Sentiment analysis on a large dataset There is a huge database available at http://help.sentiment140.com/for-students/ to be used for sentiment analyses I ran the code in an iMac. It took 13 hours to create a lexicon, train a model and test it. The accuracy is 74%

If you want to download these files and run, here are some suggestions: Recreational training: If you want to check if the code compiles and have a quick fun to see the response of LSTM, make the following changes to the settings: in file utils_senti.py, set hm_lines = 100 or 1000 (or any number in between) in NN_senti.py, set hm_data and hm_test_data = 100 or 1000 (or any number in between)

For learning: If you are moderately interested in testing the system and also learn programming LSTMs, make the following changes to the settings: in file utils_senti.py, set hm_lines = 1000 or 50000 (or any number in between) in NN_senti.py, set hm_data and hm_test_data = 1000 or 50000 (or any number in between) Th above settings will take a couple hours to a few hours for compilation

For Hardcore enthusiasts: If you are highly interested in becoming an expert in LSTMs, run the files as-is with the original settings The files will have a day or two for compilation. If you have a faster instance on GCP or AWS, you are in better luck. A good GPU will also help.

I am currently working on getting higher accuracies will different hyper parameters and other changes.

Settings: Tensorflow: 1.0, Python >= 3.5

Execution steps:

Step1: Download the *.py files Step2: Download the datasets (training.1600000.processed.noemoticon.csv and testdata.manual.2009.06.14.csv) from the above link into the same directory Step3: Open and terminal window and type python utils_senti.py Step 4: The size of the lexicon is published in the window after the successful execution of utils_senti.py. Update the lexicon size in NN_senti.py Step 5: Run the next command python NN_senti.py