We apply transfer learning from the sentence-transformer model in Hugging Face called 'sentence-t5-base' due to its optimal accuracy and complexity, as observed from the model leaderboard in the sentence similarity task. We then utilize this model as a cross-encoder model.
The benefits of transfer learning can enhance the accuracy of the model in understanding job titles, particularly those related to the medical field such as 'doctor'.
Before training
After training
- STS train dataset. Link to our dataset
- HuggingFace STS Model Ranking. Link to the similarity task model leaderboard.
- Difference of Bi-Encoder & Cross-Encoder. Detailed explaination of bi-encoder and cross-encoder
- STS-trained-lokergo Model. Our trained model was hosted into HuggingFace repository for easy access
C23-VR01 ML Teams.