Cross-Lingual Entity Linking For Web Tables

With rapid advances in Web technologies,there are more and more HTML tables on the web,which contains rich relational semantic information.One common way to better process and understand those tables for machines is to link string mentions in each table cell to corresponding entities in a knowledge base,such as Wikipedia or Freebase.This technique is called “Entity Linking”.Knowledge Base,or Knowledge Graph,is a huge structural database of human knowledge,containing large amount of entities together with different relations between different two entities.Linking string mentions to entities in a knowledge base helps us disambiguate mention,and enables machines understand natural language better.This leads to great improvements in natural language process related applications,such as Question Answering.On the other hand,since tables tend to have consistent structures,it is easy to extract structural information or knowledge from Web tables and add it into knowledge base if it is undiscovered.We call it “Table Linking” when apply entity linking on Web tables.In general,table linking is performed in monolingual scenario,which means the the origianl table and the target knowledge base are written in the same language,such as linking an English table to an English knowledge base.However,when it comes to linking web tables in other languages,the corresponding non-English knowledge bases are often not comprehensive enough to cover all the entity mentions in the tables at hand.Therefore,this thesis trys to link non-English web tables to English knowledge base,in a novel process that we call “Cross-Lingual Table Linking”.We attempt to solve the cross-lingual table linking problem without using any non-English knowledge bases.To the best of our knowledge,this is the first attempt that attacks the cross-lingual table linking problem.We thus propose a neural network based joint model for cross-lingual table linking.We embed mention,context and entity in a continuous vector space to capture their semantics.Further,we employ a linear transformation between vector spaces of two languages.For each table,we link all the mentions simultaneously,so as to fully utilize the relationships among entities in the same row or column.We encode these correlations as a coherence feature in the model.Furthermore,we design a pairwise ranking loss function for parameter learning and propose an iterative prediction algorithm to link new tables.Experimental results report that our approach improves the accuracy of cross-lingual table linking by a relative gain of 12.1%.Detailed analysis of our approach also shows a positive and important gain brought by the joint framework and coherence feature.