Classifying Emails into Human vs Machine Category

It is an essential product requirement of Yahoo Mail to distinguish between personal and machine-generated emails. The old production classifier in was based on a simple logistic regression model. That model trained by aggregating features at the SMTP address level. We propose building deep learning models message train four individual CNN models: (1) content with subject as input, (2) sender email name (3) action analyzing recipients’ patterns generating target labels senders’ opening/deleting behaviors (4) salutation utilizing "explicit salutation" signal positive labels. Next, we final full after exploring different combinations above models. Experimental results editorial data show that our improves adjusted-recall from 70.5% 78.8% precision 94.7% 96.0% compared Also, significantly outperforms state-of-the-art BERT this task. Our new has been deployed current system (Yahoo 6).