In this challenge, we had to classify comments from online discussion that were insulting or not. One particularity of this challenge is that, as an exercise, we were not allowed to use sklearn, NLTK or any other machine learning or NLP libraries. Therefore, we had to rewrite in Python every needed algorithm.
We used several NLP tools such as TF-IDF and n-grams to build the features for our model, and we used a linear SVM to classify the comments.
We achieved a score of 0.820 (accuracy metric) and finished second among 55 participants.