Up next in 10
AI in Power BI || Women Data Summit 2021
5K views
Nov 9, 2023
Are you interested in learning about AI Insights that are in Power BI Desktop? In this session, I will briefly review the collection of pre-trained models. I will then focus on text analytics and it’s library of functions: language detection, sentiment analysis, and key phrases. Conference Website: https://globaltechconferences.com/event/women-data-summit-2021/ C# Corner - Community of Software and Data Developers: https://www.c-sharpcorner.com C# Live - Dev Streaming Destination: https://csharp.live #AI #PowerBI #Azure #Data #WomenDataSummit #Synapse
View Video Transcript
0:00
Hi everybody. As Simon was saying, we're going to talk about AI in Power BI
0:08
And let me tell you a little bit about myself first, just so you're familiar
0:14
And as they were saying, I'm a business intelligence and ytics specialist at the University of Central Florida
0:20
We love our football team, Go Nights. I have a bachelor's degree in management and a master's degree in communication
0:28
Right now I have 10 years of experience. I began my career
0:34
I was working for a small startup business in 2010, and I first started working in T-Sql
0:41
which I really love, still love it, and I really like data warehousing
0:46
My new thing, as you can tell, is artificial intelligence. I love PowerBi and business intelligence strategy
0:53
So, all know, I just love data. Certifications, I have my certification, Power BI
1:01
which is a data yst associate, and I'm also certified in AI Fundamentals
1:05
which I'm gonna be speaking to you about today. And I am the leader of the Orlando Power BI user group
1:10
Our meetings are still virtual, so if you would like to attend, just go to Meetup.com and put a search for Orlando PowerBi user group
1:18
Or you can actually go to the Microsoft community site, go to user group, and you can find us there as well
1:24
So I'm looking forward to seeing you. Okay, so how can you connect with me
1:29
Please feel free if you have any questions and you just don't want to say it, you know, publicly
1:34
Please feel free to write me or connect with me on Twitter at IT Data Deva
1:40
Or you can connect with me on LinkedIn, similar name IT Data Deva there as well
1:46
And I'm looking forward to connecting with you. All right, so let's just talk about the agenda today
1:53
First, I just want to go over some, I would say, informative stuff that kind of gives you
2:00
an idea of what AI insights is so you can understand. And then we'll get into the fun stuff, the demos
2:07
I know everyone loves the demos, but I first want to cover some introductory stuff just
2:11
so you can understand how this all fits into the picture. All right, so let's just talk about that
2:18
We're going to list the elements of artificial intelligence, and then I'm going to define a
2:22
insights with you. We're going to discuss the three AI insights in Parachary. We'll
2:28
define some functions. We'll do the demo of Sentiment's ysis, which I know
2:33
everybody loves, I've done this presentation two times so far, and it's a very
2:38
popular thing. And then we'll go over review what we discussed today, and then we'll
2:43
have some Q&A. All right. So let's just discuss some of the artificial
2:50
intelligence elements. Okay. We've got, first let me list them and then I'll just discuss them
2:56
We have machine learning, anomaly detection, natural language processing, which is also known as NLP for the acronym
3:03
We have computer vision and conversational AI. Machine learning. What is that
3:09
It's basically using algorithms to teach a model to make better decisions from data
3:14
So it will improve after experience and learn from its mistakes. There are three types of machine learning you should know about
3:22
it's supervised, unsupervised, and reinforcement learning. And that's all you need to know for now
3:27
If you wanted to look into that further, if you're free to do a Google search
3:31
The next one is a nominee detection. This is a pretty popular one, I think
3:35
is really easy to understand. It just identifies atypical patterns and data
3:40
Example, think about credit card fraud. You know, someone's using your credit card
3:45
You know, for instance, me, I'm living in Florida. If I get, you know, a ring with someone
3:50
in another state is using it filled gas. That's kind of anomaly because I haven't been driving much anywhere
3:56
The next is natural language processing. This is basically the computer has the ability to understand written and spoken language
4:06
And the next one is computer vision. And this is basically the computer has the ability to see and understand pictures and videos
4:16
and which is processes. And the next one is conversational AI. And that is pretty much, I'm sure most people have an idea what this is, it's a software agent
4:25
a bot, which can engage in a conversation. So let's dig a little deeper now into machine learning
4:34
Here are some of the features. We have automated machine learning, we have Assure machine learning designer, data and compute
4:41
management, and pipelines. Don't worry about the last two too much. Just kind of focus right now on automated machine learning and assure our machine learning
4:50
designer. Automated machine learning pretty much enables non-experts to quickly create effective machine learning model from data. You will see this in Power BI. And the next
5:02
one is a short machine learning designer. This is pretty much a graphical interface
5:07
which enables no code development of ML solutions. I really like the Azure machine
5:13
