基于点关联测度矩阵分解的中英跨语言词嵌入.PDFVIP

文章编号：1003-0077 （2011）00-0000-00 基于点关联测度矩阵分解的中英跨语言词嵌入 1,2 2 2 1,2 于东 ，赵艳 ，韦林煊 ，荀恩东 （1. 北京语言大学 大数据与教育技术研究所，北京 100083； 2.北京语言大学 信息科学学院，北京 100083） 摘要：研究基于矩阵分解的词嵌入方法，提出统一的描述模型，并应用于中英跨语言词嵌入问题。以双语 对齐语料为知识源，提出跨语言关联词计算方法和两种点关联测度的计算方法：跨语言共现计数和跨语言 点互信息。分别设计目标函数学习中英跨语言词嵌入。从目标函数、语料数据、向量维数等角度进行实验， 结果表明：在中英跨语言文档分类中以前者作为点关联测度最高得到87.04%的准确率；在中英跨语言词义 相似度计算中，后者作为点关联测度得到更好的性能，同时在英-英词义相似度计算中的性能略高于主流的 英语词嵌入。 关键词：点关联测度；词嵌入；跨语言；矩阵分解 中图分类号：TP391 文献标识码：A Chinese-English Cross-lingual Word Embeddings Based on Pointwise Relevant Measurement Matrix Factorization 1,2 2 2 1,2 YU Dong , ZHAO Yan , WEI Linxuan , XUN Endong (1. Institute of Big Data and Language Education, Beijing Language and Culture University, Beijing 10083, China; 2.College of Information Science, Beijing Language and Culture Univers ity, Beijing 10083, China) Abstract: This paper presents a unified model for matrix factorization based word embedding, and applies the model to Chinese-English Cross-lingual word embedding. It proposes a method to determine Cross-lingual relevant word on parallel corpus. Both Cross-lingual word co-occurrence and pointwise mutual information are served as pointwise relevant measurements to design objective function for learning Cross-lingual word embeddings. Experiments are carried out from perspectives of different objective function, corpus, and vector dimension. For the problem of Cross-lingual document classification, the best performance model achieves 87.03% in accuracy, as it adopts Cross-lingual word co-occurrence as relevant measurement. However, models adopt Cross-lingual pointwise mutual information get better performance in Cross-lingual word sim