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

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 stateof-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. The dataset will be made available at https://github. com/complementizer/news-tls.

Explicitly Modeling Importance and Coherence for Timeline Summarization

“…Implementation Details. We follow the experimental settings [7,13,8,10] by asking the TLS framework to summarize a single topic across multiple timelines (where the ground-truth timeline varies). The timelines are produced with l dates and k sentences per date, in which l and k are same as the ground-truth.…”

“…Oracle Summaries. Previous studies have approximated the upbound performance (Oracle) of TLS [7,18,13]. Oracle results are obtained by greedily selecting all input documents' sentences.…”

“…Based on this date selection algorithm, our derivational method is denoted as TLSUBX ‡ . We also follow the works of DATEWISE [13] to select the important dates by SUPERVISED [3]. TLSUBX † uses this date selection algorithm.…”

“…Non-submodularity approaches. CLUST [13] is a textual clusterbased method which count date mentions to rank event clusters, and then generate daily summaries in each cluster. To cluster articles, they use Markov Clustering (MCL).…”

TopicRefine: Joint Topic Prediction and Dialogue Response Generation for Multi-turn End-to-End Dialogue System

“…Natural Language Generation (NLG), is the task of generating language that is coherent and understandable to humans, has been applied to many downstream tasks such as text summary (Zhang et al, 2019a;Bar-Haim et al, 2020;Cho et al, 2020;Huang et al, 2020;Gholipour Ghalandari and Ifrim, 2020), machine translation (Li et al, 2020;Baziotis, Haddow, and Birch, 2020;Cheng et al, 2020;Zou et al, 2020), and dialogue response generation (Radford et al, 2019;Zhou et al, 2018b;Tuan, Chen, and Lee, 2019;Zhao et al, 2020;Liu et al, 2020a;Wolf et al, 2019).…”