【摘要】 针对传统的方面级情感分析模型不能很好表征深层次的字词向量信息，且无法解决该领域由于人工标注的高成本方法，使得数据集普遍较少而导致的分类效果较差的问题，提出BERTDTL-HAN方面级情感分析模型。模型通过BERT结构在获得含有丰富语义信息字词向量信息的同时，结合深层次迁移学习和层次注意网络机制将数据量大的句子级别情感分析数据集，通过单词编码层和片段编码层两个维度深层迁移到数据量小的方面级情感分析任务中，并在三个领域的数据集上进行实验。对比该领域内的三个最佳基准模型，其准确率分别提升1.40%、0.96%和0.93%。更多还原

【Abstract】 The traditional aspect-level sentiment analysis model can not well represent the deep-level word vector information,and can not solve the problem of poor classification effect due to the high cost of manual annotation method,which makes the data set generally less. Aiming at these two problems,this paper proposes a BERT-DTL-HAN aspect emotion analysis model. Model by BERT structure contains rich semantic information of words vector information at the same time,combined with deep migration mechanism and hierarchical network learning a large quantity of data to sentence level sentiment analysis data set by encoding layer and section code words deep migration to the small amount of data of two dimensions emotional level analysis task,and in three areas of experimental data set.Compared with the three best benchmark model in this field,the accuracy of the model is improved by 1.40%,0.96% and0.93% respectively.更多还原