TREC 2007 Spam Track Overview

By Gordon V. Cormack. 2007

249 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

On Security and Sparsity of Linear Classifiers for Adversarial Settings (2016, methodology): “For our experiments we use the TREC 2007 spam track data, consisting of about 25000 legitimate and 50000 spam emails [22].”
Adversarial Feature Selection Against Evasion Attacks (2016, background): “of 25 220 legitimate and 50 199 real spam emails [67].”
Slow Learners are Fast (2009, methodology): “In particular, we used two diﬀer-ent training sets that were based on e-mails: the TREC dataset (Cormack, 2007), consisting of 75,419 e-mail messages, and a proprietary (signiﬁcantly harder) dataset of which we took 100,000 e-mails.”
Feature Weighting for Improved Classifier Robustness (2009, methodology): “Dataset d #tr #va #ev Ref TREC05 208,844 24,582 6,145 61,455 [5] TREC06 133,495 10,086 2,521 25,241 [4] ECML06 206,908 3,200 800 7,500 [1] Hotmail 2,644,921 688,500 76,500 150,000 [13]”
A survey and experimental evaluation of image spam filtering techniques (2011, methodology): “A third data set consists of images extracted from the publicly available TREC 2007 spam corpus (Cormack, 2007), made up of messages collected in 2007 from different accounts, including honeypots.”
Multiple classifier systems for robust classifier design in adversarial environments (2010, methodology): “We used the benchmark TREC 2007 email corpus [15], publicly available at http://plg.”
Evaluating sentence-level relevance feedback for high-recall information retrieval (2018, methodology): “The Track coordinators used an iterative process (Cormack and Lynam, 2005a) in which a number of spam classiﬁers were applied to the corpus, and disagreements between the classiﬁers and a provisional labeling were adjudicated by the coordinators.”
Design of robust classifiers for adversarial environments (2011, methodology): “We experiment on the benchmark TREC 2007 email corpus [20], which is made up of 75,419 real emails (25,220 legitimate and 50,199 spam messages), collected between April and July 2007.”
A review of machine learning approaches to Spam filtering (2009, unknown): “However, in order to have an adequate assessment of the performance of filters, it is necessary to adopt more realistic evaluation settings (e.g., the TREC corpora, Cormack, 2006, 2007; Cormack & Lynam, 2005), that better mimic the scenario faced by a filter deployed for practical operation.”
Secure Kernel Machines against Evasion Attacks (2016, methodology): “For our experiments we use the TREC 2007 spam track data, consisting of about 25000 legitimate and 50000 spam emails [11].”
One-and-a-Half-Class Multiple Classifier Systems for Secure Learning Against Evasion Attacks at Test Time (2015, methodology): “We use the benchmark TREC 2007 email corpus [20], which consists of 25,220 ham and 50,199 spam emails.”
Semi-supervised Classification for Natural Language Processing (2014, methodology): “On the other hand, in 2007 TREC Spam Track Challenge [15], the participating spam ﬁlters were trained with publicly available emails and their model accuracy was tested on emails collected from user inboxes (i.e., personalized emails).”
Spam Filtering Using Inexact String Matching in Explicit Feature Space with On-Line Linear Classifiers (2006, methodology): “However, as seen in the 2005 TREC Spam Filtering Competition (Cormack and Lynam, 2005b), even state of the art methods yield imperfect results.”
Information Retrieval Evaluation (2011, methodology): “The toolkit also was used in the evaluation to implement the various feedback tasks and to record and score results [58].”
A survey of learning-based techniques of email spam filtering (2008, methodology): “According to the final report [18], the best performance was shown by one of the filters supplied by Jožef Stefan Institute and based on compression models [11], able to achieve spam misclassification rate of 1.”