eBay NER Machine Learning Challenge

NER labeling for handbag dataset

Baseline F1 Score to beat: 0.800

---

Model 2.3

• Method
  • 2 Bidirectional LSTMs for main and Time Distributed LSTM to capture character positioning
• Inputs
  • Word Embeddings, POS embeddings, Character encoding from LSTM
• Tags
  • Retagged with BIOS tags for multi-token
• Data
  • Training set + random expansion

Model Params: 12,852,467

epochs: 10

Result - F1 Score: 0.82313

---

Model 2.2

• Method
  • 2 Bidirectional LSTMs
• Inputs
  • Word Embeddings and POS embeddings
• Tags
  • Retagged with BIOS tags for multi-token
• Data
  • Training set + random expansion

Model Params: 12,708,947

epochs: 12

Result - F1 Score: 0.8149