Commonsense Question Answering

PhD Qualifying Examination


Title: "Commonsense Question Answering"

by

Mr. Zizheng LIN


Abstract:

Understanding commonsense is an indispensable requirement of achieving 
artificial general intelligence. Therefore, various CommonSense Question 
Answering (CSQA) tasks and benchmarks have been proposed to examine the 
commonsense comprehension and reasoning ability of AI models. These tasks 
and benchmarks either put emphases on a specific aspect on commonsense 
knowledge (e.g., social commonsense, physical commonsense, temporal 
commonsense) or covers general commonsense knowledge. To mitigate the 
reporting bias issue of commonsense knowledge, many CommonSense Knowledge 
Graphs (CSKG) have been constructed to provide models with abundant 
explicit and structured source of commonsense knowledge, which 
substantially boosts the progress of CSQA research. Recently, many 
algorithms have been proposed to solve CSQA tasks. Based on the source of 
commonsense knowledge, they can be divided into the following three 
categories: (1) Using a Pre-Trained Language Model (PTLM) as the only 
implicit knowledge source; (2) Enhance the QA model with an external 
knowledge graph as an explicit knowledge source; (3) Using the explicit 
knowledge induced from PTLM. Each category of methods has its own 
advantages and disadvantages in different perspectives such as 
interpretability and performance. In this survey, we first introduce tasks 
and resources for CSQA, specifically the tasks and benchmarks, as well as 
some prominent CSKGs. Then we describe recent methods for CSQA, classified 
as using PTLM as the only implicit knowledge source, using external 
knowledge graph as explicit knowledge source, and inducing explicit 
knowledge from PTLM. Afterward, we present the experimental results 
recording the CSQA performance of these methods. Lastly, we conclude the 
survey and point out some possible future directions for CSQA research.


Date:			Friday, 9 July 2021

Time:                  	2:00pm - 4:00pm

Zoom meeting:
https://hkust.zoom.us/j/98337880915?pwd=cWxyNGhYaCtUS3ZlUENBSnIvL1VGQT09

Committee Members:	Dr. Yangqiu Song (Supervisor)
 			Prof. Xiaofang Zhou (Chairperson)
 			Dr. Qifeng Chen
 			Prof. Nevin Zhang


**** ALL are Welcome ****