Leveraging machine learning and Crowdsourcing for knowledge graph construction in term of learning with privilege information

Frequent Google users may be familiar with the knowledge graph – it is what creates the visual display of key information related to a search query, often shown at the top of the page. The knowledge graph a store of information built up by organizing extracted information from various websites and grouping related material under their respective topics. However, these knowledge graphs are not infallible; because the graphs are built computationally and thus lack the robustness of human interpretation, if the text is written in an unusual manner, it is entirely possible to have incorrect information in the graph. This incorrect information is referred to as noise. The purpose of this project is to create a computer algorithm to better parse and interpret text, to lower the amount of noise entered into knowledge graphs. This algorithm will also be trained on human-computed (thus noiseless) knowledge graphs to increase its effectiveness.

Faculty Supervisor:

Anne Condon

Student:

Partner:

Zhejiang University

Discipline:

Computer science

Sector:

Education

University:

The University of British Columbia

Program: