TextSentimentClassification, using tensorflow. Original Data
Remove the letter whose number of repetitions is over 3 from a word...
Using word2vec and GloVe to generate word vectors...
Model | Epoch | Training Accuracy | Validation Accuracy | Parameters(word vectors excluded) |
---|---|---|---|---|
TextCNN+nonstatic | 130 | 0.8839 | 0.8142 | 281,202 |
TextRNN+nonstatic | 150 | 0.8383 | 0.8199 | 285,826 |
CRNN+nonstatic | 70 | 0.8600 | 0.8219 | 274,818 |
RCNN+nonstatic | 50 | 0.8553 | 0.8227 | 318,978 |
HAN+nonstatic | 110 | 0.8355 | 0.8188 | 209,410 |
Convolutional Neural Networks for Sentence Classification
- CNN-rand
- CNN-static
- CNN-nonstatic
- CNN-multichannel
Model | Epoch | Training Accuracy | Validation Accuracy | Parameters(word vectors excluded) |
---|---|---|---|---|
TextCNN+rand | 130 | 0.8761 | 0.8137 | 281,202 |
TextCNN+static | 60 | 0.9015 | 0.8113 | 281,202 |
TextCNN+nonstatic | 130 | 0.8839 | 0.8142 | 281,202 |
TextCNN+multichannel | 60 | 0.9225 | 0.8141 | 561,202 |
Choosing to use word vectors in a nonstatic way.
Using bidirectional RNN, and then concatenating the output of the forward process and the output of the backward process...
A C-LSTM Neural Network for Text Classification
Using CNN to extract sentences with higher-level phrase representations, and then learning long short-term dependency with bi-RNN...
Recurrent Convolutional Neural Networks for Text Classification
In addition to implementing the same structure as the paper, using bi-LSTM or bi-GRU and then concatenating their outputs... RNN for capturing contextual information and max pooling used for judging which words play key roles in the task...
Hierarchical Attention Networks for Document Classification
Transforming a sentence into a document consisting of sentences...
Uniform blending...
Using Logistic Regression as the level-2 classifier...
Model | Epoch | Training Accuracy | Testing Accuracy | Parameters(word vectors excluded) |
---|---|---|---|---|
LR+static_avg | - | 0.77364 | 0.773605 | - |
NB+static_avg | - | 0.606435 | 0.61082 | - |
TextCNN+nonstatic | 130 | 0.8703 | 0.817615 | 281,202 |
TextRNN+nonstatic | 150 | 0.8384 | 0.81969 | 285,826 |
CRNN+nonstatic | 70 | 0.8589 | 0.82449 | 274,818 |
RCNN+nonstatic | 50 | 0.8497 | 0.822935 | 318,978 |
HAN+nonstatic | 110 | 0.8330 | 0.820235 | 209,410 |
bagging | - | 0.8538 | 0.82999 | - |
stacking | - | 0.867135 | 0.831045 | - |