Upcoming SlideShare
×

# n-gramコーパスを用いた類義語自動獲得手法について

10,082 views
9,887 views

Published on

Published in: Technology
10 Likes
Statistics
Notes
• Full Name
Comment goes here.

Are you sure you want to Yes No
• Be the first to comment

Views
Total views
10,082
On SlideShare
0
From Embeds
0
Number of Embeds
329
Actions
Shares
0
36
0
Likes
10
Embeds 0
No embeds

No notes for slide

### n-gramコーパスを用いた類義語自動獲得手法について

1. 1. n-gram 1
2. 2. NLP 2
3. 3. ✓ n-gram - n-gram - n-gram ✓ n-gram - - ✓ Ameba n-gram - n-gram - - 3
4. 4. n-gram 4
5. 5. N n-1 n .... 5
6. 6. N n-1 n .... _ _ _ _ 6
7. 7. n 2-gram _ _ _ _ _ _CD _ _ _ _ 7
8. 8. n 2-gram _ _ _ _ _ _CD _ _ _ _ 8
9. 9. n 2-gram _ _ _ _ _ _CD _ _ _ _ 9
10. 10. n 2-gram _ _ _ _ _ _CD _ _ _ _ 10
11. 11. 11
12. 12. n-gram 12
13. 13. 13
14. 14. GSK ※ n MeCab 14
15. 15. 15
16. 16. ※ n MeCab ipadic 16
17. 17. 17
18. 18. 18
19. 19. 19
20. 20. ✓ n-gram - n-gram - n-gram ✓ n-gram - - ✓ Ameba n-gram - n-gram - - 20
21. 21. Baidu 21
22. 22. 22
24. 24. 24
25. 25. • • Baidu 25
26. 26. − − NG 26
27. 27. − − 27
28. 28. _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _2038_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 28
29. 29. _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _2038_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 29
30. 30. 1. w v ※ n-gram 2. w v Weight 3. wi wj Measure 30
31. 31. n-gram n-gram n > 1) 1-gram 31
32. 32. Weight freq(w) w freq(v) v freq(w,v) w v 32
33. 33. Measure [2010 ] Simpson Jaccard Simpson-Jaccard ※Vi Vj wi,wj ※Simpson Jaccard 33
34. 34. [2010 ]http://nlp.nagaokaut.ac.jp/arc/10/10NLP-asakura.pdf 34
35. 35. 35
36. 36. 10 ... 36
37. 37. 1. 2. Simpson-Jaccard [ 2010] 3. 2+ [ 2010] • • 37
38. 38. 38
39. 39. 39
40. 40. • → → • 5 9 • 40
41. 41. 10 × [ ] → → ..... 100 41
42. 42. 42
43. 43. • • • • BootStrap 43
44. 44. 44
45. 45. - FX 0.0701096506659567 - 0.06259698742992559 - 0.06042366216085525 - 0.058799844151470174 - 0.05735695738694635 - 0.055435815549429605 - 0.05232985446935419 - 0.04980454574094188 - 0.04927509063333216 - 0.04824717092086649 45
46. 46. - 0.13373671096183018 - 0.0918404883164821 - 0.08504703320048444 - 0.07961197269662577 - 0.07711954365644495 - 0.07702996997666198 - 0.07662113919947415 - 0.0725147419575286 - 0.06882302518118967 - 0.06849300043195472 46
47. 47. - 0.46469658771106315 - 0.11975034678985662 - 0.1097484574232766 - 0.09879163577196441 - 0.09813513570488815 - <EMOJI_546> 0.09719796981762122 - 0.09570307958644202 - 0.09472409933583924 - □ 0.09466494353310652 - 0.0939216442782036 47
48. 48. - 0.1315343475643398 - 0.09995353106829284 - <EMOJI_195> 0.07684385370515374 - 0.0668208753259571 - 0.064557866506281 - 0.06399098585323154 - 0.0631873841230999 - 0.059873448412962174 - 0.058499392875879214 - 0.05843526443912576 48
