Ranking Multidocument Event Descriptions for Building Thematic Timelines

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1208–1217, Dublin, Ireland, August 23-29 2014. Ranking Multidocument Event Descriptions for Building Thematic Timelines Kiem-Hieu Nguyen†, Xavier Tannier†$, and Veronique Moriceau†$ †LIMSI-CNRS $Univ. Paris-Sud Orsay, France {nguyen,xtannier,moriceau}@limsi.fr Abstract This paper tackles the problem of timeline generation from traditional news sources. Our sys- tem builds thematic timelines for a general-domain topic defined by a user query. The system selects and ranks events relevant to the input query. Each event is represented by a one-sentence description in the output timeline. We present an inter-cluster ranking algorithm that takes events from multiple clusters as input and that selects the most salient and relevant events. A cluster, in our work, contains all the events happening in a specific date. Our algorithm utilizes the temporal information derived from a large collection of extensively temporal analyzed texts. Such temporal information is combined with textual contents into an event scoring model in order to rank events based on their salience and query-relevance. 1 Introduction We aim at building thematic timelines from multiple documents relevant to a specific, user-generated query. For instance, for the query “Libya conflict”, our system will return important events related to the Libya conflict in 2011 involving Kadhafi forces, rebels, NATO intervention, etc. (Figure 1). Such a timeline can then be visualized as a textual, event-based summary, or through any existing graphical timeline visualization tool. The main contribution of this paper is a two-step inter-cluster ranking algorithm aimed at selecting salient and non-redundant events from temporal clusters, which are sets of sentences describing events related to the query and that occurred at the same day. In the first step, a scoring model is proposed to rank sentences describing events, according to their relevance and salience to the topic. In the second step, the ranked events are iteratively reranked based on their content in order to reduce information redundancy. We finally obtain an extendable, chronological summary of important events concerning the query. This paper is organized as follows: §2 introduces related work. §3 presents the resources used and gives an overview of the system. The salient date algorithm proposed by Kessler et al. (2012), that we used to build our temporal clusters, is briefly summarized in §4. §5 and §6 describe our ranking approach to event selection and a content-based reranking algorithm, respectively. The evaluations are presented in §7. §8 is dedicated to the conclusion and future work. 2 Related Work Our work is closely related to event detection and tracking (EDT) and multidocument summarization (MDS). This section introduces some important work in these fields. 2.1 Event Detection and Tracking EDT on news streams has been intensively studied. Early work concentrates on detecting events from article texts using vector-based techniques (Allan et al., 1998; Petrovi´c et al., 2010) or graphical models This work is licenced under a Creative Commons Attribution 4.0 International License. Page numbers and proceedings footer are added by the organizers. License details: http://creativecommons.org/licenses/by/4.0/ 1208 Mar 19 2011. The …

Content truncated to 500 words