Refining latent multi-view graph for relation extraction

• Main Author: Xue, Fuzhao
• Other Authors: Chng Eng Siong
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/151541
• Institution: Nanyang Technological University

Relation extraction (RE) is to predict the relation type of two entities that are mentioned in a piece of text, e.g. a sentence or a dialogue. According to various text input formats, Relation extraction can be divided into different categories, such as sentence-level RE, document-level RE and dialogue-level RE. In this thesis, we focus on addressing sentence-level RE and dialogue-level RE with a unified model. For both sentence-level RE and dialogue-level RE, the main challenge arise when the given text is long. It is difficult to identify indicative words for the relation prediction. Recent advances on RE tasks are from BERT-based models. These models predict relation by sentence-level semantic representation. Hierarchical representations are missing in BERT-based models so that the words weakly related to the target relation induce extra noise. Therefore, in this thesis, we propose a unified model, GDPNet (Gaussian Dynamic Time Warping Pooling Net), to improve relation extraction by identifying indicative words. GDPNet contains a BERT module and a Graph Module. In the BERT module, we obtain token representations from the raw text. Then, in the Graph Module, we propose to construct a latent multi-view graph to capture various possible relationships among tokens. We then refine this graph to select important words for relation prediction. Finally, the representation of the refined graph and the BERT-based sequence representation are concatenated for relation extraction. Specifically, in our proposed GDPNet, we utilize Gaussian Graph Generator (GGG) to generate edges of the multi-view graph. The graph is then refined by Dynamic Time Warping Pooling (DTWPool). On DialogRE and TACRED, we show that GDPNet achieves the best performance on dialogue-level RE, and comparable performance with the state-of-the-arts on sentence-level RE. We also perform a quantitative study to show that our model is capable of finding indicative words in an unsupervised manner.