Doc2Vecの事前学習済みモデルから単語や文書のベクトルを獲得する

事前学習済みモデルのダウンロード

モデルをloadし，単語ベクトルを確認してみる

confirm_WordVector.py

from gensim.models.doc2vec import Doc2Vec
model = Doc2Vec.load('model/enwiki_dbow/doc2vec.bin')
vector = model.infer_vector(["word"])
print(len(vector))
print(vector)

300
[ 1.40280828e-01  1.83409289e-01 -2.64408961e-02 -1.11115627e-01
 -1.84268013e-01  2.21883774e-01 -1.39962300e-03 -7.80699700e-02
 -8.71175826e-02  2.56892532e-01  1.28477469e-01 -1.32150203e-01
  9.16299447e-02 -1.08361199e-01  2.24919140e-01  2.95133799e-01
 -1.05266787e-01 -8.87315348e-02 -3.57903123e-01 -1.16929680e-01
  2.43903622e-01 -2.26327643e-01  5.54789364e-01 -4.85761255e-01
  1.12485945e-01 -2.10972637e-01  2.92448878e-01  1.36567667e-01
 -3.39356661e-01  1.54809028e-01 -4.06004965e-01  2.51045078e-01
 -5.08944094e-01 -2.72488981e-01  2.54280418e-01 -5.28896302e-02
  5.78162111e-02  1.88219622e-01  5.74344695e-01 -4.67062503e-01
 -2.05024302e-01 -5.98332509e-02 -3.44163746e-01 -3.80116701e-01
  1.05316617e-01 -1.75678745e-01 -4.92762923e-01  3.11034918e-01
 -3.04395765e-01 -6.21335721e-03 -2.51851439e-01  6.83335662e-02
  2.69055486e-01 -4.56707805e-01  3.17851663e-01 -1.69105187e-01
  3.56151521e-01 -5.05028307e-01 -2.53974706e-01 -5.85785925e-01
  1.44802809e-01  1.71069667e-01  2.14749686e-02  2.62290016e-02
 -5.90268746e-02 -4.17226970e-01 -2.58289903e-01 -1.34147465e-01
 -1.69140883e-02  2.69945771e-01 -8.30643922e-02 -2.70083934e-01
 -5.48509397e-02 -3.51466686e-01 -2.83847153e-01  5.26780486e-01
 -9.27017778e-02  3.41789305e-01  1.61628351e-01 -9.79063809e-02
  2.50723511e-01 -3.06959093e-01 -4.54114348e-01 -1.49249837e-01
 -6.02198720e-01 -3.59645128e-01  1.29344389e-01 -4.97040823e-02
  1.67680234e-01  3.98838282e-01  3.97429094e-02 -8.42189014e-01
  4.17290986e-01  9.80646759e-02  5.52689396e-02  2.00707242e-01
 -4.96996380e-02 -3.10181230e-01  7.32129142e-02  1.78322211e-01
  1.99462384e-01  1.85920909e-01  4.16447707e-02  3.06156427e-02
  5.19993417e-02 -9.45110098e-02  3.29695880e-01 -6.64467752e-01
 -4.22538340e-01  1.76553596e-02  3.59327137e-01  1.87507823e-01
 -4.77306396e-01 -1.01719558e-01 -4.10893440e-01 -1.98205486e-01
 -2.00183213e-01 -2.72218496e-01 -2.06492599e-02  3.01751882e-01
  4.59669717e-02 -2.81522602e-01  1.15110882e-01  1.12400606e-01
 -3.65632564e-01 -2.55062699e-01  2.92361856e-01 -4.80110735e-01
  1.91051483e-01 -1.09290645e-01  3.52236956e-01  2.30695501e-01
  4.36141849e-01 -4.78955433e-02  1.11169226e-01  1.39120921e-01
 -2.11431772e-01  4.52448912e-02 -2.72998810e-01 -4.09108907e-01
 -1.19410396e-01  1.38503099e-02  7.53449369e-03 -2.37264037e-01
 -1.67033702e-01 -2.26302013e-01 -1.10190071e-01 -3.45773011e-01
  1.81666419e-01 -1.88263834e-01  2.19820291e-01 -2.88389564e-01
  1.02379367e-01  4.77272905e-02  4.77848239e-02 -2.84629092e-02
 -2.28211567e-01 -2.59289056e-01  7.43009150e-04 -1.49935097e-01
  1.42509758e-01  3.70406181e-01  4.54252928e-01  2.22431928e-01
