Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EROS: Entity-Driven Controlled Policy Document Summarization

Joykirat Singh, Sehban Fazili, Rohan Jain, Md Shad Akhtar

TL;DR

The proposed model, EROS, identifies critical entities through a span-based entity extraction model and employs them to control the information content of the summaries using proximal policy optimization (PPO) and shows encouraging improvement over various baselines.

Abstract

Privacy policy documents have a crucial role in educating individuals about the collection, usage, and protection of users' personal data by organizations. However, they are notorious for their lengthy, complex, and convoluted language especially involving privacy-related entities. Hence, they pose a significant challenge to users who attempt to comprehend organization's data usage policy. In this paper, we propose to enhance the interpretability and readability of policy documents by using controlled abstractive summarization -- we enforce the generated summaries to include critical privacy-related entities (e.g., data and medium) and organization's rationale (e.g.,target and reason) in collecting those entities. To achieve this, we develop PD-Sum, a policy-document summarization dataset with marked privacy-related entity labels. Our proposed model, EROS, identifies critical entities through a span-based entity extraction model and employs them to control the information content of the summaries using proximal policy optimization (PPO). Comparison shows encouraging improvement over various baselines. Furthermore, we furnish qualitative and human evaluations to establish the efficacy of EROS.

EROS: Entity-Driven Controlled Policy Document Summarization

TL;DR

The proposed model, EROS, identifies critical entities through a span-based entity extraction model and employs them to control the information content of the summaries using proximal policy optimization (PPO) and shows encouraging improvement over various baselines.

Abstract

Privacy policy documents have a crucial role in educating individuals about the collection, usage, and protection of users' personal data by organizations. However, they are notorious for their lengthy, complex, and convoluted language especially involving privacy-related entities. Hence, they pose a significant challenge to users who attempt to comprehend organization's data usage policy. In this paper, we propose to enhance the interpretability and readability of policy documents by using controlled abstractive summarization -- we enforce the generated summaries to include critical privacy-related entities (e.g., data and medium) and organization's rationale (e.g.,target and reason) in collecting those entities. To achieve this, we develop PD-Sum, a policy-document summarization dataset with marked privacy-related entity labels. Our proposed model, EROS, identifies critical entities through a span-based entity extraction model and employs them to control the information content of the summaries using proximal policy optimization (PPO). Comparison shows encouraging improvement over various baselines. Furthermore, we furnish qualitative and human evaluations to establish the efficacy of EROS.
Paper Structure (17 sections, 5 equations, 2 figures, 7 tables)

This paper contains 17 sections, 5 equations, 2 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Motivation and Problem Definition:
  3. Contribution:
  4. Reproducibility:
  5. Related Work
  6. Dataset Construction
  7. Data collection and Filtering:
  8. Data Annotation:
  9. Proposed Methodology - EROS
  10. Contrastive Loss:
  11. Span Prediction:
  12. Reward calculation:
  13. Experiments and Results
  14. Baselines:
  15. Result Analysis:
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Annotation labels with their definitions, inter-relationships, and an annotated instance. Best viewed in colors.
  • Figure 2: Left: Proposed Model for EROS. The reference model is a frozen pre-trained BART-based model with modified loss. We initialize the controlled summarization model in a similar way, which is subsequently updated through a PPO framework on a combination of rewards and KL-divergence loss. Right: Entity extraction model jointly learns a entity classification and entity identification module with the assistance of contrastive loss. Further to minimize the effect of false positives in identification, we supplement it with a entity classification module in a joint framework.