Sentiment Classification By Leveraging The Shared Knowledge From A Sequence Of Domains

This paper studies sentiment classification in a setting where a sequence of classification tasks is performed over time. The goal is to leverage the knowledge gained from previous tasks to do better on the new task than without using the previous knowledge. This is a lifelong learning setting. This paper proposes a novel deep learning model for lifelong sentiment classification. The key novelty of the proposed model is that it uses two networks: a knowledge retention network for retaining domain-specific knowledge learned in the past, and a feature learning network for classification feature learning. The two networks work together to perform the classification task. Our experimental results show that the proposed deep learning model outperforms the state-of-the-art baselines.