1 Introduction
-
This paper proposes the concept of semantic composition based on the linguistic role of negative and intensive words. Without dropping any context information, we develop a backward LSTM to model the reversing effect of negative words and the valence that modified by the intensive words on the following content.
-
Unlike previous lexicon-based methods which directly employ external sentiment lexicons, we generate the sentiment strength under different sentiment polarities of each word in the corpus, which can solve the domain-specific problem for methods based on external sentiment lexicons.
-
Our method can adapt to the situation that two negative or intensive words exist in a sentence, and can be employed to several deep neural network models to improve the performance on sentiment classification.
2 Related Work
2.1 Lexicon-Based Sentiment Classification
2.2 Deep Neural Networks for Sentiment Classification
3 Proposed Model
Notation | Description |
---|---|
tw
| The target word, i.e., a negative or intensive word |
ntw
| The number of target words |
\(S=[x_{1}, x_{2}, \ldots , x_{n}]\)
| A sentence with n words |
\(x_{i} \in R^{d}\)
| A d-dimensional word embedding of the ith word |
\(V=[v_{1}, v_{2}, \dots , v_{n+ntw}]\)
| The vector representations generated by our method |
\(F=[f_{1}, f_{2}, \ldots , f_{m}]\)
| A hidden vector containing sentiment feature of each sentence |
\(r_{\mathrm{avg}}\)
| A vector containing the average information |
\(f_{i}\)
| The sentiment feature of the ith word |
\(W\in ws*d\)
| A convolutional filter applied to continuous word embeddings |
ssinfo
| Sentiment supplement information extract by the backward LSTM |
ssvec
| A sentiment supplement vector generate by \(\lambda\) times the ssinfo |
\(s_{i}^{1}\)
| The positive polarity strength of the ith sentence |
\(s_{i}^{0}\)
| The negative polarity strength of the ith sentence |
\(C_{\mathrm{p}}\)
| The predicted value of the positive label |
\(C_{\mathrm{n}}\)
| The predicted value of the negative label |
3.1 Sentiment Supplementary Information Generation
3.2 Sentence Encoding
3.3 Model Training and Sentiment Prediction
4 Experiments
4.1 Datasets
Dataset |
\(N_S\)
|
\(L_S\)
| |V| | |N| (%) | |I| (%) |
---|---|---|---|---|---|
MR | 10,662 | 20 | 18,376 | 33.7 | 53.2 |
SST | 9613 | 17 | 17,439 | 25.8 | 49.8 |
SLS | 3000 | 12 | 5170 | 27.8 | 39.0 |
Negative words | Cannot |
Negate | |
Neither | |
Never | |
No | |
Nobody | |
... | |
Intensive words | Cannot |
Absolutely | |
Completely | |
Even | |
Just | |
Mostly | |
... |
4.2 Experiment Design
-
CNN [12]. It generates sentence representation by a convolutional layer with multiple kernels (i.e., kernels’ size of 3, 4, 5 with 100 feature maps each) and pooling operations. Note that the dropout operation is added to prevent over-fitting.
-
LSTM [9]. The whole corpus is process as a single sequence, and LSTM generates the sentence representation by calculating the mean of the whole hidden states of all words. The hidden state size was empirically set to 128.
-
CharSCNN [8]. It employs two convolutional layers to extract features from character and sentence levels, and the output of the second convolutional layer is passed to two fully connected layer is passed to two fully connected layers to calculate the sentiment score. Empirically, the context windows of word and character were set to 1. The convolution kernel size of the character-level layer and that of the sentence-level layer were, respectively, set to 20 and 150.
-
Supplementary information modeling-based methods [32]. Such methods incorporate a kind of sentiment supplementary information into three neural networks, i.e., CNN, LSTM, and CharSCNN. These new models are denoted as NIS-CNN, NIS-LSTM, and NIS-CharSCNN, where “NIS” means “Negative and Intensive Supplement.”
