GPLSIUA: Combining Temporal Information and Topic Modeling for Cross-Document Event Ordering

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 820–824, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics GPLSIUA: Combining Temporal Information and Topic Modeling for Cross-Document Event Ordering Borja Navarro-Colorado and Estela Saquete Natural Language Processing Group University of Alicante Alicante, Spain borja@dlsi.ua.es, stela@dlsi.ua.es Abstract Building unified timelines from a collection of written news articles requires cross-document event coreference resolution and temporal re- lation extraction. In this paper we present an approach event coreference resolution accord- ing to: a) similar temporal information, and b) similar semantic arguments. Temporal in- formation is detected using an automatic tem- poral information system (TIPSem), while se- mantic information is represented by means of LDA Topic Modeling. The evaluation of our approach shows that it obtains the highest Micro-average F-score results in the SemEval- 2015 Task 4: “TimeLine: Cross-Document Event Ordering” (25.36% for TrackB, 23.15% for SubtrackB), with an improvement of up to 6% in comparison to the other systems. How- ever, our experiment also showed some draw- backs in the Topic Modeling approach that de- grades performance of the system. 1 Introduction Since access to knowledge is crucial in any domain, connecting and time-ordering the information ex- tracted from different documents is a very important task. The goal of this paper is therefore to build or- dered timelines for a set of events related to a tar- get entity. In doing so, our approach is dealing with two problems: a) cross-document event coreference resolution and b) cross-document temporal relation extraction. In order to arrange event mentions in a timeline it is necessary to know which event mentions co-refer to the same event or fact and occur at the same mo- ment. Our approach attempts to formalize the idea that two or more event mentions co-refer if they have not only temporal compatibility (the events occur at the same time) but also semantic compatibility (the event mentions refers to the same facts, location, en- tities, etc.). Of a set of event mentions in one or more texts, our proposal groups together the event mentions that (i) have the same or a similar temporal reference, (ii) have the same or a similar event head word, and (iii) whose main arguments refer to the same or sim- ilar topics. In order to evaluate the system, we have participated in the SemEval-2015 Task 4 “TimeLine: Cross-Document Event Ordering”. In the following sections we will present the the- oretical background to our approach (section 2) and the main technical aspects (sections 3 and 4). Then we will present the results obtained (section 5) and some conclusions. 2 Background Two or more event mentions co-refer when they re- fer to the same real fact or event. Two events can denote the same fact whereas the linguistic mentions have a different syntax structure, different words, or even a different meaning. Whatever the case may be, both event mentions must be semantically related. An event mention is formed of an event head (usually a verb or …

Content truncated to 500 words