Challenge: Documents revealing sensitive information about individuals must typically be de-identified by masking all mentions of PII .
Approach: They propose a method that attempts to reverse the process of re-identification based on a database of documents representing background knowledge.
Outcome: The proposed method can recover as many as 80% of de-identified text spans and improve accuracy along with the level of background knowledge.

Similar Papers

De-Identification of Sensitive Personal Data in Datasets Derived from IIT-CDIP (2024.emnlp-main)

Copied to clipboard

Challenge: Large volumes of data are becoming increasingly important for training machine learning models for document understanding tasks like classification, information extraction, and visual question answering.
Approach: They propose a data de-identification pipeline that replaces sensitive data with synthetic, but realistic, data that preserves the utility of de-identified documents.
Outcome: The proposed method preserves the utility of the de-identified documents so that they can continue to be used in various document understanding applications.
Adversarial Learning of Privacy-Preserving Text Representations for De-Identification of Medical Records (P19-1)

Copied to clipboard

Challenge: De-identification is the task of detecting protected health information (PHI) in medical text.
Approach: They propose to create shareable representations of medical text that contain no PHI and can be shared between organizations to create unified datasets for training de-identification models.
Outcome: The proposed representation allows training a simple LSTM-CRF model to an F1 score of 97.4%.
Gumbel Reranking: Differentiable End-to-End Reranker Optimization (2025.acl-long)

Copied to clipboard

Challenge: Existing distillation-based approaches suffer from training-inference misalignment and fail to capture interdependencies among candidate documents.
Approach: They propose a method to optimize rerankers by learning a stochastic, document-wise Top-k attention mask using the Gumbel Trick and Relaxed Top-K Sampling.
Outcome: The proposed framework minimizes the overall language loss and improves recall on hotpotQA.
Iterative Self-Correction for Text-Driven Person Re-Identification with Large Vision-Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Existing methods for Person Re-Identification (ReID) adopt a static "one-pass" paradigm, converting images to text once for retrieval.
Approach: They propose a framework that reformulates ReID as an iterative "Think-and-Refine" process.
Outcome: The proposed framework outperforms state-of-the-art methods in complex occlusion scenarios.
Assisting the Human Fact-Checkers: Detecting All Previously Fact-Checked Claims in a Document (2022.findings-emnlp)

Copied to clipboard

Challenge: Recent years have brought us a proliferation of false claims online, which spread fast . fact-checkers have been using automated fact-finding to verify claims .
Approach: They propose a system that can detect claims that can be fact-checked by a given database . they create a manually annotated document dataset and propose evaluation measures .
Outcome: The proposed system achieves sizable performance gains over strong baselines.
Not What the Doctor Ordered: Surveying LLM-based De-identification and Quantifying Clinical Information Loss (2025.emnlp-main)

Copied to clipboard

Challenge: De-identification is an application of NLP where automated algorithms remove identifying information of patients and providers.
Approach: They propose to use generative large language models to de-identify patients and providers . they propose to validate existing metrics to quantify extent of inappropriate removal .
Outcome: The proposed method is based on a survey of LLM-based de-identification research . it shows that the models perform poorly in identifying clinically relevant changes .
RedactOR: An LLM-Powered Framework for Automatic Clinical Data De-Identification (2025.acl-industry)

Copied to clipboard

Challenge: Existing de-identification methods suffer from recall errors, limited generalization, and inefficiencies, limiting their real-world applicability.
Approach: They propose a multi-modal framework for de-identifying electronic health records using a retrieval-based entity relexicalization approach.
Outcome: The proposed framework achieves competitive performance while optimizing token usage to reduce LLM costs.
Privacy Ripple Effects from Adding or Removing Personal Information in Language Model Training (2025.findings-acl)

Copied to clipboard

Challenge: PII is a sensitive information that can be removed from large-language model training due to evolving curation techniques, or because it was recently scraped for retraining.
Approach: They characterize a phenomenon where PII that appeared earlier in training becomes extractable at a later step after fine-tuning on other PI I.
Outcome: The authors show that PII memorization is a dynamic property of a model that evolves throughout training pipelines and depends on commonly altered design choices.
Downstream Task Performance of BERT Models Pre-Trained Using Automatically De-Identified Clinical Data (2022.lrec-1)

Copied to clipboard

Challenge: Automatic de-identification systems introduce errors due to their imperfect precision and may negatively impact the utility of the de-identified dataset.
Approach: They propose to de-identifie a large clinical corpus in Swedish by removing entire sentences containing sensitive data or by replacing sensitive words with realistic surrogates.
Outcome: The proposed models are safe to distribute to other academic researchers and reduce privacy risks.
Generating Synthetic Free-text Medical Records with Low Re-identification Risk using Masked Language Modeling (2025.naacl-srw)

Copied to clipboard

Challenge: Existing methods to generate medical records using Causal Language Modelling are limited due to privacy concerns.
Approach: They propose a method for generating medical records using Masked Language Modelling using Causal language models.
Outcome: The proposed method produces high-quality synthetic data with a re-identification risk of only 3.5% and a patient recall of 96%.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations