Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DQNC2S: DQN-based Cross-stream Crisis event Summarizer

Daniele Rege Cambrin, Luca Cagliero, Paolo Garza

TL;DR

This work proposes an online approach to crisis timeline generation based on weak annotation with Deep Q-Networks that selects on-the-fly the relevant pieces of text without requiring neither human annotations nor content re-ranking, which makes the inference time independent of the number of input queries.

Abstract

Summarizing multiple disaster-relevant data streams simultaneously is particularly challenging as existing Retrieve&Re-ranking strategies suffer from the inherent redundancy of multi-stream data and limited scalability in a multi-query setting. This work proposes an online approach to crisis timeline generation based on weak annotation with Deep Q-Networks. It selects on-the-fly the relevant pieces of text without requiring neither human annotations nor content re-ranking. This makes the inference time independent of the number of input queries. The proposed approach also incorporates a redundancy filter into the reward function to effectively handle cross-stream content overlaps. The achieved ROUGE and BERTScore results are superior to those of best-performing models on the CrisisFACTS 2022 benchmark.

DQNC2S: DQN-based Cross-stream Crisis event Summarizer

TL;DR

This work proposes an online approach to crisis timeline generation based on weak annotation with Deep Q-Networks that selects on-the-fly the relevant pieces of text without requiring neither human annotations nor content re-ranking, which makes the inference time independent of the number of input queries.

Abstract

Summarizing multiple disaster-relevant data streams simultaneously is particularly challenging as existing Retrieve&Re-ranking strategies suffer from the inherent redundancy of multi-stream data and limited scalability in a multi-query setting. This work proposes an online approach to crisis timeline generation based on weak annotation with Deep Q-Networks. It selects on-the-fly the relevant pieces of text without requiring neither human annotations nor content re-ranking. This makes the inference time independent of the number of input queries. The proposed approach also incorporates a redundancy filter into the reward function to effectively handle cross-stream content overlaps. The achieved ROUGE and BERTScore results are superior to those of best-performing models on the CrisisFACTS 2022 benchmark.
Paper Structure (16 sections, 1 equation, 1 figure, 2 tables)

This paper contains 16 sections, 1 equation, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Methodology
  4. Weak Annotation
  5. DQN-based Text Retrieval
  6. Topic Modeling and Abstraction
  7. Experiments
  8. Dataset
  9. Evaluation metrics
  10. Baselines
  11. Setup
  12. Quantitative results overview
  13. Inference time.
  14. DQN training and testing.
  15. Conclusion and Future Works
  16. ...and 1 more sections

Figures (1)

  • Figure 1: Mean percentage of "take" action per episode during training (a) and the mean number of retrieved text per event (b). The red lines indicate the mean value and the black line the maximum number of daily facts according to the NIST annotation.