Hi, everybody, thanks for joining. We're really excited to introduce you today and walk you through a new service called AWS natural language processing. My name is Nino Bice, a little bit about myself. I'm a product manager on the team, I work with the team of engineers. And together, we've gone out, thought extensively around, what are the problems in the space? How could we bring a solution to you that solves those problems? And we work day in and day out with our customers to continue in evolution of a roadmap. But that's a little bit on myself, I've been at the AWS about a year now. The majority of my background is in data systems and cloud in general, and semantic data experiences. So it's why I'll be talking to you about NLP specifically today. The course will take you through service introduction, of course, we'll talk about some overview and use cases. So you can understand not only what it is, but what it can do for you and how you can use it. And I'll take you through a brief demonstration of our console. Which is really helpful in understanding the service, and even allows you to play with it with your own information as well. So before we dive in to introducing the features of the service, let's set the stage for. Why are we here, and why are we talking about this service in general? It's really important to understand that the unstructured text, so text that is not in a schema and is not in a relational table. It's frankly exploding, It's growing exponentially. So if you think about the '70s, '80s, '90s, a lot of us were inputting information into computer systems in a structured way. Forms, we were writing them through data input, a lot of things like Excel. This information was coming in structured and it was therefore being stored in a structured way. And that means that there's a set of technology that we've built to allow you to store and query that at that structured data. So now we're entering an era where a lot of information being generated is unstructured. So you can think of things like social media, you can think of things like Twitter. You can think of the the way that your brand company or service is interacting with your customers. So those customers are feeling like with with Chatbots, they're interacting with you in conversational ways. They're interacting with your brand or service in comments and reviews. This is all data that's important and it's growing exponentially, because it's easier to communicate that way. And more people will continue to do that value is locked inside of this text. So to a machine it looks like a string of unstructured text. To a brand manager, it looks like what somebody is saying about their price. Or the experience they had staying at a specific hotel. Or the fact that when they stayed somewhere, they really enjoyed the coffee shop down the street. These are all these are all elements of information that are important to any business, or really anyone. So the reason why we're able to bring something of high value like this today. Is because of that machine learning and artificial intelligence. Text analytics and NLP specifically has been around for a while. But it's really been rules-based, allowing you allowing you to parse unstructured data. So you can do things like keyword counting and sorting. Now with deep learning models, were able to train this technology to bring human-like context and awareness. To that text extraction, to that NLP experience. And the last thing we want to mention, that's really important. Is that we've thought deeply around how to bring this technology to market, so that it's for everyone. It doesn't require an advanced skill-set, or maybe a three-month exercise. Where you learn about models deeply, you learn about training models. This service specifically is brought so that everyone that works with data today. With the skills you have today, can now look at approaching natural language processing. That's AI-based, using the skills that you have, so let's introduce the service. The service itself offers five main capabilities and we'll talk about it this way. And it's important to remember that all of these capabilities are based on deep learning. The first one is sentiment. Sentiment allows you to understand whether what the user is saying is positive or negative. Or even neutral, sometimes that's important as well. You want to know if there's not sentiment, that might be a signal. The next one is entities. This feature goes through the unstructured text and extracts entities and actually categorizes them for you. So things like people, or things like organizations will be given a category. And we'll walk through more detail what that means. The third capability is language detection. So for a company that has a multilingual application, with a multilingual customer base. You can actually determine what language the text is in. So you know if you have to translate the text itself, or take some other kind of business action on the text. The fourth capability is key phrase, think of this as noun phrases. So where entities are extracted, is maybe proper nouns. The key phrase will catch everything else from the unstructured text, so you actually can go deeper into the meaning. What were they saying about the person? What were they saying about the organization for example? And then the fifth capability is topic modeling. Topic modeling works over a large corpus of documents. And helps you do things like organize them into the topics contained within those documents. So it's really nice for organization and information management. So let's talk a little bit deeper around the APIs