model2json.py

#Author: Phi Van Thuy
#Purpose: Convert word2vec models to JSON database by cosine distance metric

from gensim.models import Word2Vec

#Trained model
# model_path = "/cl/work/thuy-ph/word2vec/GoogleNews-vectors-negative300.bin"

print "Loading model..."
model = Word2Vec.load('emma-model')
# model = word2vec.Word2Vec.load_word2vec_format(model_path, binary=True)  # C binary format
print "Loading model: Done"

#Name of output file
f = open('en_data_cosine_skipgram_original.json','w')

f.write("{\n")

number_words = len(model.vocab)
#number_words = 10000
for i in range(0, number_words):
  stringA = model.vocab.items()[i][0]
  f.write("\n\"" + stringA.encode("utf-8") + "\":[\n")

  nearest_words = model.most_similar(positive=[stringA], negative=[],  topn=20)
  number_nearest_words = len(nearest_words)

  for j in range(0, number_nearest_words):
    f.write("{\"w\":\"" + nearest_words[j][0].encode("utf-8") + "\",\"d\":" + str(round(nearest_words[j][1], 3)) + "}")
    if j != number_nearest_words - 1:
      f.write(",\n")
    else:
      f.write("]")

  if i != number_words - 1:
    f.write(",\n")
  else:
    f.write("\n")

f.write("\n}\n")

f.close()

print "Finished!"