Stop Using word2vec | Stitch Fix Technology – Multithreaded
Stop Using word2vec When I started playing with word2vec four years ago I needed (and luckily had) tons of supercomputer time. But because of advances in our understanding of word2vec, computing word vectors now takes fifteen minutes on a single run-of-the-mill computer with standard numerical libraries1. Word vectors are awesome but you don’t need a neural network – and definitely don’t need deep
Word2Vec ã®ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯å¦ç¿’éŽç¨‹ã‚’ç†è§£ã™ã‚‹ · ã‘ã‚“ã”ã®ãŠå±‹æ•·
Word2Vec ã¨ã„ã†ã¨ã€æ–‡å—通りå˜èªžã‚’ベクトルã¨ã—ã¦è¡¨ç¾ã™ã‚‹ã“ã¨ã§å˜èªžã®æ„味をã¨ã‚‰ãˆã‚‹ã“ã¨ãŒã§ãる手法ã¨ã—ã¦æœ‰åãªã‚‚ã®ã§ã™ãŒã€æœ€è¿‘ã 㨠Word2Vec ã‚’å”調フィルタリングã«å¿œç”¨ã™ã‚‹ç ”究 (Item2Vec ã¨å‘¼ã°ã‚Œã‚‹) ãªã©ã‚‚ã‚るよã†ã§ã€ã“ã® Word2Vec ã¨ã„ã†ãƒ„ールã¯è‡ªç„¶è¨€èªžå‡¦ç†ã®åˆ†é‡Žã®å£ã‚’超ãˆã¦æ´»èºã—ã¦ã„ã¾ã™ã€‚ 実㯠Item2Vec を実装ã—ã¦ã¿ãŸã㦠Word2Vec ã®ä»•çµ„ã¿ã‚’ç†è§£ã—よã†ã¨ã—ã¦ã„ãŸã®ã§ã™ãŒã€Word2Vec ã®å†…部ã®è©³ç´°ã«è¸ã¿è¾¼ã‚“ã§è§£èª¬ã—ãŸæ—¥æœ¬èªžè¨˜äº‹ã‚’見ã‹ã‘ã‚‹ã“ã¨ãŒãªã‹ã£ãŸã®ã§ã€ä»Šæ›´æ„Ÿã¯ã‚ã‚Šã¾ã™ãŒè‡ªåˆ†ã®çŸ¥è˜ã®æ•´ç†ã®ãŸã‚ã«ã‚‚ブãƒã‚°ã«æ®‹ã—ã¦ãŠãã¾ã™ã€‚ãªãŠã€ã“ã®è¨˜äº‹ã¯ Word2Vec ã®ã‚½ãƒ¼ã‚¹ã‚³ãƒ¼ãƒ‰ã¨ã„ãã¤ã‹ã®ãƒšãƒ¼ãƒ‘ーをèªã‚“ã§è‡ªåŠ›ã§ç†è§£ã—ãŸå†…容ã«ãªã‚Šã¾ã™ã€‚é–“é•ã„ãŒå«ã¾ã‚Œã¦ã„ã‚‹å¯èƒ½æ€§ã‚‚ã‚ã‚Šã¾ã™ã®ã§ã”了承ãã ã•ã„。もã—é–“é•ã„を見ã¤ã‘ãŸå ´åˆã¯æŒ‡æ‘˜ã—ã¦ã‚‚らãˆã‚‹ã¨
Globally Scalable Web Document Classification Using Word2Vec, Mon, Apr 27, 2015, 6:30 PM | Meetup
Digital Garage Development LLC 717 Market Street · San Francisco, CA Main Talk: Globally Scalable Web Document Classification Using Word2Vec Abstract Extracting information from unstructured web documents is a common problem for many applications and determining which category they belong to can be especially challenging at planetary scale. In this talk, we will show how SmartNews achieves globall
自然言語処ç†ã®æœ€æ–°æ‰‹æ³•"word2vec"ã§è‰¦ã“ã‚ŒåŠ è³€ã•ã‚“ã‹ã‚‰ä¹³ã‚’引ã„ã¦ã¿ã‚‹ - ã‚ã‚“ã¡ã¹ï¼
