An evaluation of Naive Bayesian anti-spam filtering

By Ion Androutsopoulos, John Koutsias, Konstantinos V. Chandrinos, G. Paliouras, Constantine D. Spyropoulos. 2000

697 citations recorded by the source corpus.

It has recently been argued that a Naive Bayesian classifier can be used to filter unsolicited bulk e-mail (“spam”). We conduct a thorough evaluation of this proposal on a corpus that we make publicly available, contributing towards standard benchmarks. At the same time we investigate the effect of attribute-set size, training-corpus size, lemmatization, and stop-lists on the filter’s performance, issues that had not been previously explored. After introducing appropriate cost-sensitive evaluation measures, we reach the conclusion that additional safety nets are needed for the Naive Bayesian anti-spam filter to be viable in practice.

Subfields: Artificial Intelligence, Information Systems

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

Text Categorization as a Graph Classification Problem (2015, methodology): “Another popular approach is the use of Naive Bayes and its multiple variants (McCallum and Nigam, 1998), in particular for the subtask of spam detection (Androutsopoulos et al., 2000).”
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].”
On Attacking Statistical Spam Filters (2004, methodology): “One of the widely used methods, Bayesian classication, attempts to calculate the probability that a message is spam based upon previous feature frequencies in spam and legitimate e-mail [9, 8, 1 , 4]. Other notable learning algorithms applied to spam ltering include boosting [2] and support vector ”
Feature Cross-Substitution in Adversarial Classification (2014, methodology): “Our evaluation uses three data sets: Enron email data [21], Ling-spam data [22], and internet advertisement dataset from the UCI repository [23].”
A comparison of machine learning techniques for phishing detection (2007, methodology): “In addition to the previous metrics we use cost-sensitive measures namely the “weighted accuracy” and the “weighted error” measures proposed in [1] and [25].”
ACAS: automated construction of application signatures (2005, methodology): “A system based on Naive Bayes can be continuously updated as new labeled data becomes available, this is why it is widely used for applications such as SPAM filtering [1].”
A survey of the applications of text mining in financial domain (2016, background): “Androutsopoulos et al. (2000a) proposed Naive Bayes based anti-spam model.”
Machine learning for email spam filtering: review, approaches and open research problems (2019, methodology): “A Naive Bayes (NB) classi ﬁ er simply apply Bayes' theorem on the context classi ﬁ cation of each email, with a strong assumption that the words included in the email are independent of each other [38].”
An improved K-nearest-neighbor algorithm for text categorization (2012, methodology): “…filtering, weighted accuracy (WAcc), total cost ratio (TCR), spam recall (SR), spam precision (SP), SF 1 (the combination of SR and SP) are widely used to evaluate the spam filter performance (Androutsopoulos et al.,2000a; Androutsopoulos et al.,2000b; Sakkis et al.,2001; Koprinska et al.,2007).”
Text Mining in Big Data Analytics (2020, background): “[114], was one of the first publicly available datasets.”
Can machine learning be secure? (2006, background): “Because this approach often fails to detect novel intrusions, a variety of researchers have proposed incorporating machine learning techniques into intrusion detection systems [1, 16, 18, 24, 38, 41].”
Applying lazy learning algorithms to tackle concept drift in spam filtering (2007, background): “with Applications legitimate message can be higher than letting a spam message pass the filter, and this difference must be taken into account during both training and evaluation tasks ( Androutsopoulos et al., 2000; Hidalgo, Lo ´ pez, & Sanz, 2000).”
Spam Detection Using Feature Selection and Parameters Optimization (2010, methodology): “The other approach is to perform parameters optimization every time when we eliminate each feature.”
Adversarial Examples for Malware Detection (2017, methodology): “Starting with the use of naive Bayes classifiers for spam detection [1], machine learning has been increasingly applied to solve core security problems.”
Random-Walk Term Weighting for Improved Text Classification (2006, methodology): “These datasets are commonly used for text classiﬁcation evaluations (Joachims, 1996; Craven et al., 1998; Androutsopoulos et al., 2000; Mihalcea and Hassan, 2005).”
Every Document Owns Its Structure: Inductive Text Classification via Graph Neural Networks (2020, methodology): “Traditional methods for text classiﬁcation include Naive Bayes (Androutsopoulos et al., 2000), k-Nearest Neighbor (Tan, 2006) and Support Vector Machine (Forman, 2008).”
Learning to Filter Spam E-Mail: A Comparison of a Naive Bayesian and a Memory-Based Approach (2000, methodology): “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 ).”