Core Concepts
TAR 1.0
Definition
A one-shot TAR workflow: train a model on a fixed sample, apply it to the full set, then validate the result.
Why it matters
It produces a clean statistical story but assumes the collection is stable.
In practice
A TAR 1.0 workflow has distinct phases. A subject-matter expert codes training documents in rounds until the model stops improving. The model then classifies the whole collection, and a cutoff score splits it into documents to review or produce and documents to set aside.
Progress is usually measured against a control set: a random sample coded at the start and never used for training. When control-set metrics stop improving between rounds, training is treated as stable.
The weak point is change. If new documents arrive after training, the model and control set may no longer represent the collection, and parts of the process may need to be repeated.
Compare with
- TAR 2.0: TAR 2.0 trains and reviews at the same time and has no separate training phase.