Learning to Search in Long Documents Using Document Structure
This addresses the problem of efficient question answering over long documents for NLP applications, representing an incremental improvement by combining tree-based navigation with existing methods.
The paper tackles the bottleneck of sequential processing in reading comprehension models for long documents by proposing a framework that models documents as trees and uses an agent to navigate and extract answers, resulting in improved question answering performance over DQN and IR baselines, with further gains from ensembling.
Reading comprehension models are based on recurrent neural networks that sequentially process the document tokens. As interest turns to answering more complex questions over longer documents, sequential reading of large portions of text becomes a substantial bottleneck. Inspired by how humans use document structure, we propose a novel framework for reading comprehension. We represent documents as trees, and model an agent that learns to interleave quick navigation through the document tree with more expensive answer extraction. To encourage exploration of the document tree, we propose a new algorithm, based on Deep Q-Network (DQN), which strategically samples tree nodes at training time. Empirically we find our algorithm improves question answering performance compared to DQN and a strong information-retrieval (IR) baseline, and that ensembling our model with the IR baseline results in further gains in performance.