tweet2vec

THE project.

Preprocessing

To run preprocessing:

python preprocess.py data.txt

This will run a test which will iterate a few times through the lines of data.txt, and print the raw text, clean text, and hashtags.

Running without arguments (i.e. just python preprocess.py) will default to looking for the file ./data/sample.csv

If the file is not found, it will default to just running on 4 sample tweets provided in the code.

To prepare the data for the model:

python preprocess.py --prepare data.txt

This will generate the file in ./models/mlb.pickle, which is a MultiLabelBinarizer object. This is the object that turns a list of hashtags into an encoded vector (one-hot for each hashtag), for our model.

By default it will filter out hashtags that don't appear more than 10 times. To change this number to say, 100:

python preprocess.py --prepare --threshold 100 data.txt

Note: you will want to remove the ./models/mlb.pickle file if you are generating a new one. If a mlb.pickle exists, the script will load it and then filter out any hashtags that aren't already in the model.

TweetIterator

The main object in prepocessing.py is the TweetIterator object. It allows you to iterator through lines in a text file and yield any number of things useful for our model. This allows you to process and generate features for input files in a memory efficient way (i.e. you never have to load all of your data into memory).

The second argument determines whether you skip samples which do not have any hashtags.

To iterate through the raw text:

tweet_iterator = TweetIterator('source.txt', False, 'raw_text')
for t in tweet_iterator:
    print(t)

To iterate through the hashtags:

tweet_iterator = TweetIterator('source.txt', False, 'hashtags')
for t in tweet_iterator:
    print(t)

If you have an MultilabelBinarizer object prepared, iterate through the label vectors:

tweet_iterator = TweetIterator('source.txt', False, 'label')
for t in tweet_iterator:
    print(t)

Iterate through character matrices:

tweet_iterator = TweetIterator('source.txt', False, 'char_mat')
for t in tweet_iterator:
    print(t)

If you have a word2vec model saved as ./models/w2v.pickle or ./models/w2v.bin, you can iterate through word2vec matrices:

tweet_iterator = TweetIterator('source.txt', False, 'word_mat')
for t in tweet_iterator:
    print(t)

You can combine any number of these:

tweet_iterator = TweetIterator('source.txt', False, 'raw_text', 'clean_test', 'hashtags', 'char_mat', 'word_mat')
for rt, ct, h, cm, wm in tweet_iterator:
    print(rt, ct, h, cm, wm)

etc...