Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
文献紹介 8月10日
Search
gumigumi7
August 10, 2018
0
120
文献紹介 8月10日
Improving Distributional Similarity
with Lessons Learned from Word Embeddings
gumigumi7
August 10, 2018
Tweet
Share
More Decks by gumigumi7
See All by gumigumi7
文献紹介 1月24日
gumigumi7
0
230
文献紹介 11月7日
gumigumi7
0
120
文献紹介 10月3日
gumigumi7
0
320
文献紹介 9月3日
gumigumi7
0
250
文献紹介 7月16日
gumigumi7
0
260
文献紹介 6月12日
gumigumi7
0
330
文献紹介 5月16日
gumigumi7
0
180
文献紹介 4月18日
gumigumi7
0
140
文献紹介 12月15日
gumigumi7
0
110
Featured
See All Featured
The Invisible Customer
myddelton
114
12k
Docker and Python
trallard
35
2.7k
It's Worth the Effort
3n
180
27k
From Idea to $5000 a Month in 5 Months
shpigford
378
45k
The Art of Programming - Codeland 2020
erikaheidi
43
12k
The MySQL Ecosystem @ GitHub 2015
samlambert
244
12k
A Modern Web Designer's Workflow
chriscoyier
689
190k
The Invisible Side of Design
smashingmag
294
49k
Pencils Down: Stop Designing & Start Developing
hursman
117
11k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
126
32k
Mobile First: as difficult as doing things right
swwweet
217
8.6k
Designing on Purpose - Digital PM Summit 2013
jponch
111
6.5k
Transcript
( Improving Distributional Similarity with Lessons Learned from Word Embeddings
▪ ▪ Omer Levy, Yoav Goldberg, Ido Dagan.
▪ Improving Distributional Similarity with Lessons Learned from Word Embeddings. ▪ Transactions of the Association for Computational Linguistics. 2015. ▪ ▪ , Skip-gram, LSA 2
▪ , pi ur nu p i S a
clulroug a S ▪ ) ▪ pi u tea , pi pi a a 3
▪ V ▪ 2 count predict ▪ Don’t count,
predict! (Baroni et al., 2014) ▪ Word2Vec Skip-gram Shifted-PMI ▪ Neural word embeddings as implicit matrix factorization (Levy and Goldberg, 2014) 4
▪ 2 ▪ hyperparameter ▪ Word2Vec 5
▪ ))( (- 2 ▪ )( D L AW
P ▪ 2 - - 2 ▪ - - 2 - 2 VI M W ▪ GW))( (- 2 V M P ▪ ▪ 6
▪ 2 ▪ C D 7 The quick brown
fox jumps over the lazy dog 4 4 3 4 2 4 1 4 4 4 3 4 2 4 1 4 Word2Vec : 1 1 1 2 1 3 1 4 1 1 1 2 1 3 1 4 Glove :
▪ ▪ '()&$,.-/+ ▪ #
# % # ▪ !"$) 8 ! = 1 − % & &: , %:
▪ ▪ R G MR ▪ - - S
▪ Shifted PMI ▪ - 2 PID ▪ −log(&) W S NV 9 ( ) *( + = -./ 0, 2 − log(&) 3--./ 0, 2 = max(-./ 0, 2 − log(&), 0)
▪ 7 0 7 577 2 ▪ 7 0.SW
0 0 5 SV I ca db S W M b e P ▪ V S ▪ α WD C 10
▪ ▪ 2 G SV SV D ▪ G
D AC 11
▪ ▪ E D ▪ E D C D
12
▪ ▪ ▪ 2 13
14
▪ ▪ PPMI-Matrix ▪ SVD ▪ SGNS (Word2Vec)
▪ Glove ▪ ▪ 6$-"!#% ▪ &,-" 2$-"!#% ▪ '*(-")$+ ▪ 672)$+
16 ▪ # "!# ▪ SVD
17 ▪ )%+ '*(-$ ▪ ▪
&,"-$# SGNS
▪ - Se N ▪ - - - n
y bc ▪ N smp n y i ▪ G tu d H uo a 18