Machine Learning Algorithms - Second Edition

Introducing Word2vec with Gensim

One of the most common problems in NLP and topic modeling is represented by the semantic-free structure of the Bag-of-Words strategy. In fact, as discussed in the previous chapter, Chapter 13, Introducing Natural Language Processing, this strategy is based on frequency counts and doesn't take into account the positions and the similarity of the tokens. The problem can be partially mitigated by employing n-grams; however, it's still impossible to detect the contextual similarity of words. For example, let's suppose that a corpus contains the sentences John lives in Paris and Mark lives in Rome. If we perform a Part-of-Speech (POS) and Named Entity Recognition (NER) tagging, we can discover that John and Mark ...