Dataset |
d
|
p
|
---|---|---|
MR | 128 | 0.5 |
SST | 256 | 0.3 |
SLS | 128 | 0.2 |
4.3 Evaluation Metrics
4.4 Results and Analysis
Model | Dataset | Accuracy | Precision | Recall | F1-measure |
---|---|---|---|---|---|
SentiWordNet [3] | MR | 58.3 | 56.3 | 77.8 | 65.3 |
SLS | 64.9 | 59.7 | 86.1 | 70.5 | |
SST | 60.1 | 59.0 | 78.4 | 67.3 | |
SCL-NMA [14] | MR | 60.9 | 59.2 | 82.0 | 68.7 |
SLS | 69.5 | 66.4 | 88.8 | 76.0 | |
SST | 65.0 | 63.4 | 84.7 | 72.5 | |
Opinion Lexicon [10] | MR | 69.0 | 68.8 | 73.9 | 71.3 |
SLS | 80.6 | 76.5 | 94.9 | 84.7 | |
SST | 73.7 | 74.6 | 78.4 | 76.5 | |
CNN [12] | MR | 78.9 | 79.5 | 77.9 | 78.7 |
SLS | 87.8 | 89.3 | 85.9 | 87.6 | |
SST | 81.6 | 82.5 | 80.2 | 81.3 | |
NIS-CNN [32] | MR | 79.8 | 79.7 | 80.0 | 79.8 |
SLS | 88.3 | 89.7 | 86.5 | 88.1 | |
SST | 82.1 | 82.6 | 81.3 | 82.0 | |
IM-CNN | MR | 79.1 | 79.2 | 78.9 | 79.1 |
SLS | 88.0 | 89.0 | 86.7 | 87.8 | |
SST | 81.7 | 82.3 | 80.1 | 81.5 | |
NM-CNN | MR | 79.7 | 79.9 | 79.4 | 79.6 |
SLS | 88.4 | 89.5 | 87.0 | 88.2 | |
SST | 82.2 | 82.8 | 81.3 | 82.0 | |
NIM-CNN | MR | 80.1 | 79.8 | 80.6 | 80.2 |
SLS | 88.6 | 89.9 | 87.0 | 88.4 | |
SST | 82.3 | 82.9 | 81.4 | 82.1 | |
LSTM [9] | MR | 75.9 | 75.4 | 76.9 | 76.1 |
SLS | 85.8 | 86.0 | 85.5 | 85.8 | |
SST | 75.8 | 77.6 | 72.3 | 75.0 | |
NIS-LSTM [32] | MR | 76.2 | 76.0 | 76.6 | 76.3 |
SLS | 86.1 | 86.3 | 85.8 | 86.1 | |
SST | 76.3 | 77.5 | 74.1 | 75.8 | |
IM-LSTM | MR | 76.0 | 75.3 | 77.4 | 76.3 |
SLS | 85.9 | 85.8 | 86.0 | 85.9 | |
SST | 75.8 | 77.9 | 72.0 | 74.9 | |
NM-LSTM | MR | 76.2 | 75.8 | 77.0 | 76.4 |
SLS | 86.2 | 86.3 | 86.1 | 86.2 | |
SST | 76.2 | 78.4 | 72.3 | 75.2 | |
NIM-LSTM | MR | 76.3 | 76.2 | 76.5 | 76.3 |
SLS | 86.4 | 86.5 | 86.3 | 86.4 | |
SST | 76.4 | 78.7 | 72.4 | 75.4 | |
CharSCNN [8] | MR | 74.0 | 75.1 | 71.8 | 73.4 |
SLS | 86.4 | 88.6 | 85.3 | 86.9 | |
SST | 81.7 | 83.1 | 79.6 | 81.3 | |
NIS-CharSCNN [32] | MR | 74.4 | 75.5 | 72.2 | 73.8 |
SLS | 86.9 | 88.4 | 85.9 | 87.3 | |
SST | 82.0 | 83.5 | 79.8 | 81.6 | |
IM-CharSCNN | MR | 74.1 | 75.1 | 72.1 | 73.6 |
SLS | 86.6 | 88.4 | 85.2 | 86.7 | |
SST | 81.9 | 83.3 | 79.8 | 81.5 | |
NM-CharSCNN | MR | 74.4 | 75.7 | 71.9 | 73.7 |
SLS | 86.9 | 88.9 | 85.7 | 87.3 | |
SST | 82.2 | 83.4 | 80.4 | 81.8 | |
NIM-CharSCNN | MR | 74.6 | 75.2 | 73.4 | 74.3 |
SLS | 87.3 | 88.8 | 85.9 | 87.3 | |
SST | 82.3 | 83.6 | 80.4 | 82.0 |
Sentence |
NW
|
C
| CNN | NM-CNN | ||
---|---|---|---|---|---|---|
Pos
|
Neg
|
Pos
|
Neg
| |||
You cannot help but get caught up | Cannot | Positive | 38.1 | 61.9 | 58.3 | 41.7 |
Hollywood wouldn’t have the guts to make | Not | Positive | 41.7 | 58.3 | 61.4 | 38.6 |
The story is nowhere near gripping enough | Nowhere | Negative | 69.1 | 30.9 | 32.9 | 67.1 |
Sentence |
IW
|
C
| CNN | IM-CNN | ||
---|---|---|---|---|---|---|
Pos
|
Neg
|
Pos
|
Neg
| |||
An extremely unpleasant film | Extremely | Negative | 22.8 | 77.2 | 5.3 | 94.7 |
Really quite funny | Really | Positive | 73.5 | 26.5 | 88.0 | 12.0 |
Too silly to take seriously | Too | Negative | 19.8 | 80.2 | 11.5 | 88.5 |
The tenderness of the piece is still intact | Still | Positive | 52.8 | 47.2 | 42.7 | 57.3 |