Fine-grained sentiment analysis analyzes the emotional polarity of text from multiple angles, and it has become a hot issue in the field of sentiment analysis. Different from previous LSTM networks in which the algorithm uses attribute information as the embedded vector and single-layer attention mechanism, this paper proposes a multi-layer network based on hierarchical attention mechanism, which can give different attention weights to words and sentences. To help the model increase the attention to important parts, on the other hand use the entity information as the embedded vector, which is more representative of the meaning of the target phrase than the attribute information. The experimental results show that the model has achieved excellent results on the SemEval 2014 dataset and is superior to the existing algorithms.
Paper
Full text
Fine-Grained Sentiment Analysis Based on Hierarchical Attention Networks
Semantic Scholar · Computer Science · 2019
Abstract
Fine-grained sentiment analysis analyzes the emotional polarity of text from multiple angles, and it has become a hot issue in the field of sentiment analysis. Different from previous LSTM networks in which the algorithm uses attribute information as the embedded vector and single-layer attention mechanism, this paper proposes a multi-layer network based on hierarchical attention mechanism, which can give different attention weights to words and sentences. To help the model increase the attention to important parts, on the other hand use the entity information as the embedded vector, which is more representative of the meaning of the target phrase than the attribute information. The experimental results show that the model has achieved excellent results on the SemEval 2014 dataset and is superior to the existing algorithms.