Knowledge Graph Completion

Projection Embedding

With the large volume of new information created every day, determining the correctness of information in a knowledge graph and filling in its missing parts becomes a curical task. If researchers can obtain a knowledge graph which has all the information in this universe without any error, then we can increase the accuracy of a lot knowledge graph based models (for example, the Fact Checking work at Weninger Lab).

In this project, we developed a neural network model called ProjE to validate an arbitrary information in RDF triple style (subject,relation,object).

 
ProjE Architecture
 

Knowledge Networks

Baoxu Shi and Tim Weninger Discriminative Predicate Path Mining for Fact Checking in Knowledge Graphs. Knowledge Based Systems, 104(15), 123-133, 2016.

Paper

Baoxu Shi and Tim Weninger Fact Checking in Heterogeneous Information Networks. International Conference on the World Wide Web (WWW), Montreal, Canada. April 2016.

Paper

Code

Fact Checking

Take relation capitalOf as an example. We first process millions of relation information from the knowledge we obtained, and then give the defintion of capitalOf in the context of United States, which represented as a set of discriminative predicate paths. Then for a claim to be checked, we generate the confidence based on how good the given claim following the mined defintion.

Fact Checking Example Fact Checking Architecture