A Fine-Grained Sentiment Analysis Method Using Transformer for Weibo Comment Text

A Fine-Grained Sentiment Analysis Method Using Transformer for Weibo Comment Text

Piao Xue (School of Marxism Studies, Xi'an Mingde Institute of Technology, China) and Wei Bai (School of Mathematics and Computer Science, Ningxia Normal University, China)
DOI: 10.4018/IJITSA.345397
Article PDF Download
Open access articles are freely available for download

Abstract

Many existing fine-grained sentiment analysis (FGSA) methods have problems such as easy loss of fine-grained information, difficulty in solving polysemy and imbalanced sample categories. Therefore, a Transformer based FGSA method for Weibo comment text is proposed. Firstly, the RoBERTa model with knowledge augmentation was used to dynamically encode the text so as to solving the polysemy issue. Then, BiLSTM is used to effectively capture bidirectional global semantic dependency features. Next, Transformer is used to fuse multi-dimensional features and adaptively strengthen key features to overcome the problem of fine-grained information loss. Finally, an improved Focal Loss function is utilized for training to solve the issue of imbalanced sample categories. As demonstrated by the experimental outcomes on the SMP2020-EWECT, NLPCC 2013 Task 2, NLPCC 2014 Task 1, and weibo_senti_100k datasets, the suggested method outperforms the alternatives for advanced comparison methods.
Article Preview
Top

Introduction

The digital age has democratized access to information, requiring the business and academic sectors to develop new methods to analyze and process this information. In addition, the tools and software for processing information are constantly being improved and updated (Gandhi et al., 2023; Li et al., 2022; Liu et al., 2022; Zeng et al., 2023). Structured data analysis alone is not enough to meet people’s needs. Unstructured text data, like public opinion expressed through social media and surveys, is gaining importance as a valuable information source. For public stability, government departments should actively monitor public opinion, understand citizen sentiment, and keep them informed through transparent information flow (Fan, 2023). Public opinion control is not only a matter for relevant staff but also one that concerns the vital interests of all citizens (Djaballah et al., 2021; Li et al., 2021; Tamil et al., 2022; Zhang & Cui, 2023; Zhang et al., 2020).

It is common to use English texts as research objects in unstructured data analysis (Xiao et al., 2022; Yu et al., 2021; Zhang et al., 2022; Y. Zhang et al., 2023). However, when dealing with Chinese unstructured data, many problems remain, such as difficulty in sentence segmentation due to the absence of spaces between words, distinguishing words or phrases that may have different meanings in different contexts, and determining the impact of adverbs, conjunctions, and negations on sentence meaning (Hazarika et al., 2020; Kaur & Kautish, 2022; Li et al., 2020; Zhang et al., 2022).

New media platforms like Weibo have become essential for capturing and shaping public opinion. These platforms provide a breeding ground for online discussions, where public sentiment surrounding current events can simmer, spread, and erupt (Liao et al., 2022; Lin et al., 2020; Liu, 2023; Poria et al., 2023). Netizens comment on events in the news through Weibo and express, disseminate, and interact with their emotions, thereby forming public opinion on these events (Fang et al., 2023; Yu et al., 2021; Zhu et al., 2022). However, information asymmetry also fuels the spread of online rumors, which can significantly impact public sentiment. Such effects typically last for an extended period, are difficult to control, and may bring about a certain amount of social instability. Therefore, it is necessary to conduct an effective sentiment analysis for Weibo comments (Jiang et al., 2023; Li et al., 2019; Lu et al., 2021). However, traditional sentiment analysis methods usually classify Weibo comment texts into three categories: positive, negative, or neutral. These coarse-grained sentiment analysis methods cannot capture the rich emotional information in Weibo texts. They can lead to confusion. Fine-grained sentiment analysis (FGSA) goes beyond simply classifying emotions as positive or negative. Depending on the requirements, it can identify more nuanced emotions, such as pleasure, anger, hatred, or sadness. This allows for a more fine-grained understanding of the emotional tendencies of Weibo texts (Huang et al., 2019).

Nonetheless, current FGSA methods are plagued by significant flaws, including the frequent loss of fine-grained data, the resolution of polysemy issues, and unbalanced sample categories. Therefore, this paper proposes a novel Transformer-based FGSA method for Weibo comment text. Its primary innovation lies in the following:

Complete Article List

Search this Journal:
Reset
Volume 18: 1 Issue (2025)
Volume 17: 1 Issue (2024)
Volume 16: 3 Issues (2023)
Volume 15: 3 Issues (2022)
Volume 14: 2 Issues (2021)
Volume 13: 2 Issues (2020)
Volume 12: 2 Issues (2019)
Volume 11: 2 Issues (2018)
Volume 10: 2 Issues (2017)
Volume 9: 2 Issues (2016)
Volume 8: 2 Issues (2015)
Volume 7: 2 Issues (2014)
Volume 6: 2 Issues (2013)
Volume 5: 2 Issues (2012)
Volume 4: 2 Issues (2011)
Volume 3: 2 Issues (2010)
Volume 2: 2 Issues (2009)
Volume 1: 2 Issues (2008)
View Complete Journal Contents Listing