Abstract: P2P streaming media has taken up a lot of bandwidth, and is prone to spread the virus, so is required to be identified accurately. This paper analyzed the shortcomings of the method P-Abacus, and proposed a kind of P2P streaming media recognition method P-Abacus based on SVM probabilistic output. P-Abacus can express it in probabilistic output, which reflects the extent of the sample belonging to known applications. We ordered the output, and according to maximum probability,made a judgement on that whether the sample belongs to class of maximum probability or unknown,or needs a further judgement If a further judgement is needed,we calculated the probabilistic output difference of the SVM built between the two largest classes, and made sure that whether the sample belongs to one of the two largest classes,or unknown. Thus P-Abacus has a better recognition effect,because probabilistic output contains more information that can be utilized. Experiments show that P-Abacus has a higher recognition rate and a lower false positive rate than Abacus, and has a limited increase of time overhead.

Key words: P2P streaming media, Recognition, SVM, Probabilistic output, Endpoint