Research On Cross-lingual Taxonomy Alignment Algorithm

Cross-lingual taxonomy alignment is a task of discovering the most relevant category in the target taxonomy of one language for each category in the source taxonomy of another language.It is an important way of realizing cross-lingual knowledge sharing and fusion,as well as one of the main applications regarding Semantic Web and natural language processing technologies.For this task,researchers have proposed a variety of methods.However,since these methods rely on domain-specific features,neglect structural information of taxonomies,and introduce lots of noise by translation tools,there still exist some limitations in cross-lingual taxonomy alignment.Therefore,how to choose an effective way to solve these problems is worth to study.Based on the problems mentioned above,this thesis proposes two algorithms leveraging topic models to align cross-lingual taxonomy.Specifically,the main contributions of this thesis are as follows:(1)This thesis proposes an algorithm using bilingual biterm topic model based on the hierarchies of taxonomies.This algorithm solves the problems of relying on domain-specific features and neglecting structural information of taxonomies.(2)This thesis proposes an algorithm leveraging a monolingual biterm topic model based on the hierarchies of the taxonomies and the canonical correlation analysis(CCA).CCA is used for mapping the different topic vector space into a single one,which solves the problem of noise introduced by the translation tools.(3)These two cross-lingual taxonomy alignment algorithms proposed in this thesis are implemented and experimented on two real world datasets.The experimental results show that these two cross-lingual taxonomy alignment algorithms significantly improve the performance.(4)This thesis designs and implements a cross-lingual taxonomy alignment visualization system.