An experimental comparison of naive Bayesian and keyword-based anti-spam filtering with personal e-mail messages

By Ion Androutsopoulos, John Koutsias, Konstantinos V. Chandrinos, Constantine D. Spyropoulos. Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, 2000

563 citations recorded by the source corpus.

The growing problem of unsolicited bulk e-mail, also known as “spam”, has generated a need for reliable anti-spam e-mail filters. Filters of this type have so far been based mostly on manually constructed keyword patterns. An alternative approach has recently been proposed, whereby a Naive Bayesian classifier is trained automatically to detect spam messages. We test this approach on a large collection of personal e-mail messages, which we make publicly available in “encrypted” form contributing towards standard benchmarks. We introduce appropriate cost-sensitive measures, investigating at the same time the effect of attribute-set size, training-corpus size, lemmatization, and stop lists, issues that have not been explored in previous experiments. Finally, the Naive Bayesian filter is compared, in terms of performance, to a filter that uses keyword patterns, and which is part of a widely used e-mail reader.

Subfields: Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/46711283e72e45dc2efe1b86db8db6ce65efd82c

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

Hidden Voice Commands (2016, unknown): “Attacks on machine learning have been studied extensively in other contexts [1, 4, 5, 10, 13, 22, 31, 40]; In particular, Fawzi et al.”
A comparison of machine learning techniques for phishing detection (2007, background): “According to [2], spam recall measures the percentage of spam messages that the ﬁlter manages to block (ﬁlter’s effectiveness).”
An evaluation of statistical spam filtering techniques (2004, methodology): “WAcc was introduced by Androutsopoulos et al. [2000] and has been used in several spam .ltering benchmarks [Androutsopoulos et al. 2000; Carreras and M´ arquez 2001].”
Bayesian reasoning and machine learning (2012, unknown): “From this data, the count vector for person 1 is [13, 3, 4] and for person 2, [4, 9, 7].”
A survey of learning-based techniques of email spam filtering (2008, background): “The shortest among the popular definitions characterizes spam as " unsolicited bulk email " [3] [90].”
Hierarchical Label Propagation and Discovery for Machine Generated Email (2016, background): “As a result of these complexities, greylists and machine learning techniques have become the default methods for spam detection, many of which utilize textual analysis methods [21, 3, 25].”
Aiding the Detection of Fake Accounts in Large Scale Social Online Services (2012, methodology): “However, they follow the random walk sampling paradigm instead of our power-iteration method (§4.2), incurring a cost as high as SybilLimit.”
Using naive bayes to detect spammy names in social networks (2013, methodology): “In typical email spam classiﬁers using Naive Bayes (e.g., [25, 26, 1]) the features w consist of words or phrases in a document, with separation indicated by whitespace characters or punctuation.”
Spam Filtering Based On The Analysis Of Text Information Embedded Into Images (2006, methodology): “To this aim, the minimization of a weighted combination of false positive and false negative misclassification rates (denoted from now on respectively as FP and FN) was proposed in several works (for instance, Androutsopoulos et al., 2000; Wang & Cloete, 2005).”
Transfer learning using computational intelligence: A survey (2015, background): “Though this assumption is not suitable in most real scenarios, naïve Bayes classifiers have nevertheless been proved to work quite well in some complicated applications, especially automatic medical diagnosis [68], spam filtering [69] and text categorization [70], in which they may even outperform m”
Machine learning for email spam filtering: review, approaches and open research problems (2019, methodology): “According to [50,51], the cost of false positives is much more ( λ times) than that of false negatives, where lambda ( λ ) is a numerical factor that stipulates how ‘ risky or ‘ harmful it is to wrongly classify a valid e-mail as spam.”
A survey and experimental evaluation of image spam filtering techniques (2011, methodology): “…techniques based on machine learning have been widely investigated by several authors, to implement textual content analysis (Sahami et al., 1998; Drucker et al., 1999; Androutsopoulos et al., 2000; Graham, 2002a; Robinson, 2003; Meyer and Whateley, 2004; Koprinska et al., 2007; Lai et al., 2009).”
A three-layer back-propagation neural network for spam detection using artificial immune concentration (2009, methodology): “Two benchmark corpora used to test our proposed CFC approach based BPNN classifier are the PU1 corpus ( Androutsopoulos et al. 2000a, b) and Ling corpus 1”
Learning to Filter Spam E-Mail: A Comparison of a Naive Bayesian and a Memory-Based Approach (2000, methodology): “4 An earlier summary of our experiments with the Naive Bayesian classifier, not comprising the experiments with TiMBL and Outlook 2000, can be found in (Androutsopoulos et al. 2000a).”