We measure the precision against other algorithms in the two phases in the approach: 

  1. Intent induction. The precision of the categories extracted is almost 100% as the semantic contexts are mathematically present in the corpora. For our users approximately 70-90% prove useful for their goals. Regarding recall, we are usually between 50 and 80% of the data with at least one intent.

  2. Real-time classification. Precision in real time is around 85% and recall is around 80%.

Did this answer your question?