Automated big text security classification

Alzhrani, Khudran; Rudd, Ethan M.; Boult, Terrance E.; Chow, Chee-Onn

doi:10.1109/isi.2016.7745451

Cited by 14 publications

(9 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The majority of the literature on sensitivity labelling is associated with Data Loss Prevention (DLP) systems [8][9][10][11][12], notably focusing on classifying sensitivity at the document level. Other research uses sensitivity classification for confidential information redaction on declassified documents [13][14][15], where classification is often performed at finer granularity that reaches the token level.…”

Section: Related Workmentioning

confidence: 99%

“…Later, MacDonald et al built a novel SVM sensitivity classifier by mixing concepts from both NLP and machine learning for government document declassification, using POS n-gram tags as a sensitivity load indicator [15]. Alzhrani et al [9] proposed another DLP system with more fine-grained granularity. Their work effectively combined unsupervised and supervised methods to create a similarity-based classifier operating on a paragraph level and trained on an ad-hoc annotated WikiLeaks corpus.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Using BiLSTM Networks for Context-Aware Deep Sensitivity Labelling on Conversational Data

Pogiatzis

Samakovitis

2020

Applied Sciences

View full text Add to dashboard Cite

Information privacy is a critical design feature for any exchange system, with privacy-preserving applications requiring, most of the time, the identification and labelling of sensitive information. However, privacy and the concept of “sensitive information” are extremely elusive terms, as they are heavily dependent upon the context they are conveyed in. To accommodate such specificity, we first introduce a taxonomy of four context classes to categorise relationships of terms with their textual surroundings by meaning, interaction, precedence, and preference. We then propose a predictive context-aware model based on a Bidirectional Long Short Term Memory network with Conditional Random Fields (BiLSTM + CRF) to identify and label sensitive information in conversational data (multi-class sensitivity labelling). We train our model on a synthetic annotated dataset of real-world conversational data categorised in 13 sensitivity classes that we derive from the P3P standard. We parameterise and run a series of experiments featuring word and character embeddings and introduce a set of auxiliary features to improve model performance. Our results demonstrate that the BiLSTM + CRF model architecture with BERT embeddings and WordShape features is the most effective (F1 score 96.73%). Evaluation of the model is conducted under both temporal and semantic contexts, achieving a 76.33% F1 score on unseen data and outperforms Google’s Data Loss Prevention (DLP) system on sensitivity labelling tasks.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Using BiLSTM Networks for Context-Aware Deep Sensitivity Labelling on Conversational Data

Pogiatzis

Samakovitis

2020

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…The data collection must also adhere to the conditions as in Table 8. Table 8 shows that the data to be used in this study must be reported incidents, endorsed by authorized organization which having their own of third party security auditor, and should be open to public as it indicates a collective and ability to produce a predictive trend [39]. In market, the vulnerability database can be from local authorities or international authorities.…”

Section: Data Collectionmentioning

confidence: 99%

Enhancing the Secured Software Framework using Vulnerability Patterns and Flow Diagrams

Hassan¹,

Bahaman²,

Hussin³

et al. 2018

ijacsa

View full text Add to dashboard Cite

This article describes the process of simplifying the software security classification. The inputs of this process include a reference model from previous researcher and existing Common Vulnerabilities and Exposure (CVE) database. An interesting aim is to find out how we can make the secured software framework implementable in practice. In order to answer this question, some inquiries were set out regarding reference model and meta-process for classification to be a workable measurement system. The outputs of the process are the results discussion of experimental result and expert's validation. The experimental result use the existing CVE database which serves as an analysis when a) the framework is applied on three mix datasets, and b) when the framework is applied on two focus datasets. The first explains the result when the framework is applied on the CVE data randomly which consist mix of vendors and the latter is applied on the CVE data randomly but on selective vendors. The metric used in this assessment are precision and recall rate. The result shows there is a strong indicator that the framework can produce acceptable output accuracy. Apart from that, several experts' views were discussed to show the correctness and eliminate the ambiguity of classification rules and to prove the whole framework process.

show abstract

“…In the ACESS paper [1], we generated a wide range of clusters and built a classification model based on each of these clusters. The intuition behind this was to find the group of classification models with the best results via a crossvalidation technique.…”

Section: B Automated Security Classification Enabled By Similarity (mentioning

confidence: 99%

“…Unfortunately, real sensitive text datasets are generally kept private, so these works were not able to perform realistic evaluations, instead using fictitious "sensitive" texts (e.g., collected from public Twitter feeds). To perform a more realistic evaluation of the problem, we introduced the first public sensitive text dataset in [1], consisting of diplomatic cables made public by the WikiLeaks organization, and performed evaluations using a novel system design -Automated Classification Enabled by Security Similarity (ACESS) -which performs local classifier learning over clusters of similarity groups. This work, while seminal, had the disadvantage that the protocol was inherently expensive to tune hyperparameters over and had no separate validation set.…”

Section: Related Workmentioning

confidence: 99%

Automated U.S diplomatic cables security classification: Topic model pruning vs. classification based on clusters

Alzhrani

Rudd²,

Chow

et al. 2017

2017 IEEE International Symposium on Technologies for Homeland Security (HST)

Self Cite

View full text Add to dashboard Cite

The U.S Government has been the target for cyberattacks from all over the world. Just recently, former President Obama accused the Russian government of the leaking emails to Wikileaks and declared that the U.S. might be forced to respond. While Russia denied involvement, it is clear that the U.S. has to take some defensive measures to protect its data infrastructure. Insider threats have been the cause of other sensitive information leaks too, including the infamous Edward Snowden incident. Most of the recent leaks were in the form of text. Due to the nature of text data, security classifications are assigned manually. In an adversarial environment, insiders can leak texts through E-mail, printers, or any untrusted channels. The optimal defense is to automatically detect the unstructured text security class and enforce the appropriate protection mechanism without degrading services or daily tasks. Unfortunately, existing Data Leak Prevention (DLP) systems are not well suited for detecting unstructured texts. In this paper, we compare two recent approaches in the literature for text security classification, evaluating them on actual sensitive text data from the WikiLeaks dataset.

show abstract

Automated big text security classification

Cited by 14 publications

References 12 publications

Using BiLSTM Networks for Context-Aware Deep Sensitivity Labelling on Conversational Data

Using BiLSTM Networks for Context-Aware Deep Sensitivity Labelling on Conversational Data

Enhancing the Secured Software Framework using Vulnerability Patterns and Flow Diagrams

Automated U.S diplomatic cables security classification: Topic model pruning vs. classification based on clusters

Contact Info

Product

Resources

About