I'm new to relation extraction and I'm currently reading related papers.
The term "entity surface form" is repeatedly mentioned in most of them. What does it mean?
For example, in the REBEL paper, the author mentions that "a relation is considered correct only if the head and tail entity surface forms are correctly extracted."
What is meant by "head" and "tail"?
Thanks.
Surface form refers to raw text that represents the entity. While the head and tail represent the entities of some relation.
e.g. sentence: "Doom II, also known as Doom II: Hell on Earth, is a first-person shooter game"
Let's say that the bolded parts here are two entities, one of those entities could be seen as a subject of the sentence (Doom II) and one as an object (first-person shooter game), subjects and objects are sometimes called heads and tails in the literature (as the relations fit this form (head, relation, tail)).
To address the surface form more, let's say your model predicted (extracted) that the head entity is "Doom", or perhaps "Doom 2", if we compare it to original "Doom II", by the REBEL standards, that wouldn't be a correct extraction.
As someone who is also diving into the field of RE, I suggest you visit the Tsinghua university NLP lab github: https://github.com/thunlp/OpenNRE