You can add sentiment analysis to an app without training a single model
A hotel review that praises the location but complains about room service leaves people and applications unsure whether the customer was happy or not. Microsoft AI MVP and Valorem Reply technical manager Liji Thomas shows how pre-built cloud AI services let everyday developers handle that nuance with simple API calls. The question is how much these services can really understand, and how far a developer gets with no machine learning expertise.
See Azure language AI demoed live, from intent detection to opinion mining, without writing code
Liji Thomas, a Technical Manager at Valorem Reply and Microsoft MVP in AI, argues that pre-built cloud AI services let everyday developers add intelligence to applications without machine learning expertise. She walks through Azure Cognitive Services for language, including intent and entity recognition, sentiment analysis and opinion mining, and demonstrates them without code in Language Studio before closing on responsible AI principles.
“Because what gives us our intelligence is our ability to see, hear, speak, understand and make intelligent decisions.”
“Now imagine giving your apps, your applications that same ability. So this is the promise of Azure Cognitive Services.”
What you will learn
Identify the three barriers that most often block AI adoption: data, app development skills and machine learning expertise.
Explain how pre-built cloud AI services let developers add AI through simple API calls in C#, Java or Python.
Distinguish intent from entities when designing a chatbot that must understand many phrasings of one question.
Choose between plain sentiment analysis and opinion mining when analyzing mixed customer reviews.
Try Azure's language services without code in Language Studio using a free trial Azure subscription.
Key takeaways from the speaker
Learn the capabilities of the AI services that power Microsoft tools and products
Learn the use cases of Azure AI and who's making it real
Learn how to infuse intelligence into your applications using simple API calls
Details
Level
Intermediate
Format
Talk
Technical depth
Introductory technical
For
Early career, Mid-career
Length
19:47
Published
July 2022
Questions this talk answers
What stops organizations from infusing AI into their applications?
According to the research she presented, more than half of leaders cite inadequate data, a lack of app development skills and a shortage of data science or machine learning expertise.
What are cloud AI services?
They are pre-built, configurable AI models that any developer can invoke with simple API calls, deployable in the cloud or on the edge, with no data science skills needed to start.
How can a chatbot understand the same question asked in different ways?
Conversational language understanding identifies the shared intent behind different utterances, such as finding a session, and extracts the variable part, such as the session name, as an entity.
What is the difference between sentiment analysis and opinion mining?
Sentiment analysis returns a score between 0 and 1 for a piece of text, while opinion mining goes one level deeper with aspect-based sentiment that labels each target as positive or negative.
How can I try Azure language AI without writing code?
Sign up for an Azure subscription (a free trial with credits works), create a resource and use Language Studio at language.cognitive.azure.com to run the services on samples or your own text.
About this talk
With cloud AI services and tools, now every developer can leverage AI to create innovative applications that solve complex problems. Learn how to create the next generation of intelligent applications using Azure AI. You’ll be able to build apps on your terms- using the tools and frameworks of your choice and deploy your AI models anywhere. You can leverage prebuilt models or build custom models with code-first or code-free experiences. This session will introduce the Azure AI services to help you bring capabilities like speech, vision, language, decision, and search to your applications, no matter your skill level. Build apps that can see, hear, speak and understand- with zero machine learning expertise.
One intent, many questions: how chatbots understand (72 s)
Why mixed reviews need opinion mining (85 s)
Welcome to this session on applying AI with zero machine learning expertise. My name is LIJI THOMAS. I'm a manager with Valorum Reply. I'm also an MVP in AI. It's a Microsoft Most Valuable Professional in the AI space. My team and I have been working in this space for quite a few years and I'm happy to share my expertise and experience in this space with you.
So, in the interest of time, let's dive right in. So, we are at a stage in history where the opportunity with AI is now real. It's no longer a hype and industries and business organizations across the world have started to realize that at length and in quantifiable terms. So, apparently there is a lot of research and studies.
The one that you see on screen here is one by Forrester, which quantifies the benefits that AI will bring into every kind of application in terms of increased automation, business agility, revenue, customer experience, etc. Now, the question is, if all this is true, what is the friction or what are the challenges that organizations face with infusing AI into applications? At a very high level, more than 50% of leaders, tech decision makers believe that these are the major challenges that they are dealing with.
