JCSE, vol. 18, no. 2, pp.80-92, June, 2024

DOI: http://dx.doi.org/10.5626/JCSE.2024.18.2.80

Sentiment Classification Model of Fresh Agricultural Product Comments based on Semantic Structure-Combined Dictionary Optimization

Yindong Dong, Shundong Shao, Guohua Fan, Deli Chen, Yu Zhang, and Xin Luo
School of Information and Artificial Intelligence, Anhui Agricultural University, Hefei, China

Abstract: The increasing popularity of various types of online services has led to a growing number of consumers purchasing fresh agricultural products through online platforms. To address the discrepancy between the actual emotional disposition expressed in fresh produce reviews and the sentiment labels that are assigned to them, the current paper proposes a sentiment classification model based on the optimization and integration of a composite semantic structure dictionary. First, the TextRank algorithm is used to extract sentiment-bearing keywords from fresh produce reviews, which are used in the formation of a sentiment dictionary. Sentiment category matching scores are then calculated based on this dictionary. Concurrently, the text vectors processed by bidirectional encoder representations from transformers (BERT) are fed into an enhanced deep pyramid convolutional neural network (DPCNN) and a bidirectional long short-term memory network (BiLSTM), which are tasked with extracting global semantic information and contextual information, respectively. Subsequently, a weighted attention mechanism is used to consolidate these feature representations. Finally, a sentiment classification model, which is designated as Scores-BERT-IDPCNN-BiLSTM-Attention (S-BIDBA), is trained by incorporating sentiment category matching scores. The experimental results obtained herein demonstrate that S-BIDBA significantly enhances the accuracy of sentiment classification for online reviews of fresh agricultural products.

Keyword: Fresh produce reviews; Sentiment classification; Keyword extraction; Improved DPCNN

Full Paper:   132 Downloads, 491 View