Improved method of filtration of undesired traffic

The study provides an improved method of analysis and filtering of information flow based on Bayesian networks, which takes into account the thematic focus of communications. Preliminary identification of electronic communication has allowed not only to improve the efficiency of a filter based on Bayesian networks, but also to reduce the primary and secondary errors, when deciding on the classification of mail as junk mail. The software product that implements the proposed technique has been developed, and pilot testing of the validity of the proposed approach has been made. The obtained results show that the effectiveness of the developed method is higher by 13 % than the base case efficiency in filtering spam. In addition, when using this approach the probability of the primary and secondary errors, the main indicators of the quality of spam filters is reducing. Secondary error, when a legitimate message was considered as spam, was 2.8 times less than when using the basic option.