Towards a Classifier for Digital Sensitivity Review

Abstract

Abstract. The sensitivity review of government records is essential before they can be released to the official government archives, to prevent sensitive information (such as personal information, or that which is prejudicial to international relations) from being released. As records are typically reviewed and released after a period of decades, sensitivity review practices are still based on paper records. The transition to digital records brings new challenges, e.g. increased volume of digital records, making current practices impractical to use. In this paper, we describe our current work towards developing a sensitivity review classifier that can identify and prioritise potentially sensitive digital records for review. Using a test collection built from government records with real sensitivities identified by government assessors, we show that considering the entities present in each record can markedly improve upon a text classification baseline.

Enhancing Sensitivity Classification with Semantic Features Using Word Embeddings

“…preserving the privacy of personal data [6,7]. McDonald et al [3] was the first work to address the automatic classification of FOI exemptions. In that work, the authors presented a proof-of-concept classifier for classifying specific FOI exemptions, and found that extending text classification with additional features, such as the number of subjective sentences and a country risk score, could improve the effectiveness of text classification for specific sensitivities.…”

“…However, automatic sensitivity classification is a difficult task. For example, the UK Freedom of Information Act 2000 (FOIA) has 24 FOI exemptions 3 , each with wide-ranging sub-categories of exemptions. Figure 1 illustrates the scope of potentially sensitive information from just 2 of these 24 exemptions, namely International Relations and Personal Information.…”

“…Text classification [2] is one approach that has been shown to be promising as a basis for automatic sensitivity identification algorithms [3,4]. Usually, a text classification model is learned by observing statistical patterns in the distributions of individual key terms from example documents.…”

Opening Digital Archives and Collections with Emerging Data Analytics Technology: A Research Agenda

“…One potential cause for the need of restructuring of datasets could be based on the outcome of a sensitivity review process (cf. McDonald et al 2014). Archival information packages need also to be prepared for potentially many-sided use of data analytics technologies that may need to gather information from them, sometimes even dynamically, for enabling new kinds of dissemination information packages to respond varying user needs who want to access to digital archives and collections from varying viewpoints;…”

“…While the traditional approach has been a human-conducted sensitivity review record-by-record (cf. McDonald et al 2014), the concept of analytics-enhanced opening needs to go beyond plain human-conducted sensitivity reviews to innovate new approaches to creating opened dissemination information packages from the digitally preserved archival information packages that have been, in turn, produced through the process of analyticsprepared preservation. Data analytics can make different approaches to open digital archives possible: for example, potentially sensitive AIPs can be transformed to intermediary anonymized or pseudonymized resources to be opened in various ways later on, or less sensitive AIPs can be opened as such for further data analytics in connection to use.…”