0
1
00:00:00,090 --> 00:00:00,440
All right.
1

2
00:00:00,450 --> 00:00:01,860
Welcome back.
2

3
00:00:01,920 --> 00:00:08,520
In this module, we're going to be building something really exciting. And in order to do it, we're leveraging
3

4
00:00:08,520 --> 00:00:15,110
one of the newest APIs that has come out in the IOs 12 bag of goodies that we got from Apple.
4

5
00:00:15,240 --> 00:00:23,100
And this is, of course, CoreML 2. Now, CoreML 2 gives us access to the same machine learning framework
5

6
00:00:23,370 --> 00:00:29,760
that's used across the Apple products including what they have inside Siri, inside the Camera, inside
6

7
00:00:29,760 --> 00:00:30,800
Photos.
7

8
00:00:30,900 --> 00:00:37,230
And it basically allows you to utilize all of the hard work that Apple has put into their own machine
8

9
00:00:37,230 --> 00:00:42,540
learning models and place this directly inside your own apps.
9

10
00:00:42,540 --> 00:00:49,650
Now, the part of this that I was most excited about is that they've now included inside CoreML 2
10

11
00:00:50,010 --> 00:00:53,110
a Natural Language processing framework.
11

12
00:00:53,160 --> 00:01:00,330
So this allows you to analyze text and speech and use machine learning models to understand what it
12

13
00:01:00,330 --> 00:01:01,650
actually means.
13

14
00:01:01,650 --> 00:01:08,070
So this is what we're going to be using. We're going to be creating our very own Natural Language processing
14

15
00:01:08,370 --> 00:01:14,000
machine learning model and we're going to be training it up using Create ML, and then we're going
15

16
00:01:14,000 --> 00:01:17,730
to be deploying it inside our very own app.
16

17
00:01:17,730 --> 00:01:20,180
Now, what is this app going to do?
17

18
00:01:20,190 --> 00:01:29,130
Well, recently, I was reading an article on Sentiment Analysis of Twitter Data for Predicting Stock Market
18

19
00:01:29,130 --> 00:01:30,060
Movements.
19

20
00:01:30,060 --> 00:01:36,270
And this was some research done at IIT and it's a really fascinating read, actually. I've included it
20

21
00:01:36,390 --> 00:01:39,630
as a link in the Course Resources of this module
21

22
00:01:39,630 --> 00:01:46,830
if you want to get some background info on the app that we're building. But, essentially, they looked at
22

23
00:01:47,100 --> 00:01:54,870
whether if you can go onto Twitter and search for something like, say, @CocCola, and if you take a
23

24
00:01:54,870 --> 00:02:03,240
look at the latest tweets that people have written about @CocaCola or trying to reply to Coca-Cola
24

25
00:02:03,330 --> 00:02:11,450
or talking about Coca-Cola, then you get a gauge on how people feel about a particular company,
25

26
00:02:11,460 --> 00:02:12,110
right?
26

27
00:02:12,120 --> 00:02:18,150
So, for examplee, this tweet says, how, you know, the negative impact on dental health. Because Coca-Cola is
27

28
00:02:18,150 --> 00:02:25,650
everywhere, so is diabetes. And this is an extremely negative view of Coca-Cola, right. Now if we create
28

29
00:02:25,890 --> 00:02:34,440
a machine learning model that can read or scan through this text and understand that the same way that
29

30
00:02:34,440 --> 00:02:42,780
humans do, such as you or I, being able to understand that this is a negative tweet. And having that negativity
30

31
00:02:42,810 --> 00:02:51,000
associated with a particular company, a particular brand, then some research suggests that it can be able
31

32
00:02:51,000 --> 00:02:56,040
to predict the stock market movements for that particular company.
32

33
00:02:56,040 --> 00:02:59,790
Now, here's a big disclaimer. I am not a financial advisor.
33

34
00:03:00,090 --> 00:03:01,890
I don't know anything about finance.
34

35
00:03:01,890 --> 00:03:04,710
This is not financial advice at all.
35

36
00:03:04,710 --> 00:03:10,950
So, now that we've gotten that over and done with, the interesting thing is is it possible for us to build
36

