MENU

Applying deep learning to mail flows to identify spammers

This project will explore the application of multilayer artificial neural networks (anns) in predicting the likelihood that a given sender in a multi-tenant email system will be identified as a spammer at some time t.

Malicious email typically originates within compromised machines on web hosting networks. Detecting when a machine has become compromised by a spammer is of considerable concern for network providers, whose reputation is at risk when these actors begin sending out malicious email from the network. Today's state of the art in spammer detection is based on human-generated rules and programming, such as fixed volume limits or statistical analyses. Rules-based mitigation can be effective, but requires painstaking manual maintenance to keep up with the adversary, who can easily and cheaply adapt their approach in order "game" the rules. In the present project, we will investigate the application of modern deep learning AI techniques to generate ANNs that can predict with high accuracy the likelihood that a given sender will be later detected through other means to be a spammer. The input to the ANN will be a set of time-domain variables describing the sender's behavior - things such as email volume, number of valid and invalid message recipients, and content filtering results. The output will be a set of probability scores: is this sender likely to later be identified as a spammer by other means such as content filtering or human feedback? We hypothesize that the neural network approach will be more accurate and cost effective than the manual approach.
Acronym: 
MCAB
Project ID: 
10 757
Start date: 
01-08-2016
Project Duration: 
13months
Project costs: 
200 000.00€
Technological Area: 
Computer Software technology
Market Area: 
undefined

Raising the productivity and competitiveness of European businesses through technology. Boosting national economies on the international market, and strengthening the basis for sustainable prosperity and employment.