Target Embedding and Position Attention with LSTM for Aspect Based Sentiment Analysis
Borun Xu, Xiaoxiao Wang, Bo Yang, Zhongfeng Kang · 2020
Aspect Based Sentiment Analysis (ABSA) provides fine-grained sentiment information compared with traditional sentiment analysis. There are two approaches to the task, aspect term sentiment analysis (ATSA) and aspect category sentiment analysis (ACSA). In this paper, we propose a model, namely Recurrent Neural Network with Target Embedding (RTE) using target enhance technic to improve the accuracy of both the two kinds of approaches. Specifically, RTE involves two stacked LSTMs for target term extraction and sentiment analysis, and a target enhance unit for spreading target or aspect information. Experiments are conducted on several public datasets and the results illustrate that the proposed RTE outperforms several state-of-the-art models compared.