Abstract:Text classification has been attracting increasing attention with the growth of textual data created on the Internet. Great progress has been made by deep neural networks for domains where a large amount of labeled training data is available. However, providing sufficient data is time-consuming and labor-intensive, establishing substantial obstacles for expanding the learned models to new domains or new tasks. In this paper, we investigate the transferring capability of capsule networks for text classification. Capsule networks are able to capture the intrinsic spatial part-whole relationship constituting domain invariant knowledge that bridges the knowledge gap between the source and target domains (or tasks). We propose an iterative adaptation strategy for cross-domain text classification, which adapts the source domain to the target domain. A fast training method with capsule compression and class-guided routing is designed to make the capsule network more efficient in computation for cross-domain text classification. We first conduct experiments to evaluate the performance of the capsule network on six benchmark datasets for generic text classification. The capsule networks outperform the compared models on 4 out of 6 datasets, suggesting the effectiveness of the capsule networks for text classification. More importantly, we demonstrate the transferring capability of the proposed cross-domain capsule network (TL-Capsule) by applying it to two transfer learning applications: single-label to multi-label text classification and cross-domain sentiment classification. The experimental results show that capsule networks consistently and substantially outperform the compared methods for both tasks. To the best of our knowledge, this is the first work that empirically investigates the transferring capability of capsule networks for text modeling.

Transfer Learning via Cluster Correspondence Inference

Softly Associative Transfer Learning for Cross-Domain Classification

Joint Domain Alignment and Discriminative Feature Learning for Unsupervised Deep Domain Adaptation

Topic Correlation Analysis for Cross-Domain Text Classification

Transfer Domain Class Clustering for Unsupervised Domain Adaptation

Learning the Shared Subspace for Multi-task Clustering and Transductive Transfer Classification.

Co-clustering Based Classification for Out-of-domain Documents

A deep targeted transfer network with clustering pseudo-label learning for fault diagnosis across different Machines

Multi-task clustering via domain adaptation

Unsupervised domain adaptation based on cluster matching and Fisher criterion for image classification

A Collaborative Transfer Learning Framework for Cross-domain Recommendation

Bridging Domains Using World Wide Knowledge for Transfer Learning

Transferring Topical Knowledge from Auxiliary Long Texts for Short Text Clustering

Investigating the Transferring Capability of Capsule Networks for Text Classification

Domain Consensus Clustering for Universal Domain Adaptation

Cross-Domain Collaborative Learning Via Discriminative Nonparametric Bayesian Model

Transfer Learning for Optical and SAR Data Correspondence Identification With Limited Training Labels

Semantic Consistency Knowledge Transfer for Unsupervised Cross Domain Object Detection

Transfer Learning Via Feature Isomorphism Discovery

Discriminative Feature Alignment: Improving Transferability of Unsupervised Domain Adaptation by Gaussian-guided Latent Alignment

CLIP-CID: Efficient CLIP Distillation via Cluster-Instance Discrimination