Home > Events > Oral Candidacy - Dash Shi

Oral Candidacy - Dash Shi

Start: 4/7/2017 at 3:30PM
End: 4/7/2017 at 6:00PM
Location: 315 Stinson Remick
Attendees: Faculty and students are welcome to attend the presentation portion of the defense
Add to calendar:
iCal vCal

Baoxu (Dash) Shi

Oral Candidacy
          April 7, 2017          3:30 pm        315 Stinson Remick
Adviser:  Dr. Tim Weninger
Dr. David Chiang        Dr. Jane Cleland-Huang        Dr.  Xifeng Yan


Knowledge Graph Completion on Open and Closed Knowledge Graphs

Although numerous new information is created every day, most of them are unstructured and need to be processed into structured data for downstream tasks. Knowledge Graphs (KGs) is one of the most important structured data format that empowers a variety of tasks including fact checking, question answering, entity linking, link prediction, and many others. However, KGs harvested from raw data are usually far from complete and may contain errors, which limiting the performance of KG-dependent models. Therefore, it is important and necessary to develop Knowledge Graph Completion (KGC) methods to improve the reliability of existing KGs and benefit downstream tasks. Many KGC models have been introduced recently, but there are still two main problems that we are facing: 1) lacking large-scale standard benchmark datasets with rich features; 2) most existing models have a closed-world assumption and are unable to inference on unseen entities and relationships. To address the above problems, this proposal will be focusing on 1) introducing new large-scale Knowledge Graph evaluation dataset constructed using DBPedia and Wikipedia, 2) designing a new close-world KGC model that improves the overall performance with current close world KGC settings, 3) proposing open-world KGC models that can work with new, unseen entities or relationships.