DisenCite: Graph-Based Disentangled Representation Learning for Context-Specific Citation Generation

Citing and describing related literature are crucial to scientific writing. Many existing approaches show encouraging performance in citation recommendation, but unable accomplish the more challenging onerous task of text generation. In this paper, we propose a novel disentangled representation based model DisenCite automatically generate through integrating paper graph. A key novelty our method compared with is context-specific text, empowering generation different types citations for same paper. particular, first build make available graph enhanced contextual dataset (GCite) 25K edges characterized by contained sections over 4.8K research papers. Based on dataset, encode each according both textual contexts structure information heterogeneous The resulted representations then mutual regularization between its neighbors Extensive experiments demonstrate superior comparing state-of-the-art approaches. We further conduct ablation case studies reassure that improvement comes from generating incorporating