Skip to content
Sentiment Analysis Challenge
Jupyter Notebook Python HTML Other
Branch: master
Clone or download

Latest commit

Fetching latest commit…
Cannot retrieve the latest commit at this time.


Type Name Latest commit message Commit time
Failed to load latest commit information.
benchmarks Update output for building Arabic Tweet vectors Mar 14, 2016
frameworks/docker Merge pull request #85 from kfoss/dockerupdates Mar 14, 2016
project Update project status with baseline progress Nov 7, 2015
results Migrate all raw results to results/data Dec 2, 2015
src Update ArabicTwitter class with option for tokenizer Mar 14, 2016
.gitattributes Update ikeras with GitHub commit-and-push button on notebooks Sep 17, 2015
.gitignore Update gitignore to avoid datasets Nov 24, 2015
LICENSE create license file Jul 20, 2016 Update README with table of blogs Mar 14, 2016


Lab41's foray into Sentiment Analysis with Deep Learning. In addition to checking out the source code, visit the Wiki for Learning Resources and possible Conferences to attend.

Try them, try them, and you may! Try them and you may, I say.

Table of Contents

Blog Overviews

Can Word Vectors Help Predict Whether Your Chinese Tweet Gets Censored? March 2016
One More Reason Not To Be Scared of Deep Learning March 2016
Some Tips for Debugging in Deep Learning January 2016
Faster On-Ramp to Deep Learning With Jupyter-driven Docker Containers November 2015
A Tour of Sentiment Analysis Techniques: Getting a Baseline for Sunny Side Up November 2015
Learning About Deep Learning! September 2015

Docker Environments

  • lab41/itorch-[cpu|cuda]: iTorch IPython kernel for Torch scientific computing GPU framework
  • lab41/keras-[cpu|cuda|cuda-jupyter]: Keras neural network library (CPU or GPU backend from command line or within Jupyter notebook)
  • lab41/neon-[cuda|cuda7.5]: neon Deep Learning framework (with CUDA backend) by Nervana
  • lab41/pylearn2: pylearn2 machine learning research library
  • lab41/sentiment-ml: build word vectors (Word2Vec from gensim; GloVe from glove-python), tokenize Chinese text (jieba and pypinyin), and tokenize Arabic text (NLTK and Stanford Parser)
  • lab41/mechanical-turk: convert CSV of Arabic tweets to individual PNG images for each Tweet (to avoid machine-translation of text) and auto-submit/score Arabic sentiment survey via AWS Mechanical Turk

Binary Classification with Word Vectors


python -m benchmarks/baseline_classifiers

Word Vector Models

model filename filesize vocabulary details
Sentiment140 sentiment140_800000.bin 153M 83,586 gensim Word2Vec(size=200, window=5, min_count=10)
Open Weiboscope openweibo_fullset_hanzi_CLEAN_vocab31357747.bin 56G 31,357,746 jieba-tokenized Hanzi Word2Vec(size=200, window=5, min_count=1)
Open Weiboscope openweibo_fullset_min10_hanzi_vocab2548911.bin 4.6G 2,548,911 jieba-tokenized Hanzi Word2Vec(size=200, window=5, min_count=10)
Arabic Tweets arabic_tweets_min10vocab_vocab1520226.bin 1.2G 1,520,226 Stanford Parser-tokenized Word2Vec(size=200, window=5, min_count=10)
Arabic Tweets arabic_tweets_NLTK_min10vocab_vocab981429.bin 759M 981,429 NLTK-tokenized Word2Vec(size=200, window=5, min_count=10)

Training and Testing Data

train/test set filename filesize details
Sentiment140 sentiment140_800000_samples_[test/train].bin 183M 80/20 split of 1.6M emoticon-labeled Tweets
Open Weiboscope openweibo_hanzi_censored_27622_samples_[test/train].bin 25M 80/20 split of 55,244 censored posts
Open Weiboscope openweibo_800000_min1vocab_samples_[test/train].bin 564M 80/20 split of 1.6M deleted posts
Arabic Twitter arabic_twitter_1067972_samples_[test/train].bin 912M 80/20 split of 2,135,944 emoticon-and-emoji labeled Tweets

Binary Classification via Deep Learning

CNN (Convolutional Neural Network)

Character-by-character processing From Zhang and LeCun's Text Understanding From Scratch:

#Set Parameters for final fully connected layers
fully_connected = [1024,1024,1]

model = Sequential()

#Input = #alphabet x 1014

#Input = 336 x 256

#Input = 110 x 256

#Input = 108 x 256

#Input = 106 x 256

#Input = 104 X 256


#Fully Connected Layers

#Input is 8704 Output is 1024

#Input is 1024 Output is 1024

#Input is 1024 Output is 1

#Stochastic gradient parameters as set by paper
sgd = SGD(lr=0.01, decay=1e-5, momentum=0.9, nesterov=True)
model.compile(loss='binary_crossentropy', optimizer=sgd, class_mode="binary")

LSTM (Long Short Term Memory)

# initialize the neural net and reshape the data
model = Sequential()
model.add(Embedding(max_features, embedding_size)) # embed into dense 3D float tensor (samples, maxlen, embedding_size)
model.add(Reshape(1, maxlen, embedding_size)) # reshape into 4D tensor (samples, 1, maxlen, embedding_size)

# convolution stack
model.add(Convolution2D(nb_feature_maps, nb_classes, filter_size_row, filter_size_col, border_mode='full')) # reshaped to 32 x maxlen x 256 (32 x 100 x 256)

# convolution stack with regularization
model.add(Convolution2D(nb_feature_maps, nb_feature_maps, filter_size_row, filter_size_col, border_mode='full')) # reshaped to 32 x maxlen x 256 (32 x 100 x 256)
model.add(MaxPooling2D(poolsize=(2, 2))) # reshaped to 32 x maxlen/2 x 256/2 (32 x 50 x 128)

# convolution stack with regularization
model.add(Convolution2D(nb_feature_maps, nb_feature_maps, filter_size_row, filter_size_col)) # reshaped to 32 x 50 x 128
model.add(MaxPooling2D(poolsize=(2, 2))) # reshaped to 32 x maxlen/2/2 x 256/2/2 (32 x 25 x 64)

# fully-connected layer
model.add(Dense(nb_feature_maps * (maxlen/2/2) * (embedding_size/2/2), fully_connected_size))

# output classifier
model.add(Dense(fully_connected_size, 1))

# try using different optimizers and different optimizer configs
model.compile(loss='binary_crossentropy', optimizer='adam', class_mode="binary")
You can’t perform that action at this time.