Entity Linking is the task of automatically identifying
entity mentions in a piece of text and linking them to their
corresponding entries in a reference knowledge base like Wikipedia.
Although there is a plethora of works on entity linking, existing
state-of-the-art approaches do not explicitly consider the time aspect
and specifically the temporality of an entity’s prior probability
(popularity) and embedding (semantic network). Consequently, they
show limited performance in annotating old documents like news or
web archives, while the problem is bigger in cases of short texts with
limited context, such as archives of social media posts and query logs.
This thesis focuses on this problem and proposes a modeling that
leverages time-aware prior probabilities and word embeddings in the
entity linking task