基于迁移学习的电力短文本情感分类研究

Research about Sentiment Classsification of Short Text in the Electric Power Industry by Transfer Learning

为了有效提高客服效率与主动服务意识,从电力短文本中挖掘客户的情感状态,提出了一种基于迁移学习的情感分析方法,将具有丰富标注信息的商品评论语料库作为源域,提高了目标域中的电力短文本的情感分类性能。在现有基于注意力机制的双向长短型记忆网络模型之上引入域适应层,以学习跨域知识并保留特定域的知识。实验结果表明,与其他算法相比较,该算法对电力短文本进行情感分类的效果优于非迁移学习方法,具有更好的分类性能。

To improve customer service efficiency and active service consciousness effectively in the electric power industry,a method of sentiment analysis based on transfer learning is presented.According to the shortage of marking corpus in the electric power industry,this method uses goods review corpus with rich annotation information as source domain.It can help to improve the emotional classification of short text in the target domain.Based on attention mechanism,domain adaptation layer is added in the bi-directional long short-term memory model to learn cross-domain knowledge and retain domain-specific knowledge.Experiments show that compared with other algorithms,this algorithm can get better emotional classification performance for electric power short text.