Interpretable Event Detection and Extraction using Multi-Aspect Attention

Classical encoding and extraction methods rely on fixed dictionary of keywords and templates or require ground truth labels for phrase/sentences. This prevents wide spread use of these information encoding approaches to large-scale free form (unstructured) text available on the web. Event encoding can be categorized as a hierarchical task where the coarser level task is event detection – identification of documents containing a specific event and the fine-grained task of event encoding – identifying key phrases, key sentences. Hierarchical models with attention seem like a natural choice for this problem, given their ability to differentially attend to more or less important features when constructing the document representation. In this work we present a novel factorized bilinear multi-aspect attention mechanism (FBMA) that attends to different aspects of text while constructing its representation. We use this mechanism within a hierarchical framework to extract event references or extents i.e. event-related sentences and trigger words from the text. We find that our approach outperforms state-of-the-art baselines for event detection on the Civil Unrest and Military Action and Non-State Actor datasets in two different languages. We further provide qualitative examples of the extracted event extents and trigger words giving a peek into what our model learns. ACM Reference Format: Sneha Mehta, Mohammad Raihanul Islam, Huzefa Rangwala, and Naren Ramakrishnan. 2020. Interpretable Event Detection and Extraction using Multi-Aspect Attention. In Proceedings of . ACM, New York, NY, USA, 11 pages. https://doi.org/10.1145/nnnnnnn.nnnnnnn