49. 49. - 0.05180321648370331 - 0.04568992635712712 - 0.03513069449040805 - 0.027690371652517847 - 0.024491134248894916 - 0.024070717542493413 - 0.013918762159274874 - <EMOJI_1BE> 0.013795160758545503 - 0.011283944598435737 - 0.009334867761007871 49
50. 50. • • 50
51. 51. 51
52. 52. Baidu n-gram UI 1. 2. 3. 2. Trie Double Array Trie Tree 4. Trie Common Preﬁx Search 52
53. 53. 53
54. 54. Amazon EC2 High Performance Computing(HPC) 2010/07/13 1.6\$ / hour ( 150 54
55. 55. Amazon EC2 HPC 4GB 55
56. 56. ✓ n-gram - n-gram - n-gram ✓ n-gram - - ✓ Ameba n-gram - n-gram - - 56
57. 57. Ameba n-gram 57
58. 58. Ameba n-gram n 7 MeCab ipadic 1 122,500 / 2 1,195,628 3 1,586,205 / 4 1,021,938 5 499,788 / 6 231,064 7 117,515 58
59. 59. n 59
60. 60. • • • • (<EMOJI_xxx>) • 7,500 • Ameba 60
61. 61. • MeCab ipadic • Hadoop • Map/Reduce 5 • -gram • • → • → • Unicode NFKC • • 61
62. 62. Unicode ※ • • • • 62
63. 63. • • 63
64. 64. 64
65. 65. <EMOJI_031> - <EMOJI_123> 0.25245663283888076 <EMOJI_031> - 0.23865945647496203 <EMOJI_031> - 0.2233787922434289 <EMOJI_031> - <EMOJI_139> 0.21505030767963476 <EMOJI_031> - <EMOJI_320> 0.20905272608162664 <EMOJI_031> - <EMOJI_176> 0.20767832299973635 <EMOJI_031> - ww 0.20396530274266259 <EMOJI_031> - <EMOJI_035> 0.1952646415960877 <EMOJI_031> - <EMOJI_029> 0.19118380193173534 <EMOJI_031> - 0.18532801238518545 65
66. 66. <EMOJI_022> - 5.05433098481788 <EMOJI_022> - <EMOJI_106> 0.21255982715221317 <EMOJI_022> - <EMOJI_178> 0.20371824309787556 <EMOJI_022> - <EMOJI_105> 0.19694261277461542 <EMOJI_022> - <EMOJI_324> 0.1960739883132334 <EMOJI_022> - <EMOJI_028> 0.19591157145392102 <EMOJI_022> - <EMOJI_143> 0.19367459051404434 <EMOJI_022> - <EMOJI_089> 0.1894972913601391 <EMOJI_022> - <EMOJI_319> 0.1853140205685605 <EMOJI_022> - <EMOJI_187> 0.18316119166397835 66
67. 67. <EMOJI_021> - 0.1980766824375306 <EMOJI_021> - 0.160789228797033 <EMOJI_021> - <EMOJI_143> 0.15832172025846883 <EMOJI_021> - <EMOJI_325> 0.15752899334689507 <EMOJI_021> - 0.15600103135195192 <EMOJI_021> - <EMOJI_089> 0.15588927293536636 <EMOJI_021> - <EMOJI_144> 0.15288842785042273 <EMOJI_021> - <EMOJI_141> 0.15264128047608577 <EMOJI_021> - <EMOJI_022> 0.1524110403297163 <EMOJI_021> - 0.1520281112500744 67
68. 68. <EMOJI_077> - 0.20100120017122702 <EMOJI_077> - <EMOJI_075> 0.16814782000472359 <EMOJI_077> - <EMOJI_188> 0.16407662484190644 <EMOJI_077> - <EMOJI_134> 0.1598560562097877 <EMOJI_077> - 0.15753861586037332 <EMOJI_077> - <EMOJI_007> 0.1520971768068388 <EMOJI_077> - <EMOJI_014> 0.15101611087607572 <EMOJI_077> - <EMOJI_162> 0.15081721574211884 <EMOJI_077> - <EMOJI_044> 0.14580103348348722 <EMOJI_077> - 0.13778630432134603 68
