About the indegree

[yhjiujiu]

Hi, thanks your excellent work. There is a question about the indegree. Specially, the in-degree of a node is the number of times it is a tail entity? And how to calculate the MRR value with degree range in [0,100], does this MRR value mean the degree range of the predicted entity is in [0,100]?

[TimDettmers]

Hi! Thank you, for your question! That is right, the indegree is how many outwards connections point to a node, or in other words, how many tail entities are related to the head entity. One example in the paper for (head, relationship, tail) is (United States, born in, X). United states has a high indegree for born in.

To calculate the MRR value for a range, I would do what you describe, calculate the MRR only on nodes that have an in-degree [0, 100]. You could also perform a bit different experimental setup where you remove all nodes from the dataset that have an indegree of more than 100. Then you can train models on that, and from this data you can compare how models do if a graph has a certain indegree. In this case, you calculate MRR over the entire test set.