Anonymization and Information Loss

Ke Wu, Baozhong Yang, Zhenkun Ying +1

Anonymizing financial texts prevents large language models (LLMs) from exploiting look-ahead bias, but inadvertently weakens the extracted signal. We propose a framework disentangl…