Answers are grounded in this paper and cite their sources.
Sentiment analysis of movie reviews reflects users' behavioral preferences and emotional responses to films, aiding in understanding audience needs and enhancing user experience. In recent years, deep ensemble learning has significantly improved sentiment polarity prediction by combining multiple deep models. However, existing research often relies on limited architectural diversity and does not explicitly exploit heterogeneous encoders. The objective of this study is to design and systematically evaluate a heterogeneous deep meta‐ensemble (HDME) framework for movie‐review sentiment classification by (i) integrating structurally diverse base learners (CNN, BiGRU, BiLSTM, and BERT) and (ii) comparing multiple integration strategies, including voting, stacking, double‐stacking, and stacking‐voting, to quantify their performance gains over single models. Experiments on the IMDb dataset show that the proposed HDME variants consistently outperform individual base learners in accuracy, precision, recall, and F1‐score. Notably, the stacking‐weighted voting method achieved the best performance, reaching 96.61% accuracy, 96.33% recall, 96.89% precision, and a 96.60% F1‐score. These results highlight the potential of the HDME framework for capturing complementary textual features, which provides practical guidance for building high‐performance sentiment analysis systems.
No takes yet. Share an insight, caveat, or question.
Cai et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: