[Question]:

[keshavgarg139]

Question

I am working on a Sequence Labelling problem using the FLAIR module.

I have dummy e-commerce data with 3 different types of entities and each entity has approx ~1K sub-entities. Training Data (size ~200K) is synthetically created with a combination of 3K labels.

I tried to validate the FLAIR Sequence Labelling with a Query Classification model (with 3K labels). The FLAIR model (F1-score: 60%) seriously underperforms than Classification model (F1-score: 80%).

I am reluctant to develop a Sequence Labelling module because I expect the Sequence Labeller to detect and propose new entities as well.

Can you help me understand where I could go wrong and what other models I could try?