learning designer. I've been playing with that a lot. And don't worry too much about, like I was
5:17
saying, about data and compute and pipeline. data and compute is basically just a cloud-based storage and compute resources that professional
5:26
data scientists can use to run data experiments code scale and pipelines
5:30
It's where data scientists, engineers can define them to orchestrate model, training, deployment, and management tasks
5:37
All right, so now that we discussed machine learning features, let's just discuss computer vision services
5:45
And I know this one is kind of playing this screen, compare the other ones, so I apologize
5:50
First thing I'm going to speak about is computer vision services. And computer vision is basically used to yze images and video to extract descriptions
6:02
tags, objects, and text. Basically, they're seeing the world and it's making sense of it
6:09
The next is computer vision. And I'm, excuse me, custom vision. And this one is, I really liked this one, I found this one really interesting
6:18
Use the service to train custom images, classification, odd detection models, using your own images
6:25
So you basically, if you want to take, you know, for instance, you're running grocery store
6:29
and you want the computer to recognize apples and bananas, you kind of stick a picture in there
6:33
and, you know, put apple on banana and it will recognize it for you
6:36
It's really cool if you have any time to play with it. You can go to the Azure Labs and they have some fun stuff in there
6:45
The next is Face. I know this is a real popular one and most people can guess just from the name face service
6:51
which enables you to build face detection and facial recognition solutions. It can basically recognize your face and I think you can even name your face if it's
7:02
see as you if you train it that way. And the next is forum recognizer, which you know, you won't have to worry about these in
7:10
Power BI, but this is basically used to extract information from scan forms and invoices
7:17
it's actually really neat. All right, so we just listed some computer vision services
7:24
So now I'm just going to move over to what we'll be focusing on, which is natural language processing
7:29
And as I said the acronym is NLP There four basically considered in here It speech which is used to recognize and synthesize speech
7:40
It used to translate spoken languages, and then we have language understanding intelligence service
7:46
You can train a language model can understand spoken or text-based commands
7:51
You have translator texts. It can do, I think, between more than 60 languages
7:56
And then you have text ytics, which is the big one, To me, that feels a big deal
8:02
It's giving a demo of the service to yze text documents, extract key phrases, detect entities
8:10
such as dates and peoples and can evaluate sentiment, which is what we'll be talking about today
8:15
Basically how positive or negative the sentiment is. All right, so we discussed natural language processing
8:23
Now I want to talk about the six principles for responsible AI
8:29
You may be thinking, why is she bringing this up? This is very important for Microsoft when I was earning my certification
8:37
It was definitely had a decent amount of questions on it, and I can see where they're coming from
8:42
So I do want to view this with everybody, and I'll kind of give you a little short story about
8:47
what happened in my organization with the data modeling. So we have privacy and security, accountability, liability and safety, transparency
8:58
inclusiveness and fairness. For privacy and security, you want to increase reliability
9:06
that data needs to keep that data source secure. You want to make sure that the data is not leaked
9:14
And then you want to have accountability. Designers and developers should work within a framework of governance
9:19
and principles and ensure that the solution needs ethical and maker standards
9:24
And you have reliability and safety. This is very important. ideas and values and principles. You don't want to create harm. You should reduce risk to human
9:34
life. Think here driving cars, self-driving cars, right? You have robots driving them. You want
9:40
these to be safe. You don't want to hear about accidents. So that's a very important one there
9:45
Transparency is something that I feel is another very, very important one. You should be open to how and why
9:51
what's going on behind the scenes. How is this data modeling coming up with this? That's very
9:58
important to know so you don't know then you can't understand and it's kind of creates like a black box right inclusiveness AI systems should empower and
10:09
engage people and then of course fairness you want to reduce fairness any kind
10:14
of stereotypes you want greater diversity of people deploying these AI systems
10:20
and it should be fair for everyone like I said and transparency is very important I was using sentiment ysis so while back ago and
10:28
And my supervisor know too much about it. And they want to use sentiment ysis to hold our employees possibly responsible for bad service
10:39
But the only thing is right now, sentiment ysis, as you will see, it's not 100% accurate
10:44
And you don't really want to rely on people's jobs with this sentiment ysis
10:50
It's not 100% reliable. So that's something really to look at. And, you know, I had to go back and really show them how the logarithms were going
10:58
so we wouldn't possibly hurt anybody. So that's just another important thing to know
11:05
So the next is about AI Insights. So it was released in June 2020
11:12
