The paper contains a detailed overview of the proposed solution, but I have questions on some details.
-
Entity Linking
a. The paper says it uses TIARA's entity linker - then how are questions with literals handled? Because AFAIK TIARA's entity linker does not give literals as output.
b. How are candidates created when mentioned entity is literal? What is the first step of the KG query?
c. How are candidates created when there are more than one mentioned entities in a question?
d. How are candidates created when there are no mentioned entities in a question?
-
Negative Sampling
a. What are the number of negative samples used during training?
b. How is negative sampling performed ?
The paper contains a detailed overview of the proposed solution, but I have questions on some details.
Entity Linking
a. The paper says it uses TIARA's entity linker - then how are questions with literals handled? Because AFAIK TIARA's entity linker does not give literals as output.
b. How are candidates created when mentioned entity is literal? What is the first step of the KG query?
c. How are candidates created when there are more than one mentioned entities in a question?
d. How are candidates created when there are no mentioned entities in a question?
Negative Sampling
a. What are the number of negative samples used during training?
b. How is negative sampling performed ?