Event Time Extraction with a Decision Tree of Neural Classifiers

Event Time Extraction with a Decision Tree of Neural Classifiers Nils Reimers†, Nazanin Dehghani‡∗, Iryna Gurevych† † Ubiquitous Knowledge Processing Lab (UKP) and Research Training Group AIPHES Department of Computer Science, Technische Universit¨at Darmstadt ‡ School of Electrical and Computer Engineering, University of Tehran www.ukp.tu-darmstadt.de Abstract Extracting the information from text when an event happened is challenging. Documents do not only report on current events, but also on past events as well as on future events. Often, the relevant time information for an event is scattered across the document. In this paper we present a novel method to auto- matically anchor events in time. To our knowl- edge it is the first approach that takes tempo- ral information from the complete document into account. We created a decision tree that applies neural network based classifiers at its nodes. We use this tree to incrementally infer, in a stepwise manner, at which time frame an event happened. We evaluate the approach on the TimeBank-EventTime Corpus (Reimers et al., 2016) achieving an accuracy of 42.0% com- pared to an inter-annotator agreement (IAA) of 56.7%. For events that span over a single day we observe an accuracy improvement of 33.1 points compared to the state-of-the-art CAEVO system (Chambers et al., 2014). Without re- training, we apply this model to the SemEval- 2015 Task 4 on automatic timeline generation and achieve an improvement of 4.01 points F1-score compared to the state-of-the-art. Our code is publically available.1 1 Introduction Knowing when an event happened is useful for a lot of use cases. Examples are in the fields of time-aware information retrieval, text summarization, automated timeline generation, and automatic knowledge base population. Many facts in a knowledge base are ∗During author’s internship in the research training group AIPHES at UKP Lab, TU Darmstadt. 1https://github.com/ukplab/ tacl2017-event-time-extraction only true for a certain time period, for example the presidency of a person. Hence, the population of a knowledge base can highly benefit from high quality event and event time2 extraction (Surdeanu, 2013). Inherent to events is the connection to time. Allan (2002) defines an event as “something that happens at some specific time and place”. The challenges for automatic event time extraction are manifold. The temporal information in news articles which states when an event happened is, in most cases, not in the same or in neighboring sentences with the event (Reimers et al., 2016). It can be mentioned far before the event or far after the event. Even worse, for more than 60% of events, the specific day at which the event happened is not mentioned. However, from the world knowledge and causal relations, the reader can infer a lot of temporal information about those events and can often infer that the event happened before or after some specific point in time. In this paper we describe a new classifier for auto- matic event time extraction. We use the TimeBank- EventTime Corpus (Reimers et al., 2016) to train and evaluate our proposed architecture. In contrast to other corpora …

Content truncated to 500 words