Deep Learning for Sentiment Analysis

Sentiment Analysis trained on Sentiment140

Installation

The following script contains the (Python) requirements and training data.

bash ./scripts/0_install_prerequisites.sh

Usage

Training

from DeepSentiment.Networks.Tensorflow.Model import Model as TFModel

args = {'num_train_epochs': 1} #override Trainings Settings see  [DeepSentiment/Consts/Training.py]

model, history = TFModel().trainModel(CNN_LAYER = True,  #self-trained word2vec embedding layer
            POOLING_LAYER = True, 
            BiLSTM_Layer = True, 
            args=args)
            
#model, history = TFModel().trainModel(GLOVE = True, 
#            CNN_LAYER = True 
#            POOLING_LAYER = True 
#            GRU_LAYER = True 
#            BiLSTM_Layer = True 
#            LSTM_Layer = True 
#            DENSE_LAYER = True,
#            logger = logger)

Load Model

model = TFModel().loadModel(CNN_LAYER = True,  #self-trained word2vec embedding layer
            POOLING_LAYER = True, 
            BiLSTM_Layer = True)

Prediction

self-trained embedding layer (Word2Vec)

from DeepSentiment.Preprocessing.CleanText import CleanText

sample_text = ('The movie was not good. The animation and the graphics '
                    'were terrible. I would not recommend this movie.')
sample_text_cleaned = CleanText().cleanText(sample_text)
model.predict([sample_text])

GloVe embedding layer (subject to change)

from DeepSentiment.Preprocessing.CleanText import CleanText
from DeepSentiment.Dataset import Sentiment140
from DeepSentiment.Consts import (
    Global as Global, 
    Glove as Glove, 
    Paths as Paths, 
    Preprocessing as Preprocessing, 
    Training as Training 
)

args = Training.trainArgs

s140 = Sentiment140.Dataset(path=Paths.SENTIMENT140_DATASET, 
                                parsedPath=Paths.SENTIMENT140_DATASET_PARSED,
                                embeddingDim=Glove.GLOVE_DIM, 
                                MAX_SEQUENCE_LENGTH=Preprocessing.MAX_SEQUENCE_LENGTH,
                                args=args)
train_data, test_data, labelDecoder = s140.load(padInput=True,                                                 
                                                        DEBUG=Global.DEBUG, 
                                                        cleanFN = CleanText().cleanText,
                                                        Tokanize = True,
                                                        BERT = False)

sample_text = ('The movie was not good. The animation and the graphics '
                    'were terrible. I would not recommend this movie.')
sample_text_cleaned = CleanText().cleanText(sample_text)
sample_text_glove = s140.padInput(sample_text_cleaned)
predictions = model.predict([sample_text_glove])