37
00:03:11,040 --> 00:03:17,490
a machine learning model that can understand the sentiment or the emotions that are being expressed
37

38
00:03:17,970 --> 00:03:25,590
in the text inside tweets, and then use that inside our app to be able to tap the pulse on Twitter of
38

39
00:03:25,590 --> 00:03:30,000
how people feel about a certain company, a certain topic.
39

40
00:03:30,000 --> 00:03:36,780
Now, by the end of this module, we would have built a very simple looking app, but it does some really
40

41
00:03:36,780 --> 00:03:38,010
powerful stuff.
41

42
00:03:38,280 --> 00:03:48,170
So I'll be able to put in a handle, for example. Let's see how people feel about at Apple. They feel kind
42

43
00:03:48,240 --> 00:03:50,610
kind of meh, not so great.
43

44
00:03:50,610 --> 00:03:58,890
We can also put in hashtags for certain stock symbols, so TWTR is the symbol for Twitter's stock.
44

45
00:03:58,890 --> 00:04:04,840
So let's go ahead and put that in here and let's see how do people feel about Twitter Stock.
45

46
00:04:04,950 --> 00:04:05,660
Pretty good.
46

47
00:04:05,700 --> 00:04:06,860
Well, it looks--
47

48
00:04:06,920 --> 00:04:08,920
Yeah, it looks pretty good as well.
48

49
00:04:08,930 --> 00:04:10,770
Now, what about something else?
49

50
00:04:11,070 --> 00:04:12,420
What about Snapchat?
50

51
00:04:16,440 --> 00:04:19,890
Less favorable. Or how do people feel about Tesla?
51

52
00:04:22,490 --> 00:04:23,380
Equivocal.
52

53
00:04:23,420 --> 00:04:29,630
So even though this app looks incredibly simple, but actually what it does is that whenever we put in
53

54
00:04:30,200 --> 00:04:34,090
any search query in here, for example, Coca-Cola,
54

55
00:04:34,580 --> 00:04:40,780
then it will go to Twitter, looking for tweets that have the handle Coca-Cola,
55

56
00:04:41,180 --> 00:04:48,890
and then it will look at the latest 100 tweets that all mention it and it'll filter them by English
56

57
00:04:48,890 --> 00:04:49,460
language.
57

58
00:04:49,460 --> 00:04:57,020
So only look for the English language tweets. And once it has done that, it will run all hundred tweets
58

59
00:04:57,170 --> 00:05:06,350
through a sentiment analysis, and it will determine whether if it's negative, positive, or neutral. And
59

60
00:05:06,350 --> 00:05:09,580
it does that for each and every single tweet.
60

61
00:05:09,620 --> 00:05:16,190
So then we end up with a hundred tweets that are either negative, positive, or neutral. And then we tally
61

62
00:05:16,190 --> 00:05:17,240
up the scores.
62

63
00:05:17,570 --> 00:05:22,170
So plus 1 for positive, minus 1, for negative, 0 for neutral.
63

64
00:05:22,170 --> 00:05:30,080
And we get a score of how a particular handle or search term performs in terms of the tweets sentiment,
64

65
00:05:30,440 --> 00:05:35,930
and then that gets interpreted as a particular emotion or sentiment.
65

66
00:05:35,930 --> 00:05:39,650
The stronger the sentiment, the more extreme the emoticon.
66

67
00:05:39,650 --> 00:05:46,880
So whereas a few years ago, there were companies charging millions for the ability to be able to analyze
67

68
00:05:46,880 --> 00:05:53,480
Twitter data for sentiment and use it to predict things such as stock prices or voting polls.
68

69
00:05:53,480 --> 00:06:02,150
We're going to do this with just a laptop, our knowledge of IOs development and CoreML 2 and 
69

70
00:06:02,180 --> 00:06:02,750
Create ML.
70

71
00:06:02,840 --> 00:06:08,940
So I'm super excited about this module because it's actually a really, really fun thing to build.
71

72
00:06:08,990 --> 00:06:12,950
So once you're ready, head over to the next lesson and we'll get started.