So it's been around for about a year before it was in a preview feature. So some of you may have seen it before then
11:18
It's used to gain access to a collection of pre-trained machine learning models
11:23
You can think of cognitive services, right? So as you see in my next image
11:27
text ytics, right, cognitive services. And then we have vision, which is cognitive services
11:33
And then we have Azure machine learning, which is invoked by Assure machine learning functions
11:40
the model. So someone would actually in Azure, create that machine learning model and then share it with you
11:45
Okay. So you may be thinking, how do you access these features
11:53
Well, first off, you must have premium. All notes and A3, EM3 and above are premium per user, which recently just came out
12:05
And for those of you who don't know, pretty much you can get a subscription on a user level
12:11
And if you're ready to have the pro, you're just going to be spending extra $10
12:16
For those who don't, it's extra $20 a month. And you can play with some of the AI features there, which is really awesome
12:22
And then in Azure, we have a cognitive services endpoint and key
12:26
endpoint and key. And also, you have to access to Azure Machine Learn models
12:33
It's available to all users. So in sure, if you only have pro
12:38
you can actually use cognitive services, and that's what I'm going to show you today
12:43
So let's talk about a little table that I created, you know, what I've just discussed with you
12:50
We have a few plans, which are premium, premium per user and pros, I was saying
12:55
And I just listed, do you have access to content services? Right, with premium, yes you do
13:00
You pay for that in your subscription. Premium per user you do as well
13:04
And pro, you don't have any usage of that. However, if there is a workaround, which I was telling you
13:11
you can see here, my tip, you will need a key and an endpoint from Azure
13:16
Over here for premium, a tip, the AI workload needs to be enabled for premium
13:22
So that's important to remember. All right, so let's talk about the AI Insight buttons
13:28
Where are they? Well, there are in Parquiry. And if you go to add column tab, you can see over to your right
13:34
AI Insights, we have Text ytics, Vision, and Assure Machine Learning. Then on below, you can find it on the other tab
13:42
which is the Home tab, and that is kind of listed vertically, which is Text ytics Vision and Assure Machine Learning
13:48
So that is where you can find them in two spots, which I think is awesome. Next, exactly
13:54
What is text ytic function? So let's look at this a little deeper
13:59
Okay, language detection returns the name and the ISO identify, for example
14:05
Spanish is ES or English is EN. We won't really be going over that today though
14:10
Then we have key phrase extraction. This is actually one of my favorites and I'll show you this in my demo
14:15
It pretty much evaluates unstructured text and returns a list of key phrases
14:20
It's a really neat when you see it. And then you have sentiment scarring
14:24
And as I was saying earlier, it returns positive or negative sentiment in social media, customer views, and forums
14:30
The scars are between 0 and 1 and 5 is in the middle, neutral
14:37
Okay, so now that we discuss tax ytics function, let's just briefly talk about custom vision function
14:46
So basically, the computer vision function, excuse me, tags images. It has output tags for more than 2,000 recognizable objects
14:54
Some examples are human, scenery, and actions. All right, so now I want to kind of discuss
15:06
the sure machine learning model You must have as I was saying before you must have read access to your Azure subscription And I provided the link for you to read about these roles in case you are not familiar with them
15:21
And that's all for about Azure machine learning models. Let's move over to sentiment ysis
15:27
Okay, as I was saying, it's an area of AI and then returns a score. So where can you get a scar from social media, right
15:35
Facebook or Instagram, you can get it from customer reviews, or you can even get it from
15:40
discussion forums. So there's many places where you can do the sentiment ysis. So I think on time
15:49
how am I doing on time? Is it 415? I think I have to stop
15:58
All right. So let me see if I could take a demo. I may have to share the screen again, but I
16:03
actually may run past this part. Okay, so let's go to the Azure portal
16:08
If I have some time, we'll go back to that later. I didn't want to share my account
16:14
because there's some personal stuff here. So what I did do was take some screenshots
16:19
You want to make sure, first off, that you have an account
16:23
And if you don't, you can go to Azure Microsoft.com, US, I think you may have to change the language
16:31
if you're in another country, and then just go to free. You create your resource or you can look for cognitive services
16:38
So let's just see, let's just click over here to Azure portal example
16:44
Okay, and as I was saying, you can select Create and you will see Text ytics
16:50
And over here, here's Create, so there's a few ways of going to it
16:57
And then once you're here, you're going to want to add the following information
17:01
going to want to add your subscription, your resource group, your region, and just be careful
17:07
with the region because I know once you pick it, if you have to change it again afterwards
17:12
there may be a costing curve with it, your name, and then of course your pricing tier
17:19
So you have to have that all set up. Now what will you need once you create your resource
17:25
