ベクトルの近傍探索ライブラリfaissの操作備忘録を書きたかったのですが、それだけだとつまらなかったので、Word2Vec等で有名な単語ベクトルの演算がBERTにより獲得されたベクトルでもできるのか調べてみました。 事前準備 ライブラリのインストール python3 -m venv .env source .env/bin/activate pip install faiss-cpu transformers numpy import torch from transformers import AutoModel, AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("cl-tohoku/bert-base-japanese-whole-word-masking") model = AutoModel.from_pretrain
![【自然言語処理】BERTの単語ベクトルで「王+女-男」を計算してみる](https://cdn-ak-scissors.b.st-hatena.com/image/square/5d83eb8b70bbe7865902b79f8bf14d0916ecc2cf/height=288;version=1;width=512/https%3A%2F%2Fres.cloudinary.com%2Fzenn%2Fimage%2Fupload%2Fs--7oEJ-8tu--%2Fc_fit%252Cg_north_west%252Cl_text%3Anotosansjp-medium.otf_55%3A%2525E3%252580%252590%2525E8%252587%2525AA%2525E7%252584%2525B6%2525E8%2525A8%252580%2525E8%2525AA%25259E%2525E5%252587%2525A6%2525E7%252590%252586%2525E3%252580%252591BERT%2525E3%252581%2525AE%2525E5%25258D%252598%2525E8%2525AA%25259E%2525E3%252583%252599%2525E3%252582%2525AF%2525E3%252583%252588%2525E3%252583%2525AB%2525E3%252581%2525A7%2525E3%252580%25258C%2525E7%25258E%25258B%25252B%2525E5%2525A5%2525B3-%2525E7%252594%2525B7%2525E3%252580%25258D%2525E3%252582%252592%2525E8%2525A8%252588%2525E7%2525AE%252597%2525E3%252581%252597%2525E3%252581%2525A6%2525E3%252581%2525BF%2525E3%252582%25258B%252Cw_1010%252Cx_90%252Cy_100%2Fg_south_west%252Cl_text%3Anotosansjp-medium.otf_37%3Aschnell%252Cx_203%252Cy_121%2Fg_south_west%252Ch_90%252Cl_fetch%3AaHR0cHM6Ly9saDMuZ29vZ2xldXNlcmNvbnRlbnQuY29tL2EtL0FPaDE0R2pHckJWV1dtWFVVeDN4Y1Vlb2xmWE0xM0hoZThBZ2V2bGQ1ZlFqPXM5Ni1j%252Cr_max%252Cw_90%252Cx_87%252Cy_95%2Fv1627283836%2Fdefault%2Fog-base-w1200-v2.png)