*Mentor: **Nilesh Chakraborty* *Student: Peng Xu* In the first phase of the project, I complete a approach to find new mappings for languages with few existing mappings based on languages with enough mappings and cross-language links. This approach can achieve fairly good evaluation results. And I get 456 high-quality new mappings for Chinese after my manual check.
Due to the incomplete coverage of mapping on DBpedia, we need to predict ontology types for instances without a type. In the second phase, I try different methods including tensor factorization and graph embeddings on DBpedia to do type prediction. The experiments show that tensor factorization can achieve good performance on small languages like Bulgarian. However, for larger languages it performs badly due to the limit of memory. All the scripts I wrote can be easily applied to other languages if datasets are downloaded to the proper path. All my code and detailed documents can be found here: https://github.com/dbpedia/mappings-autogeneration. Further work: Currently, I ignore the literals in DBpedia when doing tensor factorization. The next step is to add the literal information. Furthermore, considering the issues about time and memory complexity, a distributed implementation of the algorithm can be useful if we want to apply the ideas on languages like English. Best Regards Peng Xu ---------------------------------- http://billy-inn.github.io/ M.Sc., Department of Computing Science University of Alberta
------------------------------------------------------------------------------
_______________________________________________ DBpedia-discussion mailing list DBpedia-discussion@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion