How to use pretrained BERT in News Recommend

Hello everyone.
I want to use the BERT pre-training model for the news recommendation problem. In this problem, a news item has a lot of information such as title, summary, and content.
Should I use a single BERT model for all 3 of the above fields of the news, or declare 3 separate BERT models (of course the same) for each of the above fields.
Thank you.