It seems to me that entity linking and entity matching are very similar.
Entity linking is also named named entity disambiguation; while entity matching is also called record linkage or deduplication, and sometimes entity resolution. From their names, they look almost the same.
And the methods the two employ overlap very much, making it more complicated to differentiate them. Could any one please illustrate their key difference using a simple example?
I'll try to answer but I'm not completely sure myself.
To my knowledge, the terms entity matching, record linkage and deduplication refer to some kind of approximate method meant to capture the different forms of the same entity. Typically these methods rely on approximate/fuzzy string matching techniques (aka string similarity/distance measures). This implies that generally these tasks don't consider the full semantic links, for example they won't try to match something like "Joe Biden" with "The president of the US" because there are no surface clues available. Afaik record linkage and deduplication are more used in the fields of IR and databases, whereas entity matching is closer to NLP terminology (this may be subjective).
I think that entity resolution is quite different, it refers to the NLP task of finding the cases where the same entity is mentioned, most often in a text (i.e. with the context available, as opposed to a database). Normally the task is more subtle, it's supposed to use semantic information (e.g. "Joe Biden" with "The president of the US") and grammatical clues (e.g. "Peter sent a letter to Mary in which he was telling her that John... ").
There's indeed a lot of overlap and it's likely that these terms are occasionally not used this way.
As depicted below, entity linking is linking mentions to entities/identifiers in a reference repository or knowledge base.
Source: Entity linking-Wikipedia
However, in entity matching the existing reference repository or entities in a knowledge base do not exist.
Source: Lecture 9: Entity Resolution