融合词典的BERT-BiGRU的藏语句子情感分类方法  

Tibetan sentence emotion classification method based on BERT-BiGRU of fusion dictionary

作　　者：公确多杰 索南才让[1,2,3] 才藏太[1,2,3] Gong Que Duojie;Suo Nan CaiRang;Cai ZangTai(School of Computer Science and Technology,Qinghai Normal University,Xining 810008,China;The State Key Laboratory of Tibetan Intelligent Information Processing and Application,Qinghai Normal University,Xining 810008,China;Key Laboratory of Tibetan Information Processing of Ministry of Education,Qinghai Normal University,Xining 810008,China)

机构地区：[1]青海师范大学计算机学院,青海西宁810008 [2]青海师范大学藏语智能信息处理及应用国家重点实验室,青海西宁810008 [3]青海师范大学藏文信息处理教育部重点实验室,青海西宁810008

出　　处：《计算机工程与设计》2025年第3期918-926,共9页Computer Engineering and Design

基　　金：国家自然科学基金项目(62266037);国家社科基金项目(23BYY078);教育部社科基金项目(2021XJA740001);教育部人文社科基金项目(21XJA740001)。

摘　　要：针对藏文情感分析研究中,由于藏文缺乏公开的情感词典和能结合上下文的深度学习模型等导致的分类效果欠佳问题,构建一个规模可观的藏文情感词典,结合基于改进的BERT-BiGRU的藏语句子情感分类方法,在哈工大开源的CINO多语言的通用预训练语言模型上进行微调。实验结果表明,本文方法能够有效提高情感分类的准确率,验证了方法的有效性。模型在分类准确率和泛化能力上的表现优于相关工作,为进一步研究藏文文本情感分类问题提供一种思路和实验证据。A substantial Tibetan sentiment lexicon was constructed to address the issue of poor classification performance in Tibetan sentiment analysis research due to the lack of publicly available sentiment dictionaries and contextual deep learning models.This lexicon was combined with an improved BERT-BiGRU-based Tibetan sentence sentiment classification method,and fine-tuned on the CINO multilingual pretrained language model provided by Harbin Institute of Technology.Experimental results demonstrate that the proposed text sentiment classification method significantly enhances the accuracy of sentiment classification,thereby validating its effectiveness.The proposed model exhibits superior performance compared to existing methodologies,surpassing them in both the classification precision and the capacity for generalization,offering insights and empirical evidences for further research on Tibetan text sentiment classification.

关 键 词：情感分类 情感词典 情感词汇本体库 BERT-BiGRU CINO 藏语句子 深度学习 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]