Relation Extraction (RE) is to predict the relation type of two entities that are mentioned in a piece of text, e.g., a sentence or a dialogue.
Given two entities and a piece of text where the two entities are mentioned in, the task of relation extraction (RE) is to predict the semantic relation between the two entities. The piece of text serves as the context for the prediction, which can be a short sentence, a long sentence, or even a dialog.
Example: Shantam likes to eat pizza, and he is a fan of burger too, unlike his friend Vidushi who loves to eat rice.
per: Shantam
per: Vidushi
food: pizza, burger, rice
Entities: words referring to the same object are called entities. For example, Shantam is an entity of type per .
Relation: A relation is defined in the form of a tuple t = (e1,e2,...,en) where the e_i are entities in a predefined relation r within document D.
relation triple —> (subject, relation type, object)
In a nutshell, the RE task I am focussing on is in supervised setting, where we try to train a stacked binary classifier (for multi-label) and classify relations for the entities extracted. Assumption is we already have the entities extracted/ or have a pipeline to have that available.