One, on the data side, they don't have the adequate data, the quality of data to, you know, build AI solutions, app development skills. It's one of the hardest, you know, skill sets that they require and they feel that their teams are not ready, their resources are not adequate. Data science skills are not adequate. Data science skills or machine learning expertise, as we call it, teams would require to have advanced machine learning expertise and knowledge of statistical models and mathematics to be able to build algorithms ground up.
So with three primary challenges facing them while they're building AI solutions, several leaders in this ecosystem have come up with what is now called as cloud AI services. Now, cloud AI services, the promises of cloud AI services is to provide pre-built AI services or models that any application developer or any, you know, everyday developers will be able to invoke using simple API calls.
You know, just like if you are familiar, just like if you are familiar with calling a REST API or you are familiar with working in a programming language of your choice, C#, Java, Python, you will be able to invoke AI services into your applications starting today. So though they are pre-built, they are configurable and you have ongoing support from their partners, you can deploy it on the cloud, you can deploy it on the edge, you know, they're very responsibly built and they're tested.
And these are services that you can actually trust. So you have a lot of benefits from these cloud AI services adoption as you see on screen. So it's quick to build AI solutions, you faster to market, faster to deploy, you've got better troubleshooting expertise, and you don't need to have machine learning expertise or data science skills to get started with or to use these services.
So Microsoft's offering in this space, before we go into that, I'm happy to tell you a little back history of story around this. So even before AI was a thing and started gaining all this traction, decades before Microsoft started investing in this space and we have, so that's basically based out of the Microsoft Research Centers across the globe.
And research, advanced research that was done in these research centers came up with these AI services that today have even achieved human parity. What that means is, say, for example, we talk about image captioning. Image captioning services that Microsoft has built, the Microsoft research teams have built, can caption an image as good as a human can. And these services, what they did with them initially is that they infused it into their own applications.
So if you're used to using any of the Microsoft applications on a daily basis, like Word or PowerPoint or Excel or Teams or whatnot. I mean, even Microsoft Edge for that matter, the image captioning feature that we're talking about right now is a part of Microsoft Edge. So not everyone adds alternative text to their images, but what Microsoft Edge does is it automatically captions a text, even if the author has not added an alternative text.
So in terms of accessibility, if someone's using a narrator to have that experience with browsing, you know, you don't have to rely on an author. The browser themselves because of this image captioning feature is able to automatically caption an image for you. So long story short, what they did with that after that is that they decided that they will open it up for the developer community, the very AI services that they use inside these products.
They will democratize it, they will democratize it, they will open it up so that you and I will be able to use those services and build AI applications. So before we go into that, I want to give you a very high level understanding of the landscape here. When we talk about AI services in Microsoft or in Azure specifically, there are various flavors.
One is this pre-built AI services, the one that we're talking about in this session. We also have other flavors which includes conversational AI and custom AI. So yes, you can build this using deep machine learning expertise from scratch. But if you don't want to and you want to get started today, you have the option of choosing Azure Cognitive Services.
So these are the pre-built AI services to be talking about. And that in a few lines of code with able to, you know, you'll be able to call, call invoke API services. And you're able to interact with applications just the way that a human would. Because what gives us our intelligence is our ability to see, hear, speak, understand and make intelligent decisions.
Now imagine giving your apps, your applications that same ability. So this is the promise of Azure Cognitive Services. And this is a huge, you know, the breadth of services and capabilities across each of this is very interesting and it's very, very powerful. But again, for the sake of time, we might just look at a few of this and I'd love to talk you through, you know, some of the capabilities on the language services.
Now, with language, what we would typically do is, you know, the ability to understand a human language or the ability to have an intelligent conversation with a human is a very powerful one. So imagine being able to talk to a machine or an application just the way that you and I talk. Right. So the the technology would be should be able to understand what I'm trying to ask or what I'm trying to say should be able to go find an appropriate response after that.
Maybe should be able to understand my sentiments or emotions, able to understand key phrases, entity, recognize entities in that translate my text if I'm talking in my native language. And, you know, the first the technology can help translate that across hundreds of languages and and provide me an accurate response. So these are the the several capabilities of the language service and we look at some of them in detail.
So let's say that we are trying to create a bot for the Women Tech Conference. Right. So one of the things that I could ask the bot is when is Women Tech 2020 happening? And it would say June 7 to 10. Great. So the catch with that is that each of us would ask that in different ways.