æ¦‚è¦ ã“ã®è¨˜äº‹ã¯è‡ªç„¶è¨€èªžå‡¦ç†ã¨ã„ã†åˆ†é‡Žã®æœ€æ–°æ‰‹æ³•word2vec を利用ã—ã¦èª°ã§ã‚‚éŠã¹ã‚‹ã‚ˆã†ã«ã™ã‚‹ãŸã‚ã®æ‰‹é †ã‚’説明ã™ã‚‹ã‚‚ã®ã§ã™ã€‚ word2vecを利用ã™ã‚‹ã¨æ„味ã®è¨ˆç®—ãŒå®Ÿç¾ã§ãã¾ã™ã€‚ 例ãˆã°"king"ã‹ã‚‰"man"を引ã„ã¦"woman"を足ã™ã¨"queen"ãŒå‡ºã¦ããŸã‚Šã€ "æ±äº¬"ã‹ã‚‰"日本"を引ã„ã¦"フランス"を足ã™ã¨"パリ"ãŒå‡ºã¦ãã‚‹ã¨ã„ã†é¢ç™½ã„手法ã§ã™ã€‚ 自然言語処ç†ã¨ã¯äººé–“ãŒæ—¥å¸¸çš„ã«ç”¨ã„る自然言語をコンピュータã«å‡¦ç†ã•ã›ã€ 翻訳やè¦ç´„ã€æ–‡å—入力支æ´ã‚„質å•å¿œç”システムを作るãªã©ã«æ´»ç”¨ã•ã‚Œã¦ã„る分野ã§ã™ã€‚ 自然言語処ç†ã¨è¨€ã†ã¨è€³æ…£ã‚Œãªã„言葉ã‹ã‚‚ã—ã‚Œã¾ã›ã‚“ãŒã€ 実ã¯æ¤œç´¢ã‚„推薦ãªã©ã§ç§ãŸã¡ãŒæ—¥å¸¸çš„ã«åˆ©ç”¨ã—ã¦ã„ã‚‹ãªã˜ã¿æ·±ã„技術ã§ã‚‚ã‚ã‚Šã¾ã™ã€‚ 自然言語処ç†ã®é©ç”¨ç¯„囲やè¦ç´ 技術ã¯å¹…広ã„ã®ã§ã™ãŒã€ ãã®ä¸ã§ã‚‚word2vecã®ç‰¹è‰²ã¯ã€ 冒é ã§ã‚‚挙ã’ãŸã‚ˆã†ã«ã€Œæ„味ã®è¨ˆç®—ã€ãŒå‡ºæ¥ã‚‹ã“ã¨ã§ã™ã€‚ ã“ã‚Œ
自然言語処ç†ã‚’ãªã«ã‚‚知らãªã„ç§ãŒword2vecを走らã›ã‚‹ã¾ã§ - 最尤日記
googleã®ä¸ã®äººãŸã¡ãŒä½œã£ãŸword2vecã¨ã„ã†ãƒ¢ãƒŽãŒã‚ã‚Šã¾ã™ã€‚deep learningを自然言語(N-gram?)ã«é©ç”¨ã™ã‚‹ã“ã¨ã«ã‚ˆã‚Šå˜èªžã‚’100次元ãらã„ã®ãƒ™ã‚¯ãƒˆãƒ«ç©ºé–“ã«ãƒžãƒƒãƒ—ã™ã‚‹ç‰©ã ã¨æ€ã„ã¾ã™ã€‚é¢ç™½ã•ã¯ä»¥ä¸‹ã®ãƒ™ãƒ¼ã‚¸ã®é€šã‚Šã§ã™ãŒã€ãŸã£ãŸã“ã‚Œã ã‘ã®äº‹ã§ã€ã»ã¨ã‚“ã©æ„味ç†è§£ã®ä¸€æ©æ‰‹å‰ã¾ã§åˆ°é”ã—ã¦ã„ã‚‹ã¨æ€ã„ã¾ã™ã€‚ Taku Kudo : word2vec ã§å°‘ã—éŠã‚“ã§ã¿ãŸã€‚ã„ã‚ゆる deep… é¢ç™½ã„ã®ã¯ã€2ã¤ã®ãƒ™ã‚¯ãƒˆãƒ«ã®å·®ãŒã€2ã¤ã®å˜èªžã®é–¢ä¿‚をよãè¿‘ä¼¼ã—ã¦ãれるã“ã¨ã€‚ （ä¸ç•¥ï¼‰ A B C → X (A → Bã®é–¢ä¿‚ã«å¯¾ã—〠C → X ã«å½“ã¦ã¯ã¾ã‚‹Xを探ã™) グーグル ヤフー トヨタ → 日産 渋谷 新宿 æœå¹Œ → æ—å· è¦å¯Ÿ 泥棒 æ£ç¾© → ãã 平和 戦争 å·¦ → å³ ç¤¾å“¡ 会社 生徒 → å°å¦æ ¡ 空 æµ· 天井 → åºŠæ¿ ç”Ÿãã‚‹ æ»ã¬ å‹•ã → æ¢ã¾ã‚‹ ・・・ Deep-le
1
ランã‚ング
ランã‚ング
メンテナンス
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
word2vecã¨PMIãŒç‰ä¾¡ - Kazuma Hashimoto (橋本和真)
Improving Distributional Similarity with Lessons Learned from Word Embeddings · Issue #134 · arXivTimes/arXivTimes
分散表ç¾ã«é–¢ã—ã¦æ€ã†ã“㨠- Kazuma Hashimoto (橋本和真)
Google Code Archive - Long-term storage for Google Code Project Hosting.
{{#tags}}- {{label}}
{{/tags}}