1

I've been reading the paper Reinforcement Knowledge Graph Reasoning for Explainable Recommendation (by Yikun Xian et al.) lately, and I don't understand a particular section:

Specifically, the scoring function $f((r,e)|u)$ maps any edge $(r,e)$ to a real-valued score conditioned on user $u$. Then, the user-conditional pruned action space of state $s_t$ denoted by $A_t(u)$ is defined as:

$A_t(u) = \{(r,e)| rank(f((r,e)|u))) \leq \alpha ,(r,e) \in A_t\} $

where $\alpha$ is a predefined integer that upper bounds the size of the action space.

Details about the scoring function can be found in the attached paper.

What I don't understand is: What does rank mean, here? Is the thing inside of it a matrix?

It would be great if someone could explain the expression for the user conditional pruned action space in greater detail.

nbro
  • 39,006
  • 12
  • 98
  • 176
stoic-santiago
  • 1,121
  • 5
  • 18

1 Answers1

1

$\text{rank}(f((r,e)|u))$ in $A_t(u)$ means to compute the value of scoring function $f$ for all pairs $(r,e)\in A_t$ which are conditioned by $u$, then sort them in a descending order. The rank of the $f((r,e)|u)$ in this order is equal to $\text{rank}(f((r,e)|u))$. Hence $\text{rank}(f((r,e)|u)) \leqslant \alpha$ means to select the $\alpha$ top most scored pairs.

OmG
  • 1,731
  • 10
  • 19