Abstract: The current mass of malicious code reports has become a huge burden of cloud-security-based anti-virus network systems．The utilization of efficient,scientific and automatic classification mechanism is the basic premise for responding quickly to large-scale known and unknown malicious codes and their new variants．In order to achieve the automatic classification of malicious codes,we improved the Kohonen algorithm,a classic neural network model with no mentors and proposed a new neural network model A_Kohonen with part supervised learning of the process．Then the A_Kohonen-based automatic classification mechanism of malicious codes was provided to support anti-virus experts to refine and analyze malicious codes further．Experimental analysis shows that the mechanism can initially classify malicious codes effectively and accurately.

Key words: Malicious code,Report classification,Anti-virus,Neural network,Information security