Predicting Chinese Phrase-Level Sentiment Intensity in Valence-Arousal Dimensions With Linguistic Dependency Features

Phrase-level sentiment intensity prediction is difficult due to the inclusion of linguistic modifiers (e.g., negators, degree adverbs, and modals) potentially resulting in an shift or polarity reversal for modified words. This study develops a graph-based Chinese parser based on deep biaffine attention model obtain dependency structures relations. These obtained features are then used our proposed Weighted-sum Tree GRU network predict phrase-level valence-arousal dimensions. Dependency parsing results using Sinica Treebank indicate that outperforms transition-based methods such as MLP stack-LSTM with identical findings English parsing. Experimental EmoBank other transformer-based neural networks BERT, ALBERT, XLNET ELECTRA, reflecting effectiveness dependencies predication tasks. In addition, requires fewer parameters less inference time quantitative analysis, making relatively lightweight efficient.