Efficient Character-level Document Classification by Combining Convolution and Recurrent Layers

TL;DR

This paper introduces a neural network combining convolutional and recurrent layers for character-level document classification, achieving similar accuracy to convolution-only models but with fewer parameters.

Contribution

The novel architecture efficiently encodes character inputs by integrating convolution and recurrent layers, reducing model complexity while maintaining performance.

Findings

Achieves comparable accuracy to convolution-only models

Uses significantly fewer parameters

Validated on eight large-scale datasets

Abstract

Document classification tasks were primarily tackled at word level. Recent research that works with character-level inputs shows several benefits over word-level approaches such as natural incorporation of morphemes and better handling of rare words. We propose a neural network architecture that utilizes both convolution and recurrent layers to efficiently encode character inputs. We validate the proposed model on eight large scale document classification tasks and compare with character-level convolution-only models. It achieves comparable performances with much less parameters.