# Towards Automatic Comparison of Data Privacy Documents: A Preliminary Experiment on GDPR-like Laws

arXiv: https://arxiv.org/abs/2105.10117 · HTML: https://mlsanity.com/paper/2105.10117/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 19 references that are in the map.
- First posted on arXiv 21 May 2021.

## Builds on (most cited first, 7 of 7)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference](https://mlsanity.com/paper/1704.05426/) (2017)
- [A large annotated corpus for learning natural language inference](https://mlsanity.com/paper/1508.05326/) (2015)
- [TLDR: Extreme Summarization of Scientific Documents](https://mlsanity.com/paper/2004.15011/) (2020)
- [Ranking Sentences for Extractive Summarization with Reinforcement Learning](https://mlsanity.com/paper/1802.08636/) (2018)
- [Building Legal Case Retrieval Systems with Lexical Matching and Summarization using A Pre-Trained Phrase Scoring Model](https://mlsanity.com/paper/2009.14083/) (2020)
- [JNLP Team: Deep Learning for Legal Processing in COLIEE 2020](https://mlsanity.com/paper/2011.08071/) (2020)

## Abstract (arXiv)

General Data Protection Regulation (GDPR) becomes a standard law for data protection in many countries. Currently, twelve countries adopt the regulation and establish their GDPR-like regulation. However, to evaluate the differences and similarities of these GDPR-like regulations is time-consuming and needs a lot of manual effort from legal experts. Moreover, GDPR-like regulations from different countries are written in their languages leading to a more difficult task since legal experts who know both languages are essential. In this paper, we investigate a simple natural language processing (NLP) approach to tackle the problem. We first extract chunks of information from GDPR-like documents and form structured data from natural language. Next, we use NLP methods to compare documents to measure their similarity. Finally, we manually label a small set of data to evaluate our approach. The empirical result shows that the BERT model with cosine similarity outperforms other baselines. Our data and code are publicly available.