10 research outputs found
Authorship Verification
No full textIn recent years, stylometry, the study of linguistic style, has become more prominent in security and privacy applications involving written language, mostly in digital and online domains. Although literature is abundant with computational stylometry research, the field of authorship verification is relatively unexplored. Authorship verification is the binary semi-open-world problem of determining whether a document is written by a given author or not. A key component in authorship verification techniques is confidence measurement, on which verification decisions are based, expressed by acceptance thresholds selected and tuned per need. This thesis demonstrates how utilization of confidence-based approaches in stylometric applications, and their combination with traditional approaches, can benefit classification accuracy, and allow new domains and problems to be analyzed. We start by motivating the usage of authorship verification approaches with two stylometric applications: native-language identification from non-native text and active linguistic user authentication. Next, we introduce the Classify-Verify algorithm, which integrates classification with binary verification, applied to several stylometric problems. Classify-Verify is proposed as an open-world alternative to restricted closed-world attribution methods, and is shown effective in dealing with possibly missing candidate authors by thwarting misclassifications, coping with various domains and scales, and even adversarial authors who try to fool the classifier.Ph.D., Computer Science -- Drexel University, 201
Breaking the Closed-World Assumption in Stylometric Authorship Attribution
No full textPart 2: Forensic TechniquesInternational audienceStylometry is a form of authorship attribution that relies on the linguistic information found in a document. While there has been significant work in stylometry, most research focuses on the closed-world problem where the author of the document is in a known suspect set. For open-world problems where the author may not be in the suspect set, traditional classification methods are ineffective. This paper proposes the “classify-verify” method that augments classification with a binary verification step evaluated on stylometric datasets. This method, which can be generalized to any domain, significantly outperforms traditional classifiers in open-world settings and yields an F1-score of 0.87, comparable to traditional classifiers in closed-world settings. Moreover, the method successfully detects adversarial documents where authors deliberately change their styles, a problem for which closed-world classifiers fail
Forecasting dengue fever in Brazil: An assessment of climate conditions
No full textLocal climate conditions play a major role in the biology of the Aedes aegypti mosquito, the main vector responsible for transmitting dengue, zika, chikungunya and yellow fever in urban centers. For this reason, a detailed assessment of periods in which changes in climate conditions affect the number of human cases may improve the timing of vector-control efforts. In this work, we develop new machine-learning algorithms to analyze climate time series and their connection to the occurrence of dengue epidemic years for seven Brazilian state capitals. Our method explores the impact of two key variables—frequency of precipitation and average temperature—during a wide range of time windows in the annual cycle. Our results indicate that each Brazilian state capital considered has its own climate signatures that correlate with the overall number of human dengue-cases. However, for most of the studied cities, the winter preceding an epidemic year shows a strong predictive power. Understanding such climate contributions to the vector’s biology could lead to more accurate prediction models and early warning systems. [ © 2019 Stolerman et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. DOI: https://doi.org/10.1371/journal.pone.0220106
Active Linguistic Authentication Using Real-Time Stylometric Evaluation for Multi-Modal Decision Fusion
No full textPart 2: Forensic TechniquesInternational audienceActive authentication is the process of continuously verifying a user based on his/her ongoing interactions with a computer. Forensic stylometry is the study of linguistic style applied to author (user) identification. This paper evaluates the Active Linguistic Authentication Dataset, collected from users working individually in an office environment over a period of one week. It considers a battery of stylometric modalities as a representative collection of high-level behavioral biometrics. While a previous study conducted a partial evaluation of the dataset with data from fourteen users, this paper considers the complete dataset comprising data from 67 users. Another significant difference is in the type of evaluation: instead of using day-based or data-based (number-of-characters) windows for classification, the evaluation employs time-based, overlapping sliding windows. This tests the ability to produce authentication decisions every 10 to 60 seconds, which is highly applicable to real-world active security systems. Sensor evaluation is conducted via cross-validation, measuring the false acceptance and false rejection rates (FAR/FRR). The results demonstrate that, under realistic settings, stylometric sensors perform with considerable effectiveness down to 0/0.5 FAR/FRR for decisions produced every 60 seconds and available 95% of the time
