Deep Neural Network for Sentiment Analysis on Twitter
Switch branches/tags
Nothing to show
Clone or download
Latest commit 74a16c7 Dec 17, 2017
Permalink
Failed to load latest commit information.
data added data files(huge\!) Feb 22, 2015
models added data files(huge\!) Feb 22, 2015
nltk_data/tokenizers/punkt add tokenizer pickle Feb 25, 2015
report minor change Mar 3, 2015
.gitignore added data files(huge\!) Feb 22, 2015
.requirements.txt.swp modified requirements.txt Feb 24, 2015
LICENSE added MIT licence May 5, 2015
Procfile Fixed the wrong main problem name in Procfile Feb 25, 2015
README.md Update README.md Dec 17, 2017
adadelta.py isolated adadelta code and test it against some numpy implementation Mar 8, 2015
curl_test.sh fixed function name bug in web_app Feb 25, 2015
data_collection.py fixed a bug Jan 24, 2015
dcnn.py moved numpy implementation of logreg layer to numpy_impl Mar 6, 2015
dcnn_train.py Numpy implementation of DCNN done\! Feb 21, 2015
distribute.sh minor change Mar 3, 2015
hyper_param_sample.py bug fix for hyperparam sampling(when 'values' field is absent) Feb 16, 2015
k_max_pool_test.py some playboard on theano added Jan 25, 2015
logreg.py recnn compilation/running done Mar 5, 2015
numpy_impl.py moved numpy implementation of logreg layer to numpy_impl Mar 6, 2015
param_tuning.md 1. fixed argparser problem. 2. save plots to disk automatically 3. ad… Jan 30, 2015
param_util.py added missing files Feb 25, 2015
plot_util.py fixed plotting bug Feb 2, 2015
pretrain_embedding.py added word2vec embedding pretrain Jan 28, 2015
ptb.py training code runnable now, still tuning the parameters Mar 8, 2015
recnn.py training code runnable now, still tuning the parameters Mar 8, 2015
recnn_train.py training code runnable now, still tuning the parameters Mar 8, 2015
recnn_util.py training code runnable now, still tuning the parameters Mar 8, 2015
requirements.txt modified requirements.txt Feb 24, 2015
run_example.sh example command to run the main program Feb 1, 2015
sentiment.py integrated DCNN sentiment to web app Feb 21, 2015
stopall.sh much better stop-all-tasks-on-ukko bash utility Feb 8, 2015
test_2d_3d.py some playboard on theano added Jan 25, 2015
test_adadelta.py isolated adadelta code and test it against some numpy implementation Mar 8, 2015
test_conv2d.py Numpy implementation of DCNN done\! Feb 21, 2015
test_dropout.py added some simple toy program for testing purpose Jan 26, 2015
test_embedding.py added some simple toy program for testing purpose Jan 26, 2015
test_learning_rate.py added argsparse feature Jan 29, 2015
test_numpy_convfoldpool.py test for embedding done, test for convfoldpool layer fails Feb 21, 2015
test_numpy_dcnn.py Numpy implementation of DCNN done\! Feb 21, 2015
test_numpy_embedding.py Numpy implementation of DCNN done\! Feb 21, 2015
test_numpy_logreg.py minor module name change Mar 6, 2015
test_numpy_performance.py Numpy implementation of DCNN done\! Feb 21, 2015
test_recnn.py training code runnable now, still tuning the parameters Mar 8, 2015
test_recnn_train.py RecNN numpy implementation done Mar 6, 2015
test_rntn_layer.py RecNN numpy implementation done Mar 6, 2015
test_util.py minibatch training available for RecNN Mar 6, 2015
thoughts added 'stuff behind the scene', parameter tuning etc Feb 21, 2015
tree_stat.py restructured code organization and added token frequency table calcul… Mar 6, 2015
util.py Numpy implementation of DCNN done\! Feb 21, 2015
web_app.py fixed function name bug in web_app Feb 25, 2015

README.md

Twitter Sentiment Analysis using ConvNet

A tool to

  • predict sentiment "positiveness" for tweets

How to use it?

>> from sentiment import sentiment_score
>> print sentiment_score(u"I love you")
0.9999

It returns a sentiment index ranging from 0 (negative sentiment) to 1 (positive sentiment).

Online Demo

  • predict sentiment "positiveness" for single tweets
  • gain an overview of the "positiveness" of hashtags

Click here

Algorithm

Please refer to A Convolutional Neural Network for Modelling Sentences for more information about the algorithm.

Technical choices

  • Tornado as the web framework
  • Theano as the neural network training implementation
  • Scipy as the neural network classification(online version) implementation

Training techniques

  1. Fan-in, fan-out initialization
  2. Dropout
  3. AdaDelta

Contributors

Han Xiao and Yao Lu