Linking Mentions to Entities

In the previous two chapters, we have discussed the first two components of the entity linking pipeline: locating mentions and generating candidate entities. The next step in the pipeline is to select an entity from the candidate set of each mention according to the meaning of the input text. In the candidate generation step, we see that most approaches are solely based on name string similarity and entity popularity, namely, other contextual information from the input text is not used. Since the goal of candidate generation is to quickly produce a small set of entities that contains the target entity, sophisticated models might not be needed. However, in order to correctly pick the answers out of the candidate sets, contextual information should be taken into account. This step is often referred as context-sensitive inference.