 -2.51703948e-01  1.28542066e-01  7.27307573e-02 -3.60925421e-02
  6.45418346e-01 -2.29296759e-01 -2.46794242e-02 -3.51088405e-01
  2.99131393e-01 -1.01994380e-01  2.05502391e-01  5.13257325e-01
  2.81603962e-01  3.98386598e-01  7.68973529e-02 -2.05001414e-01
 -1.08222596e-01  3.70851427e-01  6.77625686e-02 -4.04938042e-01
  2.17772741e-02  2.16333512e-02 -1.00487657e-01  2.47037604e-01
  6.34489302e-03  2.80573443e-02  2.21345127e-01 -5.39463460e-01
 -1.15930647e-01  8.56445264e-03  7.61211962e-02 -1.54177174e-01
  1.10860772e-01 -6.26938343e-01 -3.82335544e-01 -6.73514232e-02
 -4.02066022e-01 -1.69688925e-01 -3.15610260e-01  6.77945558e-03
  3.97334605e-01  7.22034797e-02  1.42006814e-01 -2.81334162e-01
  1.22516409e-01  3.27533394e-01 -1.56189814e-01  4.97612879e-02
  2.15303227e-01 -8.69842410e-01 -1.04782172e-01 -2.61912316e-01
  8.11299086e-02 -1.06915340e-01 -4.68756080e-01  2.54943911e-02
  1.79967985e-01 -1.35952368e-01  1.55958846e-01 -1.72587708e-01
 -7.17891514e-01  1.18898049e-01  4.15051430e-02 -3.22812885e-01
  1.64221272e-01  5.38067400e-01  2.20012248e-01 -3.49850534e-03
 -9.77616534e-02  4.59246367e-01  2.99039483e-01  6.88107729e-01
  3.54239970e-01  8.27741176e-02 -1.64990410e-01  1.75694339e-02
 -4.47227359e-01  3.88276935e-01  1.59138501e-01 -4.20660712e-02
 -1.80355012e-01 -2.54727751e-01  1.02000490e-01 -1.31719366e-01
  2.61006087e-01  3.00956100e-01  1.64773628e-01 -1.77720655e-02
  3.05260122e-01  2.02634603e-01 -5.14772385e-02  1.07577242e-01
  1.58462778e-01 -2.40044415e-01  2.85942465e-01 -4.73183356e-02
 -4.39267427e-01 -1.93622246e-01 -5.14240086e-01 -1.80472374e-01
  7.56859004e-01 -7.72481337e-02 -1.83406934e-01 -1.04551455e-02
  1.79811299e-01 -2.92110503e-01  2.21000046e-01 -9.72364619e-02
  8.97284423e-04  3.09679478e-01 -5.13073541e-02  5.61714590e-01
 -8.19747671e-02 -4.49352294e-01  5.06668910e-02  4.01322357e-02
 -1.38008431e-01 -8.54075775e-02 -3.30738991e-01 -2.61910200e-01
  3.30826312e-01 -1.27152279e-01 -2.62047518e-02 -1.46580711e-01
  2.61440694e-01 -5.41437566e-01 -2.53788650e-01  1.29316911e-01
  4.26622987e-01  1.34229422e-01  5.41095473e-02  7.28423670e-02
  3.60608399e-02 -3.21376175e-01 -4.56876233e-02 -1.15793005e-01]

文書ベクトルの類似度を計算してみる

confirm_sim.py

doc_words1 = 'the picture was taken by me .'.split()
doc_words2 = 'i took a picture .'.split()
doc_words3 = 'the picture is very popular .'.split()

sim_value = model.docvecs.similarity_unseen_docs(model, doc_words1, doc_words2, alpha=1, min_alpha=0.0001, steps=5)
print(doc_words1)
print(doc_words2)
print(sim_value)


sim_value = model.docvecs.similarity_unseen_docs(model, doc_words2, doc_words3, alpha=1, min_alpha=0.0001, steps=5)
print(doc_words2)
print(doc_words3)
print(sim_value)

['the', 'picture', 'was', 'taken', 'by', 'me', '.']
['i', 'took', 'a', 'picture', '.']
0.47444016
['i', 'took', 'a', 'picture', '.']
['the', 'picture', 'is', 'very', 'popular', '.']
0.534385