Towards Robust and Privacy-preserving Text Representations

Li, Yitong; Baldwin, Timothy; Cohn, Trevor

doi:10.18653/v1/p18-2005

Cited by 118 publications

(127 citation statements)

References 17 publications

Supporting

Mentioning

121

Contrasting

Order By: Relevance

“…Others have attempted to remove biases from learned representations, e.g., gender biases in word embeddings (Bolukbasi et al, 2016) or sensitive information like sex and age in text representations (Li et al, 2018). However, removing such attributes from text representations may be difficult (Elazar & Goldberg, 2018).…”

Section: Fine-tuning On Target Datasetsmentioning

confidence: 99%

Don’t Take the Premise for Granted: Mitigating Artifacts in Natural Language Inference

Belinkov¹,

Poliak²,

Shieber³

et al. 2019

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics

View full text Add to dashboard Cite

Natural Language Inference (NLI) datasets often contain hypothesis-only biases-artifacts that allow models to achieve non-trivial performance without learning whether a premise entails a hypothesis. We propose two probabilistic methods to build models that are more robust to such biases and better transfer across datasets. In contrast to standard approaches to NLI, our methods predict the probability of a premise given a hypothesis and NLI label, discouraging models from ignoring the premise. We evaluate our methods on synthetic and existing NLI datasets by training on datasets containing biases and testing on datasets containing no (or different) hypothesis-only biases. Our results indicate that these methods can make NLI models more robust to dataset-specific artifacts, transferring better than a baseline architecture in 9 out of 12 NLI datasets. Additionally, we provide an extensive analysis of the interplay of our methods with known biases in NLI datasets, as well as the effects of encouraging models to ignore biases and fine-tuning on target datasets. 1 * * Equal contribution 1 Our code is available at https://github.com/ azpoliak/robust-nli.2 This hypothesis contradicts the premise and would likely not be inferred.

show abstract

Section: Fine-tuning On Target Datasetsmentioning

confidence: 99%

Don’t Take the Premise for Granted: Mitigating Artifacts in Natural Language Inference

Belinkov¹,

Poliak²,

Shieber³

et al. 2019

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics

View full text Add to dashboard Cite

show abstract

“…Users can also provide some demographic information.In the collected dataset, each review is associated with three attributes, gender (male/female), age, and location (Denmark, France, United Kingdom, and United States). We follow the same approach as in [39] and discard all non-English reviews based on LANGID.PY 5 [40], and only keep reviews classified as English with a confidence greater than 0.9. We follow the setting of [32] and categorize age attribute into three groups, over- 45, under-35, and between 35 and 45.…”

Section: Datamentioning

confidence: 99%

“…This data is obtained from Hovy et al [32] and consists of 600 sentences, each tagged with POS information based on the Google Universal POS tagset [45] and also labeled with both gender and age of the users. The gender attribute is categorized into male and female, and age attribute is categorized into two groups over- 45, under-35. We follow the setting of [39] and use Web English Tree-bank (WebEng) [16] as a pre-training tagging model because of the small quantity of text available for this task. WebEng is similar to TrustPilot datasets w.r.t.…”

Section: Task 2: Part-of-speech (Pos) Taggingmentioning

confidence: 99%

“…Textual Data Anonymization. Few works consider the privacy of textual user data [6,29,39,50]. The work of [29] introduces possible privacy threats of document repositories, 1) name entity recognition, and 2) author identification.…”

Section: Related Workmentioning

confidence: 99%

“…The work of Zhang et al [50] introduces a verified version of differential privacy specified for textual data to overcome the curse of dimensionality problem, namely, ϵ-TextIndistinguishability and satisfy it by adding Laplacian noise. Another work [39] uses the idea of adversarial learning to generate text representation. Their framework consists of a generator which generates representation w.r.t.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Privacy Preserving Text Representation Learning

Beigi

Shu

Guo

et al. 2019

Proceedings of the 30th ACM Conference on Hypertext and Social Media

View full text Add to dashboard Cite

Online users generate tremendous amounts of textual information by participating in different activities, such as writing reviews and sharing tweets. This textual data provides opportunities for researchers and business partners to study and understand individuals. However, this user-generated textual data not only can reveal the identity of the user but also may contain individual's private information (e.g., age, location, gender). Hence, "you are what you write" as the saying goes. Publishing the textual data thus compromises the privacy of individuals who provided it. The need arises for data publishers to protect people's privacy by anonymizing the data before publishing it. It is challenging to design effective anonymization techniques for textual information which minimizes the chances of re-identification and does not contain users' sensitive information (high privacy) while retaining the semantic meaning of the data for given tasks (high utility). In this paper, we study this problem and propose a novel double privacy preserving text representation learning framework, DPText, which learns a textual representation that (1) is differentially private, (2) does not contain private information and (3) retains high utility for the given task. Evaluating on two natural language processing tasks, i.e., sentiment analysis and part of speech tagging, we show the effectiveness of this approach in terms of preserving both privacy and utility.

show abstract

Generalised Differential Privacy for Text Document Processing

Fernandes

Dras

McIver

2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

We address the problem of how to "obfuscate" texts by removing stylistic clues which can identify authorship, whilst preserving (as much as possible) the content of the text. In this paper we combine ideas from "generalised differential privacy" and machine learning techniques for text processing to model privacy for text documents. We define a privacy mechanism that operates at the level of text documents represented as "bags-of-words" -these representations are typical in machine learning and contain sufficient information to carry out many kinds of classification tasks including topic identification and authorship attribution (of the original documents). We show that our mechanism satisfies privacy with respect to a metric for semantic similarity, thereby providing a balance between utility, defined by the semantic content of texts, with the obfuscation of stylistic clues. We demonstrate our implementation on a "fan fiction" dataset, confirming that it is indeed possible to disguise writing style effectively whilst preserving enough information and variation for accurate content classification tasks.

show abstract

Towards Robust and Privacy-preserving Text Representations

Cited by 118 publications

References 17 publications

Don’t Take the Premise for Granted: Mitigating Artifacts in Natural Language Inference

Don’t Take the Premise for Granted: Mitigating Artifacts in Natural Language Inference

Privacy Preserving Text Representation Learning

Generalised Differential Privacy for Text Document Processing

Contact Info

Product

Resources

About