算机科学与技术系知识工程实验室;
【基金】：国家重点基础研究发展计划(2014CB340504) 国家自然科学基金委员会与法国国家科研署双边合作协议(61261130588) 清华大学自主科研项目(20131089256) 国家科技支撑计划(2014BAK04B00) THU-NUS下一代搜索联合研究中心项目资助
【分类号】：TP391.1

【参考文献】