Seewald, Alexander K.Alkassar, AmmarSiekmann, Jörg2019-04-032019-04-032008978-3-88579-222-2https://dl.gi.de/handle/20.500.12116/21484Spam has become a problem of global impact. Most spam messages are currently sent out by captured machines organized in bot networks, which are infected with malicious software and are therefore under direct control of spammers. The connected explosion of automatically generated new malware variants has manual analysis at a great disadvantage, while classical automated analysis systems have problems keeping up with the diversity of new variants. Here, we propose using machine learning approaches to learn global (i.e. malware intent) and local (i.e. specific functionality) malware properties based on behavioral traces of malware recorded in virtual environments, and test them on a small corpus. Initial results are somewhat promising, so we also discuss areas for improvement as well as current and future challenges.enText/Conference Paper1617-5468