1
00:00:01,130 --> 00:00:04,170
- [Instructor] So back in
lesson 11 we focused on

2
00:00:04,170 --> 00:00:07,950
natural language processing
and showed all sorts of

3
00:00:07,950 --> 00:00:12,410
different ways to look at
natural language for things like

4
00:00:12,410 --> 00:00:17,410
sentiment and to find
things like parts of speech

5
00:00:17,440 --> 00:00:21,910
and to help applications
understand natural language.

6
00:00:21,910 --> 00:00:24,090
The Natural Language Understanding service

7
00:00:24,090 --> 00:00:27,740
from Watson does all of
that and a whole bunch more

8
00:00:27,740 --> 00:00:29,610
in a prepackaged manner.

9
00:00:29,610 --> 00:00:32,670
Plus you can train the service yourself

10
00:00:32,670 --> 00:00:36,460
using their tool called
Watson Knowledge Studio.

11
00:00:36,460 --> 00:00:39,860
So not only is this natural
language understand API

12
00:00:39,860 --> 00:00:42,310
capable of looking at
things like sentiment

13
00:00:42,310 --> 00:00:44,670
and emotion inside of text.

14
00:00:44,670 --> 00:00:47,510
It will find key words
in the text and rank them

15
00:00:47,510 --> 00:00:50,830
by their relevance compared
to other key words throughout

16
00:00:50,830 --> 00:00:54,220
the text, it can find
things like people, places

17
00:00:54,220 --> 00:00:58,360
and organizations, so we
did the entity recognition,

18
00:00:58,360 --> 00:01:00,620
named entity recognition
back in the natural

19
00:01:00,620 --> 00:01:02,890
language processing lesson.

20
00:01:02,890 --> 00:01:06,190
It can also represent,
or not represent but

21
00:01:06,190 --> 00:01:10,240
locate things like categories
and concepts within the text.

22
00:01:10,240 --> 00:01:12,700
So it can look for
example at a news article,

23
00:01:12,700 --> 00:01:16,230
and figure out that it's about
sports or its about politics.

24
00:01:16,230 --> 00:01:18,060
And just like we were able to do

25
00:01:18,060 --> 00:01:21,010
back in the natural
language processing chapter

26
00:01:21,010 --> 00:01:24,590
it can also pick off
parts of speech as well.

27
00:01:24,590 --> 00:01:27,390
So let me just show you
that demo quickly here,

28
00:01:27,390 --> 00:01:31,300
and again the URL is at the top
of the screen at the moment.

29
00:01:31,300 --> 00:01:32,750
Opps, there we go.

30
00:01:32,750 --> 00:01:36,120
So as I scroll down here
they have a pre-populated

31
00:01:36,120 --> 00:01:39,360
text box here with the
text of a news article,

32
00:01:39,360 --> 00:01:40,850
and if you scroll down further

33
00:01:40,850 --> 00:01:42,890
you'll see this analyze button.

34
00:01:42,890 --> 00:01:45,340
So if you'd like you can
read through this text

35
00:01:45,340 --> 00:01:48,580
and then go ahead and click
analyze on your own as well.

36
00:01:48,580 --> 00:01:51,030
But it will submit that to
the server and then it's

37
00:01:51,030 --> 00:01:54,420
going to give back and analysis
down at the bottom here.

38
00:01:54,420 --> 00:01:57,260
This particular article
has an overall sentiment

39
00:01:57,260 --> 00:01:58,470
that was negative.

40
00:01:58,470 --> 00:02:01,020
If you go to the emotion
tab it'll tell you

41
00:02:01,020 --> 00:02:03,680
the various emotions
that it thinks it found

42
00:02:03,680 --> 00:02:05,040
within the text.

43
00:02:05,040 --> 00:02:07,120
And the,

44
00:02:07,120 --> 00:02:09,270
likely hood of those emotions.

45
00:02:09,270 --> 00:02:13,200
It will tell you what the key
words were within the text.

46
00:02:13,200 --> 00:02:15,750
What entities were found within the text

47
00:02:15,750 --> 00:02:18,830
so it's finding geographic
features and people

48
00:02:18,830 --> 00:02:20,100
in this case.

49
00:02:20,100 --> 00:02:23,530
Categories that this
particular article was about.

50
00:02:23,530 --> 00:02:26,590
So it has a hierarchical
category structure

51
00:02:26,590 --> 00:02:28,840
and more information as well.

52
00:02:28,840 --> 00:02:32,230
You can look at the syntax
of the text and the semantic

53
00:02:32,230 --> 00:02:34,530
roles in the text as well.

54
00:02:34,530 --> 00:02:36,880
Lots of different information
and if you wanna see

55
00:02:36,880 --> 00:02:39,490
how that comes back to you,
you can go ahead and click

56
00:02:39,490 --> 00:02:42,620
the Json link to look at
the actual Java Script

57
00:02:42,620 --> 00:02:45,580
object notation that the service returns.

58
00:02:45,580 --> 00:02:47,450
So if we wanna do something simple here

59
00:02:47,450 --> 00:02:49,673
we could say you know, something like,

60
00:02:50,607 --> 00:02:53,763
the Avengers movie,

61
00:02:55,600 --> 00:03:00,180
if I can type that would be a
lot better, movie was awesome.

62
00:03:00,180 --> 00:03:03,000
Exclamation point, and I can
go ahead and analyze that

63
00:03:03,000 --> 00:03:07,390
and you can see that it's
a 98% positive sentiment.

64
00:03:07,390 --> 00:03:12,390
For my emotion I see 94%
accuracy with joy here,

65
00:03:13,318 --> 00:03:16,940
Avengers, movie, was a
key word within there.

66
00:03:16,940 --> 00:03:20,360
Entities in this case
it didn't return any.

67
00:03:20,360 --> 00:03:22,440
I'm surprised by that,
I would have thought

68
00:03:22,440 --> 00:03:24,890
it would've returned the movie itself.

69
00:03:24,890 --> 00:03:27,330
But again, categories, it figured out that

70
00:03:27,330 --> 00:03:28,923
I was talking about movies.

71
00:03:31,810 --> 00:03:33,790
I'm not sure exactly,
oh actually these are

72
00:03:33,790 --> 00:03:36,200
characters within the movie and

73
00:03:36,200 --> 00:03:38,070
other information as well so,

74
00:03:38,070 --> 00:03:40,920
really an interesting
service that gives you back

75
00:03:40,920 --> 00:03:43,610
a tremendous amount of information

76
00:03:43,610 --> 00:03:45,350
about natural language text,

77
00:03:45,350 --> 00:03:47,280
and the reason that
they're able to do that

78
00:03:47,280 --> 00:03:50,433
is that they've trained
their models on massive

79
00:03:50,433 --> 00:03:55,433
amounts of textual information
that has been categorized

80
00:03:55,540 --> 00:03:58,290
for you so that you can take advantage of

81
00:03:58,290 --> 00:04:00,430
those pre-existing models and again

82
00:04:00,430 --> 00:04:02,503
you can train your own as well.