Sequencing data of 5' race TCR β chain

Hi all. Recently, I constructed two libraries using 5’ race protocol and sequenced on novaseq. I got CDR3 sequences, but there was about 20% reads contain long continuous T bases or A base (more than 50 nt ). Where are they come from? How should I improve my workflow to avoid it again? Please give me an advice.

Thank you so much.