And that's the beauty of the human language. Right. So we do have a lot of nuances with the way that we we talk. And at a conversational agent or a chat bot, for example, should be is an application should be able to understand that irrespective of however you ask that the answer should be June 7 to 10.
So this is possible using the Azure Cognitive Service for language. And take it a step further. And sometimes it's important that you understand the intent or the meaning of what a piece of text is. So that's where conversational language understanding comes. So if you're familiar with natural language processing or NLP or NLU systems, you don't have to build them from scratch.
Now, you should be able to use these AI services, which will be able to give you the meaning of a piece of text in a single API call. For example, in the same chat bot example, you know, someone could ask, when is the session on AI? What are we doing right now? Is there a session on leadership?
Is there a session security on inclusion on imposter syndrome? We understand that all of these are very different questions. But how an app would deal with it is all of these irrespective of the different utterances that we've mentioned, you might note that every single speaker has the same intent here. The intention of the speaker is to find a session in the conference, right?
The difference of the variable there is the session name, which I would call an entity, right? So these kinds of nuances and things can actually be done using Azure Cognitive Service for language. Another thing that they could do is sentiment analysis and opinion mining. Now, this is interesting because these are actually two different things. So sentiment analysis is what we call fine-grained sentiment analysis.
It applies labels to a piece of text. It could even be an entire document, in fact. It could be a sentence, a paragraph, or an entire document. So if it's fed a line of text, it would come back with a single score between 0 to 1. The more it is towards 0, it is indicative of a negative sentiment.
The more it is towards 1, it is indicative of a positive sentiment. Somewhere in between is a neutral sentiment. So let's say a customer who's just back from a hotel vacation and passes this customer review says, I love the hotel location and the customer service was excellent. So what this API would do is it would return a value of 1, which is indicative of a positive sentiment.
Now, what's tricky is let's say the customer has entered something like this, which is most obvious in a practical scenario. This is what you would do. You would get a whole paragraph like you would see in Amazon reviews with a mix of, you know, sentiments of, I love the hotel location, but disappointed with the room service. Now, if that is the case, you know, typically people and even applications are left wondering if that was a positive sentiment or a negative sentiment.
So this is where you need to have, you know, better granularity and better control over the sentiment analysis here. So the hotel location was obviously a positive sentiment, but there was a negative attribution to the room service. Now, the promise of opinion mining, which is a feature of sentiment analysis itself, is that it goes one depth, one level deep.
So it is able to give you this kind of aspect based sentiment analysis. It will return value saying that, you know, what this part, the whole part was the whole sentence itself was a mixed review, but then this part was positive. This part was negative. It's important for, you know, people behind that application to know such important things, because then you know what to what your strengths are.
You know where your opportunities are as well. So it's very powerful to have those kinds of things in products and services. So if we have time, I'd love to show you a quick demo of some of these things in action. Obviously, you know, there's a little bit of technology that is involved here. But if you'd like, I would suggest that all that you would need to even try this out right now is an Azure subscription.
You can even sign up for a trial Azure subscription and you get some free credits along with that. And then what we need to do is go to this URL, which is language.cognitive.azure.com. So it's basically you just need to browse for Language Studio. Language Studio is a new offering wherein without writing a single piece of code, you can check out these services in action.
All of these that we discussed right now and possibly a few more as well. So there's a lot of services in here that you can play around with. You can understand the capabilities. We will try and look at a few of them. Like let's try sentiment analysis. Right. So at this point in time, like I said, I'm not writing any piece of code.
I'm just trying to demo and show you the capabilities of the service. So if you're interested, you can go on and read technical documentation. There's samples on GitHub. There's additional SDK links in that. You can use a programming language of your preference as well. You can use C#, Java, Python, you know, they're all supported. So here what I would do is I would select my text language.
This is the resource that I have created in Azure. So there is a billing that is associated with this. And therefore, I have just created my free tier, but I will just select that Azure resource. So it runs against the service. But this is how, this is where, you know, the action happens. Now we can choose from any of these samples.
You can drag and drop your piece of text or your document if you want to try it out on your custom piece of text as well. I will just go with one of these and let's see how this behaves. So I will just run this sample and you'll see that immediately it has identified in that piece of text, you know, what the target is.
