Text Sentiment Classification Model using Attention Mechanism and Word Vector Sentiment Computation

Accurately extracting sentiment and semantic information from long-text online reviews has significant application value. To address sentiment analysis in long-text online reviews, this paper proposes a novel sentiment analysis algorithm based on a deep feature and weighted Word2Vec fusion model, ATNPWC_BiLSTM. First, sentiment-weighted Word2Vec word vectors are used to obtain sentiment features and short-range semantic features (PWC). Simultaneously, an attention-based bidirectional Long Short-Term Memory (BiLSTM) network is employed to extract deep features and long-range semantic dependencies (ATN) from the review corpus. Then, PWC features and ATN features are fused to generate the ATNPWC feature representation. Finally, a classification model is trained using the ATNPWC features to categorize sentiment in review texts. Experimental results demonstrate that the proposed algorithm effectively captures and integrates both short- and long-range contextual semantic information, deep textual features, and sentiment characteristics, thereby improving sentiment analysis performance for long-text reviews.