How to Fine Tune BERT On unlabeled data?

Hello and Greetings ,
I have a dataset that contains job description , and I have different one that contains job skills , actually I’d like to fine tune BERT so it could better understand the the two columns.(Note the 2 columns job description and job skills are from different source)
So finally I could use cosine similarity to classify them .
any suggestions please ?

Hi Rhayem_Bannouri,

Maybe this post answers your question?