Advanced Combined LSTM-CNN Model for Twitter Sentiment Analysis
Nan Chen, Peikang Wang · 2018
In this paper, we proposed an advanced model which is based on the LSTM-CNN model presented by Pedro M. Sosa for twitter sentiment analysis. We combined the encoder-decoder framework with the regular LSTM-CNN framework. In this model, LSTM can `remember' forward information of the sequence and multilayer CNN can catch and learn local information sufficiently. Meanwhile, the multilayer CNN is also regarded as an encoder and a two-layer deconvolution part is the corresponding decoder. This encoder-decoder framework is used to reconstruct the input matrix, this process of the reconstruction of input matrix by decoder makes the features learning in CNN much more intrinsic and effective. As the result, the more effective the feature learning is, the higher accuracy rate the classifier will achieve. Furthermore, our framework can also be used for other classification issues besides sentiment analysis. This work will make sense in fields such as machine learning and natural language processing.