Event-Based Time Label Propagation for Automatic Dating of News Articles

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1–11, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Event-based Time Label Propagation for Automatic Dating of News Articles Tao Ge Baobao Chang∗ Sujian Li Zhifang Sui Key Laboratory of Computational Linguistics, Ministry of Education School of Electronics Engineering and Computer Science, Peking University No.5 Yiheyuan Road, Haidian District, Beijing, P.R.China, 100871 {getao,chbb,lisujian,szf}@pku.edu.cn Abstract Since many applications such as timeline sum- maries and temporal IR involving temporal analysis rely on document timestamps, the task of automatic dating of documents has been increasingly important. Instead of using feature-based methods as conventional mod- els, our method attempts to date documents in a year level by exploiting relative tempo- ral relations between documents and events, which are very effective for dating documents. Based on this intuition, we proposed an event- based time label propagation model called confidence boosting in which time label in- formation can be propagated between docu- ments and events on a bipartite graph. The ex- periments show that our event-based propaga- tion model can predict document timestamps in high accuracy and the model combined with a MaxEnt classifier outperforms the state-of- the-art method for this task especially when the size of the training set is small. 1 Introduction Time is an important dimension of any informa- tion space and can be useful in information re- trieval, question-answering systems and timeline summaries. In the applications involving tempo- ral analysis, document timestamps are very useful. For instance, temporal information retrieval mod- els take into consideration the document’s creation time for document retrieval and ranking (Kalczyn- ski and Chou, 2005; Berberich et al., 2007) for bet- ter dealing with time-sensitive queries; some infor- ∗Corresponding author mation retrieval applications such as Google Scholar can list articles published during the time a user specifies for better satisfying users’ needs. In addi- tion, timeline summarization techniques (Hu et al., 2011; Binh Tran et al., 2013) and some event-event ordering models (Chambers and Jurafsky, 2008; Yoshikawa et al., 2009) also rely on the timestamps. Unfortunately, many documents on the web do not have a credible timestamp, as Chambers (2012) re- ported. Therefore, it is significant to date docu- ments, that is to predict document creation time. One typical method for dating document is based on temporal language models, which were first used for dating by de Jong et al. (2005). They learned language models (unigram) for specific time periods and scored articles with normalized log-likelihood ratio scores. The other typical approach for the task was proposed by Nathanael Chambers (2012). In Chambers’s work, discriminative classifiers – max- imum entropy (MaxEnt) classifiers were used by incorporating linguistic features and temporal con- straints for training, which outperforms the previous temporal language models on a subset of Gigaword Corpus (Graff et al., 2003). However, the conventional methods have some limitations because they predict creation time of documents mainly based on feature-based models without understanding content of documents, which may lead to wrong predictions in …

Content truncated to 500 words