多粒度特征融合的分层式机器学习情感分析

打开文本图片集

中图分类号： TP391 文献标志码：A 文章编号： 1000-5013（2026）02-0164-11

Hierarchical Machine Learning Sentiment Analysis of Multi-Granularity Feature Fusion

ZHAO Jinxin，GUO Rongxin，SHI Yifan （College of Engineering，Huaqiao University，Quanzhou 362o21，China）

Abstract：To address the insufficient utilization of semantic features in esisting sentiment analysis methods for multi-category text classification tasks，a hierarchical sentiment analysis model based on multi-granularity feature fusion is proposed. First，the extreme gradient boosting（XGBoost） algorithm and support vector machine （SVM） are employed in parallel for basic classification，each generating probability distributions across 10 categories.Then，logistic regresion is adopted as a meta-classifier to perform feature-level fusion of the dual channel output results. Finally，the model is validated on a public dataset containing 62 774 comments across 10 categories. Experimental results show that the HML-MGFF model achieves an average accuracy improvement of 15.6% over traditional single-classifier models，and 4.6% over four other composite models.

Keywords：multi-granularity feature fusion； hierarchical machine learning；limit gradient lifting； support vec-tor machine; logistic regression

情感分析属于自然语言处理中备受关注的研究方向，用于识别和评估文本中的情绪和情感倾向，通常将文本划分为正面、负面或中性情感。（剩余17451字）