And so I did want to show you that I have enabled opinion mining here. Now, that is an optional thing. If you'd like to have that level of granularity, you may enable opinion mining again, not mandatory. So because I have enabled opinion mining, as you can see, what I get out of the box is I get very granular information.
So on a whole, it says that my document sentiment is mixed. I have already 85% positive and a 15% negative comments in that. So this is what I would get if I would go with plain sentiment analysis alone. But because I have chosen opinion mining, see what I get. I get about the, an opinion about the spot.
And I also get an opinion about the controls of food place. So this is, this is, again, see if you see the entire original text. So this is the short answer. This is just the first sentiment. But if you want to look at the entire piece of text that we provided there, you'll see that there's a lot of assessment that has happened with each and every target.
It is coming back and giving you a label, which is between zero to 100%, zero to one. So it, it is very clearly indicating, you know, what was, what was working for you, what was not working, so on and so forth. So you can use this. Now, how you use this depends on your specific application on your specific use case.
You can be very creative with this and, you know, use that in your applications to the, to, to suit your business needs. So we have a couple of minutes left, but I wanted to wrap it up by saying that, you know, technology aside. Any system that we build, any system that we build, especially an AI system contains not just the technology.
So there's people, there's an environment, there's people who are using it, people who have created it. There's a whole system in place. And it's, and it's therefore important that we build AI solutions very responsibly. Um, so there are AI responsible AI principles that, um, all tech giants, um, you know, evangelize. And the one center, particularly from Microsoft is they, uh, do a good job at, uh, you know, evangelizing, uh, about building fair, reliable, safe, uh, private, secure, um, inclusive and accountable applications.
Especially when you're in the hot seat building AI applications. So, um, a couple of references there, uh, especially for the forest to study and the transparency note, et cetera. So, um, feel free to reach out if you have any other questions or, uh, would like to learn more about, uh, uh, the capabilities of, um, AI on the Microsoft side or the Azure side.
Um, happy to continue the conversations and, um, I will see if we have any quick questions. If we have a minute to take them. Uh, but thank you for your time. And, uh, that's all I've got. Yes, Harita. I plan to share the presentation with reference links, uh, in my, uh, LinkedIn. Um, so I just need some time to, you know, just polish that and to make it accessible.
So if I may put my LinkedIn, uh, uh, is here, uh, so that, um, uh, you can connect and we have it. Uh, all right. What do you consider to be the most used Swift in the industry to work in AI? Um, I'm sorry, Jess. I don't know if I understood that. Right. Uh, well, the most used Swift in the industry to work in AI.
If you'd like to clarify that a little, sorry. Uh, thank you, Laurie. I'm glad it worked out well for you. You're welcome. Usha. Thank you all for joining. And I really appreciate you taking time and your patience and your attention this morning. All right. I think I'm up for time, but, uh, hope you have a good rest of the conference and, uh, take care everyone.
Thank you.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
Can developers build AI apps without machine learning expertise?
Liji Thomas shows how pre-built Azure AI services bring language understanding, sentiment analysis and opinion mining within reach of any developer who can call an API.
In short
Developers no longer need deep data science skills to build applications that see, hear and understand language. Liji Thomas, Technical Manager at Valorem Reply and a Microsoft MVP in AI, argues that pre-built cloud AI services remove the biggest barriers to AI adoption. She demonstrates Azure's language capabilities, from intent detection to opinion mining, without writing a line of code.
Key takeaways
Learn the capabilities of the AI services that power Microsoft tools and products.
Learn the use cases of Azure AI and who is making it real.
Learn how to infuse intelligence into your applications using simple API calls.
Why AI adoption stalls
Thomas opens with a clear claim about timing. "So, we are at a stage in history where the opportunity with AI is now real." (0:30) She points to a Forrester study, as the speaker cited, that quantifies the benefits AI brings to applications in automation, business agility, revenue and customer experience.
If the benefits are so clear, why is adoption hard? According to the research she presented, more than half of leaders and tech decision makers report three obstacles. They lack enough data of adequate quality. App development skills are among the hardest to find. And data science or machine learning expertise is scarce, because building algorithms from the ground up requires advanced knowledge of statistical models and mathematics.