that help you text analysis. In the example on the left, you can see that we have a snippet of unstructured text. This may have come in through a comment, or it was maybe mentioned somewhere. And you can see what the four APIs are doing here. The first one is extracting the named entity, so amazon.com is extracted as an organization. Seattle, of course, is extracted as a location. You can see that we extract noun-based phrases, or things like everyone, great customer experience. We know that the sentiment on the last sentence is positive. because of course everybody loves the great customer experience, it's generally a positive thing. And of course we've determined this snippet of text is English, of course, because it's English. The fifth capability that we've talked about is topic modeling. So topic modeling, what we've done is, we've actually brought topic modeling as a service. So for those that aren't familiar, topic modeling is doable today. It's based on an algorithm called Latent Dirichlet Allocation, LDA, and it's been kind of hard to go set up. You have to go find an environment, there's a lot of parameters to tune. You have to obviously deploy and operate that environment, to run that algorithm. Our team has done a lot of heavy lifting to make that algorithm available to you as a simple API. So we can think about topic modeling as a service. So you can just walk up, bring your documents and start using it. This service works by extracting up to 100 topics. A topic is a keyword bucket, so you can see what's in the actual corpus of documents themselves. And then the service also returns to you an automatic view which maps documents to the topics. So to give you a really basic use case, you can take a thousand blog posts. Understand what's in the blog post, from a top 100 topic perspective. And then actually map all the blog posts into those topic buckets. So if you wanted to give your users a really easy way to to explore or browse your blog posts. Based on the topics they're interested in. You could do this with a simple call to this job, and the job service itself. The next thing we'll talk about is what gets us really excited, is why the service is valuable? So like I said, NLP has been around for a while. There's a lot of folks doing NLP that's AI-based. What we've built here today is a service that's truly accurate. We have an engineering team, and a data science team behind this service. Continually working nonstop to make the service accurate. On day one you'll notice that this service is accurate out of the box, and it's in its competitive. And it's useful for the accuracy that you need for your use cases that you're dependent on. It's continuously trained, so as we've said before, we have folks behind there. Collecting data, annotating, training the model, looking for accuracy problems, fixing them. We're doing this continuously non-stop. So the more you use this service, the more that you'll be able to have the service become accurate for you, based on your own data. And then based on the fact that the team is training it on your behalf, so the service gets better over time. And the service is easy to use, so as opposed to understanding what a model is. Or how to think about training a model, or invoking a model. You can simply walk up, and it's included in the AWS SDK. And you can simply invoke the service, it's a REST API. And you could build the service in conjunction with an AWS analytic service quite easily. So now let's dive into a demo show, you a little bit about what the service actually does. And how it works and we'll show you the console itself. So let's take a moment to look at the service, and look at some real examples. So if you log in to the AWS console, you'll notice that the database NLP service comes with a really nice API explorer. Where you can enter your own text, or use example text that we provided for you. In this particular case, this is the text that comes with the console itself. And you can see over here, the entities that we've extracted. So you can see amazon.com is an organization, you can see Seattle, Washington is a location. You can even see other organizations like Starbucks and Boeing. The next thing that you'll see is that we've extracted key phrases. So these are noun-based phrases that we're extracting from this text, so some of them are the entities we've extracted. But there are also other things, more like common nouns, like customers, books, and blenders. As I mentioned earlier, combining named entities with their key phrase output. Really helps you understand what's in the text, and what's being referred to in the text. The next API that we've mentioned is language detection. So for this text, you can obviously see that we're very confident that the input text is English, and we've marked it as English. The fourth API is the sentiment API, so it sees that this statement that we've entered here, is relatively neutral. But if I erase this and I said something like, I love my Amazon deliveries. And then I analyze that text, you can now see that we're very confident this is a positive statement. This is a great example of how you can use that to understand what customers are saying. And of course if I went back up here, I'd see that Amazon the organization was mentioned. So you can quite literally understand that customers mentioning your organization. They're mentioning it in a positive sentiment way, which allows you to really take action, dive in, learn more. The fifth API that we've talked about, is topic modeling. So as I've mentioned, we've taken a fairly complex algorithm like LDA. And made it