69. 69. <EMOJI_181> - <EMOJI_150> 0.0998223599336455 <EMOJI_181> - <EMOJI_169> 0.09619906986114168 <EMOJI_181> - 0.09407484828035773 <EMOJI_181> - 0.0751037382171799 <EMOJI_181> - <EMOJI_008> 0.06918233066750942 <EMOJI_181> - 0.05804747259025906 <EMOJI_181> - 0.05277048261006679 <EMOJI_181> - 0.05059336124231738 <EMOJI_181> - 0.049448742078198986 <EMOJI_181> - <EMOJI_113> 0.04768875635910332 69
70. 70. 70
71. 71. Markov Clustering Algorithm ( = ≒ ) ※MCL:http://www.micans.org/mcl/ 71
72. 72. ( <> < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < <end> < > < > < > < > > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > <CD> < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < < > < > <CD> < > > < > < > < > < > < > < > < > < > < > < > < > < > < > < < > < > < > > < > < > <CD> < > < > < > < > < > < < > < > < > > < > < > < > < > < > <NEW> < > < > <NEW> < > < > < > < > < > < > 72
73. 73. < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > <NEW> < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > < > 73
74. 74. • • 74
75. 75. 75
76. 76. (` ω ) (` ) - 0.1768694301100037 (` ) - ⊃ 0.09913242362438676 (` ) - ({( 0.09405018119300235 (` ) - 0.031062102211042445 (` ) - 0.028857016807085342 (` ) - 0.02608930784914413 (` ) - σ 0.022498088286870093 (` ) - 0.016193426515270032 (` ) - ω 0.016122858488317616 (` ) - 0.01345138088760404 76
77. 77. (^∀^) (^∀^) - 0.06916344408426699 (^∀^) - 0.06890434881805171 (^∀^) - 0.06795858079282872 (^∀^) - 0.05823929001282186 (^∀^) - 0.05591209864361552 (^∀^) - 0.055199102468491 (^∀^) - 0.0529766713691993 (^∀^) - 0.04651788654030653 (^∀^) - 0.04506786291842584 77
78. 78. (*́д`) (*́д`) - 0.11168727359492218 (*́д`) - 0.10079146007557346 (*́д`) - 0.09144542975809382 (*́д`) - 0.08668152231221557 (*́д`) - 0.0828216114116598 (*́д`) - 0.07886784003012988 (*́д`) - 0.07516063088790395 (*́д`) - д 0.04393643491834669 (*́д`) - 0.04391789518745296 (*́д`) - 0.041770270015884854 78
79. 79. (/ω) (/ω) - 0.16208491557582996 (/ω) - /// 0.12872038557250623 (/ω) - 0.1280711164456965 (/ω) - 0.11152828905321341 (/ω) - 0.09964963425187295 (/ω) - 0.09765195595553885 (/ω) - 0.09551633971030135 (/ω) - 0.08766541396665446 (/ω) - 0.0857860000269694 (/ω) - 0.07871911676808412 79
80. 80. (^0^)/ (^0^)/ - 0.09445225976273072 (^0^)/ - 0.036873637365873665 (^0^)/ - 0.026846223878548046 (^0^)/ - 0 0.024911172513660745 (^0^)/ - 0.019634091253330398 (^0^)/ - 0.01680880907554066 (^0^)/ - ^ 0.016789481994256183 (^0^)/ - 0.016144113065194438 (^0^)/ - 0.015391342223916664 (^0^)/ - .01494851710964888 0 80