Architectural Components

Trace entity linking through two paths: training the recognizer and disambiguator, then using them on new text. Candidate generation sits inside disambiguation and reduces the set of knowledge-base entries the linking model must compare.

The architectural components diagram for entity linking is shown below. It consists of two paths:

1. Model generation path (training flow)

2. Model execution path (prediction flow)

Terminology: This lesson uses recognized entity for a mention detected in text. Keep that mention separate from the knowledge-base entity to which it may later be linked.

Model generation path

Model generation is responsible for training models for entity linking task. Let’s look at the components of this path.

Training data generation