Cross-domain Sentiment Encoding Through Stochastic Word Embedding

Cross-domain Sentiment Encoding through Stochastic Word Embedding

Abstract:

Sentiment analysis is an important topic concerning identification of feelings, attitudes, emotions and opinions from text. A critical challenge for automating such analysis is the high manual annotation cost when conducting large-scale learning. However, the cross-domain technique is a key solution for this. It reuses annotated reviews across domains and its success principally relies on the effort that has been invested to improve the cross-domain representation learning by designing increasingly more complex and elaborate model inputs and architectures. We support that it is not necessary to focus on design complexity as this inevitably consumes more time for model training. Instead, we propose to explore through a simple mapping the word polarity and occurrence information and encode such information more accurately whilst aiming at lower computational costs. The proposed approach is unique and takes advantage of the stochastic embedding technique to tackle cross-domain sentiment alignment. Its effectiveness is benchmarked with over ten data tasks constructed from two review corpora, and is compared against ten classical and state-of-the-art methods.