基于Word2vec与注意力机制的情感分析研究  

Research on Sentiment Analysis Based on Word2vec and Attention Mechanisml

作　　者：任伟建[1,2] 徐海杰 康朝海 霍凤财[1,2] 任璐 张永丰[4] REN Weijian;XU Haijie;KANG Chaohai;HUO Fengcai;REN Lu;ZHANG Yongfeng(School of Electrical Information Engineering,Northeast Petroleum University,Daqing 163318;Heilongjiang Provincial Key Laboratory of Networking and Intelligent Control,Daqing 163318;Offshore Oil Engineering Company Limited,Tianjin 300450;Planning and Design of No.2 Oil Production Plant,Daqing Oilfield Co.,Ltd.,Daqing 163318)

机构地区：[1]东北石油大学电气信息工程学院,大庆163318 [2]黑龙江省网络化与智能控制重点实验室,大庆163318 [3]海洋石油工程股份有限公司,天津300450 [4]大庆油田有限责任公司第二采油厂规划设计研究所,大庆163318

出　　处：《计算机与数字工程》2024年第10期2991-2995,3147,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61933007,61873058)资助。

摘　　要：针对传统情感分析模型对关键词特征抓取不准确、局部情感特征提取不全面造成分类效果差的问题,提出一种基于TW-BiLSTM-ATT情感分析模型。通过对TF-IDF改进,并与Word2vec结合,使权重特征融入词向量提升对关键信息的抓取能力;将词向量的位置特征融入到注意力机制中,使模型可以关注到目标词汇附近的词,进而更加全面地将情感特征提取出来。对比实验结果表明TW-BiLSTM-ATT模型在处理情感分析任务中分类效果好于同类模型。In allusion to the problems of inaccurate capture of keyword features and incomplete extraction of local sentiment features by traditional sentiment analysis models,resulting in poor classification results,a sentiment analysis model based on TW-BiLSTM-ATT is proposed.Through the improvement of TF-IDF and the combination with Word2vec,the weight feature is inte-grated into the word vector to improve the ability to capture key information.The position feature of the word vector is integrated into the attention mechanism,the model can focus on the words around the target vocabulary,and then extract the emotional features more comprehensively.The comparative experimental results show that the TW-BiLSTM-ATT model has better classification perfor-mance than similar models in processing sentiment analysis tasks.

关 键 词：Word2vec TF-IDF BiLSTM ATTENTION 情感分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]