available as a pretty easy-to-use service. In this case what you can see here, is that all we require as input, to run the topic modeling job for you. Is an S3 bucket that contains a corpus of your documents and input format. Which literally just says, tell us if you're delimiting by line. Or if each document is its own file. You can specify the number of topics. So you might want to take 1,000 documents, and put them into ten topics. Or you might want to put them into the top 100 topics. The next thing is to provide a security permission to access the bucket on your behalf. Give it a name, so this is just simply so you could track the job. And then a location of where to put the output. And as I've mentioned before you'll get to CSV files as output. One file will show you what are the topics. So if you've said show me 100 topics, we'll show you those 100 topics, and the keywords associated with them. And the next output is going to be what documents are mapping to those topics? And you can go act on that output, however you'd like. So that completes the demo, this is the console. We urge you to go in, plug in your own data, try out the service, see if it works for you. So now we're done with the demo, let's talk about some common patterns. What are we hearing from our customers, around where do they want to get started with their NLP solutions? And we've really ultimately seen it that the patterns boil down to these three areas. It's really Voice of Customer Analytics. What are your customers, what is anyone really generally saying about your brand product or service? These are really important in understanding if the new product you've just launched. How are people perceiving it? Do they like the price? Do they do they think that the color is off? These are really important things that you want to know, that you can capture from the Voice of Customer. This can be from social media, this can be from comments that they're leaving on a site somewhere. This can this can be from emails that they're sending to your company directly. It could even be support conversations that your agents are noting within support call notes. The next general pattern that we see is Semantic Search. So for example, if you're an elastic search customer. And you're currently indexing a corpus of documents to make them available to users. You can actually use the NLP service to extract things like topics, key phrases and entities. And also index on those as well. So your customers can get a better natural search experience. You could suggest other documents from the search experience based on topic contained within the search result. It just makes search better, understanding what's in the documents themselves outside of just a keyword context. The third pattern is Knowledge Management Discovery. So we hear a lot of customers say, I want to take a big corpus and organize them. I want to understand what's in these documents. I've got a variety of use cases from making this document corpus easier to navigate. All the way to, we're really looking for what's contained in these documents. To make sure that we're meeting certain standards around what information can be stored in documents. So we see a lot of customers using NLP in these these three general patterns. Let's now take a look at an example of how you would use this NLP service in context of an AWS analytics solution. In this case, we're going to talk about a social analytics application. So on the very left of the diagram we'll have tweets. Let's pretend that we have a bunch of customers tweeting about our brand service or product. We've set up a Kinesis Firehose which is calling the the the Twitter search API. And it's pulling in tweets that we've set to filter out, that we think is pertinent to us. We're then running those tweets through the NLP service to extract things like the entities in the tweets. Or the sentiment of the tweets, or even the key phrases in the tweets. We might even be determining what is the language that the tweets are in. So we really understand more about where our customer base is in the world and what they're saying. So we'll run all those tweets through the NLP service and we'll store them into a store. We could use a relational service in this case, or we could use Amazon S3. We've written all the output from the NLP service into S3, and now we can just take a query analytics tool like Athena. And start to query and analyze the NLP output. So for example, once we query that data, we can then build views inside of Amazon QuickSight. That shows us things like, who is mentioning other organizations when they're tweeting about my brand? Who is mentioning my brand or my service or my product in a negative context, and why? What are the what are the keywords that they're using are the key phrases they're using when they talk about my brand? This could allow us to do a variety of things, like, hey, in this part of the world. Customers are interpreting the product we've just launched as maybe too expensive. So bringing the NLP service together with AWS analytics capabilities. Allows you to really do text analytics at scale for a wide variety of scenarios, in this case social analytics. So thanks for attending the course on it the new AWS NLP service. We're so excited to see what you can do with the service and the solutions that you can build. It's really easy to get started, we've offered a free tier. So there's no cost to you to use your own data, we've even provided some sample data in the console. So once again, I'm Nino Bice, on behalf of the team. Thanks for considering the AWS NLP, thanks for watching. [MUSIC]
