There is an overwhelming amount of news information available online. Some news headlines are known as clickbait – they aim to attract users to click on a link but the articles that they link to may not be of value or interest to the reader. This program automatically distinguishes between clickbait and non-clickbait headlines.
Two corpora of clickbait and non-clicbait headlines are included. Each corpus counts 16,000 headlines, for a total of 32,000 headlines.
The code is informed by the paper by Chakraborty et al. (2016). The program loads the data, extracts sets of features as frequency-count vectors, and uses them to train a Naive Bayes classifier. The classifier accuracy is generated using 10-fold cross-validation and output for each feature set individually. The program extracts the following features:
- Stop words: counts for each function word (from the NLTK stopwords list)
- Syntactic: counts for the following 10 common POS tags:
['NN', 'NNP', 'DT', 'IN', 'JJ', 'NNS','CC','PRP','VB','VBG'] - Lexical: counts for 30 most common unigrams in the entire corpus
- Punctuation: counts for each punctuation mark in
string.punctuation - Complexity:
- Average number of characters per word
- Type-to-token ratio (the number of unique words / the total number of words)
- Count of long words - words with at least 6 letters
- Interrogative words: counts of the common English interrogatives
Clickbait Detector achieves 91% accuracy for all features combined, meaning that it correctly identifies over 9 out of 10 headlines.