We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Classifying Emails into Human vs Machine Category

Abstract: It is an essential product requirement of Yahoo Mail to distinguish between personal and machine-generated emails. The old production classifier in Yahoo Mail was based on a simple logistic regression model. That model was trained by aggregating features at the SMTP address level. We propose building deep learning models at the message level. We built and trained four individual CNN models: (1) a content model with subject and content as input; (2) a sender model with sender email address and name as input; (3) an action model by analyzing email recipients' action patterns and correspondingly generating target labels based on senders' opening/deleting behaviors; (4) a salutation model by utilizing senders' "explicit salutation" signal as positive labels. Next, we built a final full model after exploring different combinations of the above four models. Experimental results on editorial data show that our full model improves the adjusted-recall from 70.5% to 78.8% compared to the old production model, while at the same time lifts the precision from 94.7% to 96.0%. Our full model also significantly beats the state-of-the-art Bert model at this task. This full model has been deployed into the current production system (Yahoo Mail 6).
Comments: This paper is accepted by AAAI'22
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Cite as: arXiv:2112.07742 [cs.CL]
  (or arXiv:2112.07742v1 [cs.CL] for this version)

Submission history

From: Hongwei Shang [view email]
[v1] Tue, 14 Dec 2021 20:55:35 GMT (212kb,D)

Link back to: arXiv, form interface, contact.