The two important things you're going to need is your endpoint, which is the HTTP address at which your
17:30
services hosted and then you're going to need a key which is secret values by client
17:35
applications to authenticate themselves and please once you get your endpoint and key please keep it a
17:41
secret you don't want to share it publicly otherwise you would be charged for whoever uses your
17:45
service all right so you may be thinking okay well once I get this set up where can I test
17:52
where can I find data for sentiment ysis or any of the AI cognitive services well you can
17:58
go to my favorite is Kaggle and you can use this website to find and publish
18:03
datasets they have some great ones at Kaggle.com and or you can excuse me you can go
18:09
to UCI and that's another machine learning repository that data scientists or
18:15
ysts can use of machine learning algorithms I think they have about 58
18:20
datasets available for use and I provided the link here I suggest Kagal though I
18:25
really liked it the other UCI is good too but Kaggle is was one of my favorites
18:31
All right, so let's do the demo. So I'm going to briefly switch screens
18:39
so I think I have to stop sharing real quick. And then I am going to go over to PowerBi
18:47
Let me just grab that. All right. And I'm going to share my screen
18:58
Okay. Great. Okay
19:11
So let me just remove this. Let me just want to be these files later. Okay
19:19
All right. Let's open a Paracquiry. And can everyone see my screen in the Paracquiry
19:27
Now I'm switching. Yeah, don't. Okay, great. Don't pay attention to this error right now
19:34
I'm off my connection to my computer, so I don't have any VPN
19:39
That's the only reason why you're seeing that, but we don't need to see any of the data right now
19:45
What you just need to know is what to do. So the first thing I want to point out is
19:50
please remember to clean your data. That's one very important thing. If you have any blanks, it's not going to work
20:00
Your call won't work. So just make sure there's no blanks in your data. I usually like to create another table
20:06
That's one of the things I prefer to do. And just make sure that you have a unique identifier
20:12
And then, of course, you have your test. Your text, excuse me
20:16
So you're going to want to create a custom function first. And let's open, I think I have
20:24
Actually, let's go over to, I want to show you first because I can't open this and I don't want to switch back my screen
20:34
So let me close this and we will actually just look. Okay. And I'm going to show you this first and we'll go back and look at the custom functions since I don't want to switch in and out
20:43
But what you need to know is you need to create a custom function, which will integrate Power BI and the text ytics
20:50
And then the function is going to review the text as a parameter. and then it's going to parse the response from the API and then return a string that contains a common separate list
21:00
So then you're going to go to a new source, select a bank query, you're going to go to Vents Editor
21:04
and you're just going to put in the code that we talked about. So let's look at what I had actually downloaded
21:14
So this is Sentiment ysis, and I downloaded some data from Universal Studios
21:19
And we can see over here when I ran this, we have positive sentiment, neutral sentiment
21:26
and negative sentiment. And if you look over, this actually will total one to one two total only
21:32
So 0.95, 0, and then 0, which equals 1. The index over here is just my unique identifier, and then these are the comments
21:40
You can see some people are frustrated with Universal. You see Universal's complete disaster, stick with Disney
21:46
We went to Universal over Memorial Day weekend. It was a total train wreck
21:49
We needed to get in the parking lot for about 40 minutes. We paid for prime park and make up for all the wasted time
21:55
So you can see someone is very upset. What's the sentiment? 0.95
22:00
It's pretty right. It's, they're not happy. So let's see over here
22:06
Food is hard to get. The food service is horrible. I'm not reviewing the food
22:09
It's the wait time, 45 minute minimum to get the one cashier working at each place
22:15
It rooms the hall experience. So that's a 0.76. So you can see who is upset and I find some of these are actually pretty accurate
22:23
Let's look for one that's someone's happening. And says there's Disney, then there's Universal Landau
22:30
This has been a long time coming. I've really been coming to this park because I was a child after growing up and moving
22:34
to Florida, even working here for a year and being annual passholder
22:38
I can say I love this park. So they pretty right with this Sentiment negative is very low Positive is 0 and 0 So it might be seeing something else in there that makes it kind of towards the neutral but it mostly positive
22:53
But you can kind of see here, when people get upset, they'll mention Disney
22:57
you know, go to Disney instead. So let's look at the report
23:01
Over here is just the little bar chart that I did with the sentiment and you can see kind of along the days
23:07
you know, who had most of what. And then you over here, if you kind of wanna click
23:12
Let's say we want to do neutral. You can kind of get an idea here
23:18
and then we can click here for positive. We can kind of play with this
23:23
This is really informative and very helpful. And the next one I wanted to look at was key phrases
23:30
And what this does is it actually picks out the most important words it feels in the sentences
23:38