Towards Active Linguistic Authentication
No full textPart 8: ADVANCED FORENSIC TECHNIQUESInternational audienceBiometric technologies offer a new and effective means for securing computers against unauthorized access. Linguistic technologies and, in particular, authorship attribution technologies can assist in this effort. This paper reports on the results of analyzing a novel corpus that was developed to test the possibility of active linguistic authentication. The study collected the one-week work product of nineteen temporary workers in a simulated office environment. The results demonstrate that techniques culled from the field of authorship attribution can identify workers with more than 90% accuracy
Drexel University
No full textAbstract—Stylometry is a method for identifying anonymous authors of anonymous texts by analyzing their writing style. While stylometric methods have produced impressive results in previous experiments, we wanted to explore their performance on a challenging dataset of particular interest to the security research community. Analysis of underground forums can provide key information about who controls a given bot network or sells a service, and the size and scope of the cybercrime underworld. Previous analyses have been accomplished primarily through analysis of limited structured metadata and painstaking manual analysis. However, the key challenge is to automate this process, since this labor intensive manual approach clearly does not scale. We consider two scenarios. The first involves text written by an unknown cybercriminal and a set of potential suspects. This is standard, supervised stylometry problem made more difficult by multilingual forums that mix l33t-speak conversations with data dumps. In the second scenario, you want to feed a forum into an analysis engine and have it output possible doppelgängers, or users with multiple accounts. While other researchers have explored this problem, we propose a method that produces good results on actual separate accounts, as opposed to data sets created by artificially splitting authors into multiple identities. For scenario 1, we achieve 77 % to 84 % accuracy on private messages. For scenario 2, we achieve 94 % recall with 90% precision on blogs and 85.18 % precision with 82.14 % recall for underground forum users. We demonstrate the utility of our approach with a case study that includes applying our technique to the Carders forum and manual analysis to validate the results, enabling the discovery of previously undetected doppelgänger accounts. I
“Maybe, but it’s code’s all it is”: Thomas Pynchon, Cow Country, and Computational Stylometry
No full textIn mid-2015, Art Winslow caused something on an online furore when he suggested that the pseudonymously-authored novel by Adrian Jones Pearson, Cow Country, was, in fact, a work by Thomas Pynchon. A full-blown argument then erupted when this was countered by Nate Jones and Pynchon's own publisher. Indeed, Penguin thundered: “[w]e are Thomas Pynchon's publisher and this is not a book by Thomas Pynchon”.
While the great and the good of the contemporary republic of letters argued over authorship, however, a range of stylometric techniques exist that could assist in the debate. As the name implies, computational stylometry is the measurement (“metry”) of stylistic properties of texts (“stylo”) using computers. Stylometry, as a quantifying activity, has a long and varied history, from legal court cases where the accused was acquitted on the basis of stylometric evidence, to literary authorship attribution. In the latter case, as charted by Anthony Kenny, the discipline dates back to approximately 1851 when Augustus de Morgan suggested that a dispute over the attribution of certain epistles could be settled by measuring average word lengths and correlating them with known writings of St Paul. At the time of writing, according to Ariel Stolerman, computational forensic stylometry “can identify individuals in sets of 50 authors with better than 90% accuracy, and [can] even scaled to more than 100,000 authors”.
In this paper, I give a humanistic/critical background to stylometry and its important limitations before applying a range of stylometric techniques to the novels of Thomas Pynchon alongside that of “Pearson”. In particular, I examine the widely used unsupervised “Burrows's delta” algorithm of most-frequent-word comparisons as well as a part-of-speech frequency comparison using the Stanford PoS tagger.
At the close of the paper, I will give the results of my computational experiments, while still noting that we are far from having a perfect system for attribution. After all, literary forensics are almost always a post-facto attempt at attributing meaning, even in the anti-intentionalist schools. In this case, though, it may transpire that I have an answer (“maybe”). But as Pynchon puts it in Bleeding Edge: “it’s code’s all it is”, for sure
