Simple implementation of topic modelling using NMF and LDA

requirements.txt

sklearn>=0.18.2
nltk>=3.2.3

topic-modelling.py

from sklearn.feature_extraction.text import TfidfVectorizer, CountVectorizer
from sklearn.decomposition import NMF, LatentDirichletAllocation
from nltk.corpus import stopwords


def display_topics(model, feature_names, no_top_words):
    for topic_idx, topic in enumerate(model.components_):
        print('{}'.format(topic_idx), ' '.join([feature_names[i]
                        for i in topic.argsort()[:-no_top_words - 1:-1]]))
      
    
def main():        
    # read the file, line per line into an array        
    transcripts = []
    with open('/path/to/some/txt/file', encoding='some-encoding') as t:
        transcripts = [_.strip() for _ in t.readlines()]
    len(transcripts)

    # create a term frecuency vectorizer for target language and fit to data
    tfv = TfidfVectorizer(max_df=0.95, min_df=2, max_features=1000, stop_words=stopwords.words('target-language'))
    tfv_data = tfv.fit_transform(transcripts)
    tfv_features = tfv.get_feature_names()

    # create a count vectorizer for target language and fit to data
    cv = CountVectorizer(max_df=0.95, min_df=2, max_features=1000, stop_words=stopwords.words('target-language'))
    cv_data = cv.fit_transform(transcripts)
    cv_features = cv.get_feature_names()

    # set number of topics
    no_topics = 8

    # fit NMF
    nmf = NMF(
        n_components=no_topics, 
        alpha=.1, 
        l1_ratio=.5, 
        init='nndsvd',
    random_state=1).fit(tfv_data)    

    # fit LDA
    lda = LatentDirichletAllocation(
        n_topics=no_topics, 
        max_iter=5, 
        learning_method='online', 
        learning_offset=50., 
        random_state=1).fit(cv_data)

    # set number of words per topic to display
    no_top_words = 10

    # show results for both algorithms
    display_topics(nmf, tfv_features, no_top_words)
    display_topics(lda, cv_features, no_top_words)
    

if __name__ == '__main__':
    main()