Taking a Closed-Book Examination: Decoupling KB-Based Inference by Virtual Hypothesis for Answering Real-World Questions

Algorithm 1

Mining formulas by random walk.

Input: KB, hypothesis H (X, Y)
Output: F, W
(1)	Collect all instances H (x, y)
(2)	For H (x, y) H (X, Y)
(3)	Repeat Random walk from x
(4)	If terminal t = y
(5)	Conceptualize path x to y, as B (X, Y),
Put B (X, Y)  H (X, Y) into F
(6)	Prune fF by supports or heuristic rules
(7)	Learning weights W for F
(8)	Output F and W