【摘要】 为提高中文文本情感分析任务的准确率,优化训练时长,提出基于Reformer的文本情感分析模型。利用Reformer模型的上下文语义编码能力,充分获得文本上下文的特征,提高文本分类准确率;在Transformer模型的基础上,引入局部敏感哈希注意力机制及可逆残差,降低模型的复杂度及内存的占用。在3个公开数据集上进行实验,实验结果表明,该模型在准确率及训练时间上均优于其它模型。更多还原

【Abstract】 To improve the accuracy of Chinese text sentiment analysis tasks and to optimize the training time,a text sentiment analysis model based on Reformer was proposed.Based on Transformer model,the context semantic coding capability of the Reformer model was utilized,the characteristics of the text context were fully obtained,the accuracy of text classification was improved,and at the same time,locally sensitive hash attention mechanism and reversible residue were introduced,which reduced the complexity and memory utilization of the model.Experiments were performed on three publicly available data sets.Experimental results show that the proposed model is superior to other models in accuracy rate and training time.更多还原