Examining the State-of-the-Art in News Timeline Summarization

TL;DR

This paper reviews current news timeline summarization methods, introduces a new comprehensive dataset, and proposes a combined approach that outperforms existing techniques across multiple benchmarks.

Contribution

It provides a comparative analysis of TLS strategies, introduces a larger, more diverse dataset, and presents a simple method that improves overall performance.

Findings

Proposed combination of methods outperforms state-of-the-art on benchmarks.

New dataset is larger and covers longer time periods.

Evaluation framework enables comprehensive assessment of TLS approaches.

Abstract

Previous work on automatic news timeline summarization (TLS) leaves an unclear picture about how this task can generally be approached and how well it is currently solved. This is mostly due to the focus on individual subtasks, such as date selection and date summarization, and to the previous lack of appropriate evaluation metrics for the full TLS task. In this paper, we compare different TLS strategies using appropriate evaluation frameworks, and propose a simple and effective combination of methods that improves over the state-of-the-art on all tested benchmarks. For a more robust evaluation, we also present a new TLS dataset, which is larger and spans longer time periods than previous datasets.