This study introduces a new DQN-based Cross-Stream Crisis event Summarizer that addresses the challenges in summarizing multiple disaster-relevant data streams. It uses Deep Reinforcement Learning (DRL) to retrieve relevant texts from input streams without human supervision. The approach incorporates a redundancy filter into the reward function to handle cross-stream content overlaps. It has shown superior results on the CrisisFACTS 2022 benchmark, outperforming existing models.
Publication date: 12 Jan 2022
Project Page: https://arxiv.org/abs/2401.06683v1
Paper: https://arxiv.org/pdf/2401.06683
