Hi Nils, thank you for such an amazing video. The presented research seems to tackle some problems I encountered myself. So, I would like to ask you, what would you recommend in the following scenario: - few text sentence pairs ~ 1k-10k samples - very technical language (e.g. Fire door A20-3 PS - 1,65 x 2,36 m (RBL) aluminium tubular frame construction with glazing) My idea is to use a pre-trained BERT model and do MLM/NSP to better adapt to the very technical domain language using unlabelled data, then do fine-tuning on the labelled text pairs using a MultipleNegativeRankingLoss. The goal is to build a text matching model, which finds the best matching document to a query. Would be amazing if you answer. Cheers!
@honoratap5616
Жыл бұрын
Hey, which approach did you end up going for?
@paulntalo1425
2 жыл бұрын
Thank for sharing such enriching insights
@abhinav6343
2 жыл бұрын
Any mail id would like to connect further for project collaboration. Any mail id?
Пікірлер: 4