Personality-driven experience storage and retrieval for sentiment classification

The existing methods for sentiment classification normally ignore that the past experiences retrieved by users under particular situations would affect their sentiment expressions. Furthermore, related research may underutilize user personality to personalize the analysis of storage and retrieval of past experiences. Inspired by the cognition process of human memory, we propose a Personality-Driven Experience Storage and Retrieval (PDESR) model for sentiment classification. Concretely, we first selectively store the user’s past experiences in her/his experience bank via personalized input and forget gates. We then adopt personalized output gate to retrieve past experiences from the experience bank. Finally, we integrate the current experience with the retrieved past experiences to classify user sentiment. Specifically, personality is used to personalize the control of which past experiences should be stored in experience bank and which past experiences should be retrieved from experience bank. The experimental results show that PDESR model outperforms the related models in accuracy.