2019/11/05 by Amir Pouran Ben Veyseh, Veyseh, Amir Pouran Ben, Franck Dernoncourt +5 · 1 citation
Computer Science · #Advanced Text Analysis Techniques #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1911.01678
openalex publication_date 2019/11/05 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Definition Extraction (DE) is one of the well-known topics in Information\nExtraction that aims to identify terms and their corresponding definitions in\nunstructured texts. This task can be formalized either as a sentence\nclassification task (i.e., containing term-definition pairs or not) or a\nsequential labeling task (i.e., identifying the boundaries of the terms and\ndefinitions). The previous works for DE have only focused on one of the two\napproaches, failing to model the inter-dependencies between the two tasks. In\nthis work, we propose a novel model for DE that simultaneously performs the two\ntasks in a single framework to benefit from their inter-dependencies. Our model\nfeatures deep learning architectures to exploit the global structures of the\ninput sentences as well as the semantic consistencies between the terms and the\ndefinitions, thereby improving the quality of the representation vectors for\nDE. Besides the joint inference between sentence classification and sequential\nlabeling, the proposed model is fundamentally different from the prior work for\nDE in that the prior work has only employed the local structures of the input\nsentences (i.e., word-to-word relations), and not yet considered the semantic\nconsistencies between terms and definitions. In order to implement these novel\nideas, our model presents a multi-task learning framework that employs graph\nconvolutional neural networks and predicts the dependency paths between the\nterms and the definitions. We also seek to enforce the consistency between the\nrepresentations of the terms and definitions both globally (i.e., increasing\nsemantic consistency between the representations of the entire sentences and\nthe terms/definitions) and locally (i.e., promoting the similarity between the\nrepresentations of the terms and the definitions).\n