Cloud AI services as the shortcut
The industry's answer, Thomas explains, is cloud AI services. "Now, cloud AI services, the promises of cloud AI services is to provide pre-built AI services or models that any application developer or any, you know, everyday developers will be able to invoke using simple API calls." (2:12)
A developer who is comfortable calling a REST API in C#, Java or Python can start today. The services are pre-built but configurable, can be deployed in the cloud or on the edge, and come with ongoing support. Thomas lists the benefits: faster to build, faster to market and deploy, better troubleshooting expertise, and no need for machine learning expertise to get started.
From Microsoft Research to everyday products
Thomas traces Microsoft's offering back to decades of work in Microsoft Research centers around the world, some of which reached human parity. "Image captioning services that Microsoft has built, the Microsoft research teams have built, can caption an image as good as a human can." (4:04)
Microsoft first used these capabilities in its own products. Her example is Microsoft Edge, which automatically captions images when an author has not added alternative text, so someone browsing with a narrator does not have to rely on the author. Microsoft then opened the same services to developers. In Azure, these pre-built services sit alongside conversational AI and custom AI options, and Thomas describes their promise in human terms. "Because what gives us our intelligence is our ability to see, hear, speak, understand and make intelligent decisions." (6:15)
Teaching apps to understand language
To make the language service concrete, Thomas imagines a chatbot for the WomenTech conference. People ask when the conference takes place in countless ways, and the bot should return the same answer whatever the phrasing. Conversational language understanding goes further by extracting meaning. Questions about sessions on AI, leadership, security, inclusion or imposter syndrome look different, yet they share one purpose. "The intention of the speaker is to find a session in the conference, right?" (9:30) The session name is the variable part, which she calls an entity. Developers familiar with NLP or NLU systems, she notes, no longer have to build them from scratch.
Sentiment analysis versus opinion mining
Sentiment analysis returns a score between 0 and 1: closer to 0 is negative, closer to 1 is positive, and the middle is neutral. A review that says the hotel location and customer service were excellent scores clearly positive. Real reviews are rarely that simple. A guest who loved the location but was disappointed with room service leaves people and applications wondering what the overall verdict was. "So this is where you need to have, you know, better granularity and better control over the sentiment analysis here." (11:49)
Opinion mining, a feature of sentiment analysis, goes one level deeper with aspect-based sentiment. It labels each target as positive or negative, so the people behind an application can see both their strengths and their opportunities.
Thomas then runs a live demo. All anyone needs to try it, she says, is an Azure subscription, and a free trial with credits works. "Language Studio is a new offering wherein without writing a single piece of code, you can check out these services in action." (13:05) With opinion mining enabled, a sample text comes back as mixed overall, 85% positive and 15% negative, with separate assessments for each target. When developers are ready to move to code, documentation, GitHub samples and SDKs for C#, Java and Python are available.
Building AI responsibly
Thomas closes beyond the technology. Any AI system includes the people who create it, the people who use it and the environment around it. "And it's, and it's therefore important that we build AI solutions very responsibly." (16:52) She highlights Microsoft's principles for building fair, reliable, safe, private, secure, inclusive and accountable applications, and points to the Forrester study and a transparency note as references.
Watch the talk
"Apply AI with Zero machine learning expertise" by Liji Thomas, 20 minutes: https://www.womentech.net/video/apply-ai-zero-machine-learning-expertise-liji-thomas
About the speaker
Liji Thomas is a Technical Manager at Valorem Reply with over 15 years of experience in technical and leadership roles, having led teams that built solutions for Windows devices from mobiles, tablets and desktops to Xbox. She currently leads digital transformation initiatives for a global clientele and is passionate about problem solving, tech for good, building great teams and books.
Questions this talk answers
What stops organizations from infusing AI into their applications?
What are cloud AI services?
How can a chatbot understand the same question asked in different ways?
What is the difference between sentiment analysis and opinion mining?
How can I try Azure language AI without writing code?
Drafted with AI from the talk transcript. Editor review pending.
Frameworks and ideas mentioned
Microsoft responsible AI principles
Resources mentioned
Language Studio (language.cognitive.azure.com)
Azure free trial subscription
Azure Cognitive Services technical documentation
Azure Cognitive Services samples on GitHub
Forrester study on AI benefits
Microsoft transparency note
Cite this talk
LIJI THOMAS. "Apply AI with Zero machine learning expertise." WomenTech Network, 2022. https://www.womentech.net/video/apply-ai-zero-machine-learning-expertise-liji-thomas