一种融合Bert预训练和BiLSTM的场景迁移情感分析研究

Research on sentiment analysis of scene migration based on Bert pre-training and BiLSTM

针对机器学习和深度学习模型较难解决场景迁移和跨领域舆情事件的情感分析问题,提出一种融合Bert预训练和BiLSTM的场景迁移情感分析模型。首先构建预训练模型Bert来提取及表征文本的词向量。再融合BiLSTM和注意力机制的情感分析模型,捕获长距离依赖关系和语义特征。最后构建Softmax实现情感分析。实验结果显示,该模型的F1值和准确率分别为0.8190和0.8181,均高于对比的机器学习和深度学习模型,这表明跨场景和平台迁移的情感分析有效。

Aiming at the difficulty of machine learning and deep learning models to solve the sentiment analysis problem of scene migration and cross-domain public opinion events,we propose a scene migration sentiment analysis model that integrates Bert pretraining and BiLSTM.Firstly,a pre-trained model Bert is constructed to extract and characterize the word vectors of text.Then a sentiment analysis model with BiLSTM and attention mechanism is fused to capture long-range dependencies and semantic features.Finally,Softmax is built to implement sentiment analysis.The experimental results show that the F1value and accuracy of the proposed model are 0.8190 and 0.8181,respectively,which are higher than those of the comparative machine learning and deep learning models,and can effectively implement the sentiment analysis task of cross-scenario and platform migration.