Paper Contents
Abstract
Sentiment analysis across various domains has become a critical task inunderstanding customer feedback, social media discussions, and productreviews. However, one of the major challenges in sentiment analysis is theability to transfer sentiment knowledge effectively across differentdomains (e.g., from movie reviews to product reviews). Traditional wordembedding, such as Word2Vec and Glove, have shown promising resultsbut often fail to capture domain-specific sentiment nuances. This projectproposes a novel approach for cross-domain sentiment encoding usingstochastic word embedding. By incorporating stochastic elements intoword representation learning, the method generates domain-agnosticsentimentembeddingwhilepreservingtheuniquesentimentcharacteristics of each domain.In this study, we introduce a modified word embedding technique thatintegrates stochastic processes with sentiment analysis models. Ourapproach is based on generating probabilistic representations of words,allowing for better generalization across domains while retaining specificsentiment features relevant to each. The stochastic embedding are thenevaluated across multiple sentiment classification tasks from distinctdomains, demonstrating improvements in both accuracy and robustnesswhen compared to traditional embedding.The results show that our method effectively addresses the domain shiftproblem, offering a promising solution for applications like cross-domainsentiment analysis and opinion mining in varied contexts. Additionally, weexplore the potential of stochastic embedding in enhancing performancein low-resource domains where labeled data is limited.
Copyright
Copyright © 2025 A.Nandini. This is an open access article distributed under the Creative Commons Attribution License.
