1. Learn
    2. /
  2. Courses
    3. /
  3. Recurrent Neural Networks (RNNs) for Language Modeling with Keras
Connected

Exercise

Word2Vec

In this exercise you will create a Word2Vec model using Keras.

The corpus used to pre-train the model is the script of all episodes of the The Big Bang Theory TV show, divided sentence by sentence. It is available in the variable bigbang.

The text on the corpus was transformed to lower case and all words were tokenized. The result is stored in the tokenized_corpus variable.

A Word2Vec model was pre-trained using a window size of 10 words for context (5 before and 5 after the center word), words with less than 3 occurrences were removed and the skip gram model method was used with 50 dimension. The model is saved on the file bigbang_word2vec.model.

The class Word2Vec is already loaded in the environment from gensim.models.word2vec.

Instructions

100 XP
  • Load the pre-trained Word2Vec model.
  • Store a list with the words "bazinga", "penny", "universe", "spock", "brain" in the variable words_of_interest, keeping them in that order.
  • Iterate over each word of interest while using the .most_similar() method present on attribute wv and append the top 5 similar words to top5_similar_words as a dictionary.
  • Print the found top 5 words for each of the words of interest.