Yeah I'm very much betting on the null hypothesis when it comes to fancy machine learning-based methods and implementing the most bog-standard keyword search you could imagine.

It's not perfect, but it's scary how well it works sometimes.