So this one I just took the positive reviews. kind of see the comments that come in the positive reviews and I find this really neat
23:45
you can see when people are happy is the universal what really attracts people is harry potter
23:51
people love um harry potter i think it's the biggest attraction to universal studios um fun family vacation
24:00
parks and you know if you click here you can see the comments universal florida great price great
24:06
experience so now you may be wondering okay what are you know what are the key phrases for negative
24:12
Well, when we click over here, you know, we can see, let's click here, rise
24:18
They complain about the way, and I can tell you, since I live in Orlando and I am an annual pass holder
24:24
when you do these parks, you really got to do your research and you got to strategize
24:29
And I recommend going early in the morning, going home, taking a break, taking a nap
24:33
and then coming back in the evening because it gets very busy towards, you know, those prime hours
24:39
And then it slows down again in the evening. And I can tell you if you go to Florida in July or August or June, it's hot and it's crowded and you're not going to be happy
24:48
And that's where you're going to see a lot of these reviews is rides, which is completely right
24:53
Those lines could be very, very long. And you can see people kind of complain about it over here
25:00
This is not vacation, worst experience I've ever had. Rides are outdated
25:04
Hopefully smells like vomit. It doesn't. awful very rude yellow workers. And the interesting part is if you click over here
25:13
you can see that most people will say, go to Disney. You know, Disney's better
25:20
They need take lessons from Disney. So you can see here that there was about 11 people
25:24
who, when they're mad at Universal, mention Disney. Disney would never do something like this
25:29
Hard customer service. So you can see people are very angry and Disney seems to be their thing to mention when they're not
25:36
upset. Another thing that they mentioned is service and also mentions tickets and the
25:45
line and I was telling you you have to kind of strategize for that and let's see
25:51
if there's anything else that let's see parts kind of small so it's probably not
25:56
one on that part customer service they do mention Harry Potter but there's not
26:02
too many of that about five so anyway that's kind of the key for
26:06
phrase elements, which I think are really, really fun, really cool. I really like the Word Cloud that I created
26:15
By the way, that's the visual that you'll have to get in the App Star
26:19
and you just search for WordCloud, and it comes right up, and I think it's a lot of fun to play with
26:25
I also think the sentiment ysis is there's a lot of things you can do here
26:30
and this tends to be a big favorite for people. All right, so now I'm going to stop sharing my screen and I'm kind of going to go back over to my presentation
26:43
Let me pull that up. All right. Share. Share
26:57
Okay, so we covered our demo
27:13
We spoke about transforming data and creating a function. Okay, here is the sample that I wanted to show you, but I didn't want to switch back and forth
27:23
You can also find this online, but this is the endpoints and key that you will need
27:31
And of course you can see that I left it blank. Don't worry about the pin you mining true here
27:37
That was something I was just experimenting with and I was trying to get to work. You don't have to just include that
27:42
As long as you have that API and then your endpoint, you're good
27:46
You can run this. And you can find this online also just use. You could Google Congress of Services and the Microsoft documents you can find it
27:52
you can find it okay um we discussed that we discuss the custom function okay and as i was saying
28:04
about um sentiment ysis about um the AI rules transparency um you don't want to make big decisions
28:13
off a sentiment ysis um that's very important um you want to consider other scenarios outside
28:19
of the product and service review domain so other things that have gone on and just be careful of the filters are removing content
28:28
Another thing I wanted to show you is just this table I've been creating going to share
28:35
it probably in the Microsoft community soon. I've created a table with text ytics and where you can go and all the links
28:44
I also have created some cognitive services for Azure, which I think is really neat to
28:51
gives you all the links where you can go. It talks about what are your cognitive services
28:56
what's new in cognitive services, and then DIGSGR Accountative Service blog. So I have a few things here that you can go to
29:04
I'm gonna publish these all eventually just have the time to take around to have been really busy
29:09
And the last one, which I think is most relevant right now to this presentation, is just some links I found
29:17
for Power BI using AI. So I have a few things. Sentiment ysis of penny mining
29:26
I have a link there. Assure cognitive services, support and help options
29:31
I also have user scenarios for the text ytics API as well
29:38
All right. So let's just talk about a review. So we've listed the elements of artificial intelligence
29:48
We also defined AI, We also discussed the three AI insights, excuse me, three AI insights in Paracquiry
29:59
We also defined the functions. We did a demo of sentiment ysis
30:05
We had a review and now we also have some time for QA where we can go over some things
#Computer Science
#Educational Software
#Machine Learning & Artificial Intelligence
#Software