For this year’s Spam track we used classifiers based on language models. These models are used to compute the log-likelihood for each individual message and then classify them as either ham or spam. Different data sets were used to train these language models. Our approach is simple, we initially create simple unigram language models and smooth the probabilities of unseen tokens by means of the...

The proliferation of unrestricted Internet access has brought the community spam which has become a serious problem costing companies billions of dollars per annum. Typical anti-spam measures, such as filtering and blocking techniques, exist but focus on solving the spam problem on the message transportation level. Using such techniques may have impacts beyond the realm of spam-filters and bloc...

This submission is aimed to benchmark of Vadis methodology in the context of spam detection. The work that has been done to provide these results can be separated in two different tasks: data preparation and modelisation .

The popularity of Twitter greatly depends on the quality and integrity of contents contributed by users. Unfortunately, Twitter has attracted spammers to post spam content which pollutes the community. Social spamming is more successful than traditional methods such as email spamming by using social relationship between users. Detecting spam is the first and very critical step in the battle of ...

This paper describes two innovative analysis methods for IPv4 address sets such as antispam blacklists. First, the contents analysis provides means of measuring key properties of any set of IPv4 addresses as well as revealing relationships between such sets. Second, the behavior analysis defines behavioral attributes of querying addresses and requested addresses. Furthermore, the behavior analy...

A self-learning system for preventing unsolicited commercial or bulk e-mail (UCE or UBE) is presented. It acts at the source of each e-mail message and controls the traffic going out of a network, thus avoiding common drawbacks of standard spam filtering techniques. The system is based on a token bucket mechanism proposed earlier. In this paper, it is shown how to develop this approach into a f...

Spammers intend to increase the PageRank of certain spam pages by creating a large number of links pointing to them. We propose a novel method based on the concept of personalized PageRank that detects pages with an undeserved high PageRank value without the need of any kind of white or blacklists or other means of human intervention. We assume that spammed pages have a biased distribution of p...

Spam comprises approximately 90 to 95 percent of all e-mail traffic on the Internet nowadays, and it is a problem far from being solved. The losses caused by spam are estimated to reach up to $87 billion yearly. When fighting spam, most of the proposals focus on the receiver-side and are usually resource-intensive in terms of processing requirements. In this paper we present an approach to addr...

In this paper, we propose a new asymmetric boosting method, Boosting with Different Costs. Traditional boosting methods assume the same cost for misclassified instances from different classes, and in this way focus on good performance with respect to overall accuracy. Our method is more generic, and is designed to be more suitable for problems where the major concern is a low false positive (or...

As we think about the history of spam reduction, we can see a gradual change in the approach over time, as the spam problem has changed. Many of us may think of spam as a new problem, but in fact, it goes back at least to 1975, as noted by the late Jon Postel.[1] At the start users were mostly “techies”, and spam mostly referred to Usenet newsgroup posts that got out of hand, wherein someone wo...