This is another method, where the topics are identified from the corpus of text. The topics can be single words, patterns of words, or sequences of co-occurring words. Based on a number of words in the topic, these could be called N-Gram. So, based on context and repeatability, bigrams and trigrams could be used as features.