BI-CARU Feature Extraction for Semantic Analysis

Semantic analysis is an important part of the NLP task that enables computers to understand humans. In this paper, we present the feature extraction method of our results by applying a bidirectional RNN network with CARU units and filter the feature for each domain aspect word with major information extraction. We also developed a weighted layer to represent the combination of hidden states generated by two CARU layers. Finally, the output states are concatenated to an advanced self-weighting layer to obtain more accuracy. The experimental results show that the proposed method can slightly improve the accuracy and has great potential for improvement and analysis for any neural network using RNN architecture.