overlast さんによる Word2Vec の WebAPI である word-vector-web-api を hubot から利用する
npmでインストール
$ npm install --save knjcode/hubot-word-vector-script
or
$ npm install --save https://github.com/knjcode/hubot-word-vector-script/archive/master.tar.gz
external-scripts.jsonにhubot-word-vector-scriptを追加
$ cat external-scripts.json
["hubot-word-vector-script"]
README を参考に word-vector-web-api を導入します。
APIとしては以下のように利用できます。
$ curl "http://localhost:22670/distance?a1=寿司"
{"method": "distance", "format": "json", "items": [
{"term": "握り寿司", "score": 0.7822030782699585}, {"term": "江戸前寿司", "score": 0.77414518594741821},
{"term": "押し寿司", "score": 0.77050793170928955}, {"term": "うどん", "score": 0.76514482498168945},
{"term": "天ぷら", "score": 0.7647247314453125}, {"term": "おでん", "score": 0.75914901494979858},
{"term": "ラーメン", "score": 0.7530028223991394}, {"term": "おにぎり", "score": 0.75069379806518555},
{"term": "巻き寿司", "score": 0.75040823221206665}, {"term": "しゃぶしゃぶ", "score": 0.74389779567718506},
{"term": "日本料理", "score": 0.74245506525039673}, {"term": "酢飯", "score": 0.74163687229156494},
{"term": "ちらし寿司", "score": 0.74119293689727783}, {"term": "中華料理", "score": 0.7407233715057373},
{"term": "麺類", "score": 0.73966968059539795}, {"term": "軍艦巻き", "score": 0.73962771892547607},
{"term": "寿司屋", "score": 0.73787486553192139}, {"term": "料理", "score": 0.73675769567489624},
{"term": "丼物", "score": 0.72888869047164917}, {"term": "蕎麦", "score": 0.72767460346221924},
{"term": "刺身", "score": 0.7261509895324707}, {"term": "鍋料理", "score": 0.72471451759338379},
{"term": "料理店", "score": 0.72462576627731323}, {"term": "焼肉", "score": 0.72449952363967896},
{"term": "和食", "score": 0.72130364179611206}, {"term": "チャーハン", "score": 0.72093665599822998},
{"term": "すき焼き", "score": 0.7177349328994751}, {"term": "トンカツ", "score": 0.71647495031356812},
{"term": "早ずし", "score": 0.71499419212341309}, {"term": "ご飯", "score": 0.71361935138702393},
{"term": "にぎり寿司", "score": 0.71285635232925415}, {"term": "照り焼き", "score": 0.7118719220161438},
{"term": "鍋物", "score": 0.71128165721893311}, {"term": "たこ焼き", "score": 0.70872247219085693},
{"term": "丼", "score": 0.70830398797988892}, {"term": "エビフライ", "score": 0.70730042457580566},
{"term": "佃煮", "score": 0.70715677738189697}, {"term": "とんかつ", "score": 0.70401555299758911},
{"term": "茶漬け", "score": 0.70242542028427124}, {"term": "蒲焼", "score": 0.70182281732559204}
], "query": "寿司", "total_count": 40, "status": "OK", "sort": "cosine similarity"}
下記の環境変数を設定して hubot を起動
-
HUBOT_WORD_VECTOR_URL
word-vector-web-api の URLを指定（必須）
ローカルにインストールした場合:
export HUBOT_WORD_VECTOR_URL="http://localhost:22670" -
HUBOT_WORD_VECTOR_COUNT
検索結果の表示件数を制限する（デフォルトは5件）
30件の場合:
export HUBOT_WORD_VECTOR_COUNT=30
hubot distance <string>
<string>にベクトルが近い語を出力
user1>> hubot distance 寿司
hubot>> 握り寿司 (0.7822030782699585)
江戸前寿司 (0.7741451859474182)
押し寿司 (0.7705079317092896)
うどん (0.7651448249816895)
天ぷら (0.7647247314453125)
hubot analogy <str1> <str2> <str3>
<str1> - <str2> + <str3> にベクトルが近い語を出力
user1>> hubot analogy SONY PlayStation Nintendo
hubot>> Wii (0.7002160549163818)
ゲームボーイアドバンス (0.694932758808136)
ニンテンドーDS (0.6919329166412354)
ニンテンドー3DS (0.6773288249969482)
NINTENDO64 (0.6766645312309265)