Build a aspect-level classification model based on document-level and aspect-level data as proposed in Exploiting document knowledge for aspect-level sentiment classification. Build an attention-based aspect-level sentiment classification model with Bidirectional Long Short Term Memory networks (BiLSTM). Your model shall include:
- BiLSTM network that learns sentence representations from input sequences (Recommend to use Bidirectional provided by Keras to define the BiLSTM network).
- Attention network that predicts sentiment label, given the representation weighted by the attention score.
- Fully connected network that predicts sentiment label, given the representation weighted by the attention score
Requirements:
- You shall train your model based on transferring learning. That is, you need first train your doc-level model on document-level examples. Then the learned weights will be used to initialize aspect-level model and fine tune it on aspect-level examples.
- You shall use the alignment score function in attention network as same as the recommended
- You shall evaluate trained model on the provided test set and show the accuracy on test set.
Document-level and aspect-level data sets can be downloaded from. The raw data set contains two domains:
-
Restaurant reviews; and
-
Electronics reviews, use
lt_14as experimental data.