[MUSIC] While I knew about the power of text analytics, I was shocked at how much information could be extracted from a photo or a video. Welcome Seth. >> Hey Randy. >> I heard a rumor that if I have a photo that's tagged with my name five times on Facebook, they'll be able to identify me forever. Is that true? >> Absolutely. And they'll be able to identify you regardless of who's taken the picture, regardless of the lighting, regardless of whether it's in a video or not. Now, Google may not be able to identify you because they're not sharing information. But yes, you'll absolutely be identifiable. >> Now, I was under the impression we're gonna talk about what data could be extracted from a photo, but it looks like we're having lunch, are we? >> One of us is. >> Aww. >> So, we've talked about facial identification. So, imagine we're sitting here and talking, which we are, and I'm identified as Seth Redmore, Chief Marketing Officer of Lexalytics. My Twitter handle is @SethRedmore. Here's some biographical information about me from the web, and you're identified as yourself and who you are and your place in the world. But then also inside of here, I reach for my beverage and I drink, and we notice a brand. So there's a Taco Bell brand, on the bag and on the cup. We notice that this is a beverage, and so you can identify the fact that it is Taco Bell beverage. You can identify an action of drinking. You can identify that I'm about to have a taco in front of you. And that I'm enjoying my taco. >> And I'm upset that you're having a taco and that you're not sharing it with me. >> Exactly. And you can say, okay, these hot sauces, very identifiable colors. And I happen to like these two particular kinds of hot sauces. So you can identify what's being consumed, what's being done, and who is doing it and how they're feeling about it from the photo or from the video. >> So let's say I'm out at a bar, and I am having a beer with some friends of mine. But in the foreground at another table, somebody takes a selfie. Are you able to identify me in the background, and what sort of things could you identify from that photo? >> I could identify you. I could probably identify where you are, particularly if they geo-tagged the photo. Also, I can probably identify where you are if other photos had been taken in that institution that had been tagged with that institution's name, like somebody checked in. I can tell who you're with. So there's a lot of pieces of information that can be extracted just from that photo. >> And so, even though I'm not the subject of the photo, it is still analyzing me in the background. >> Yes. >> Does that happen with, say, videos as well? >> Absolutely. >> Can you give us an example of that? >> Yeah, sure, so imagine a party, and somebody's walking around and they're videotaping the party. And they're getting people's faces in there, and maybe those faces are being recognized, or maybe not, depending on whether they're tagged in something that's associated with wherever they uploaded it. You can identify something about those people, demographic information like racial characteristics, gender, age. You can identify whether they're happy or sad, angry. Of course, you can identify the sort of brands that they're are interacting with, are they happening to have a beer, are they enjoying their beer, are they pouring it out on the ground. So you can tell an awful lot from that video, even without knowing who the people are. But if you know who the people are, then you can associate it back to a profile, with like this person happens to generally always drink this kind of beverage. Why don't we send them some coupons? >> Oh, great. >> That sort of thing. >> What are some other examples of how video is being analyzed to help people gather information and insights from the video itself? >> Law enforcement is a huge application. Imagine you're at a riot. I don't anticipate I'd ever see you at a riot, but imagine you are, and you're taking video of it. You can extract where it's occurring, particularly if you were to tag it as this is happening where, or by street signs, or by storefronts that are there that are only in that particular area. Particular if other people have taken video of that area, so you know where it is. You could identify the people that are there. You could identify that maybe it's associated with a sporting event because everybody's wearing sporting jerseys. You can identify that people are doing bad things like, say, trying to break into something or stealing something or hurting another person. >> That's done by machines, not by people. It's all machine learning. >> So, exactly, it's done, the important thing to note here is the first pass will be done by machines. The humans will teach the machines, and machines will do it. And it provides this filter for humans to then look at. And say, oh okay, something really bad did happen here, as opposed to having to watch the 24 hours of video that are uploaded every minute. >> I also heard that with using machines, they can watch, say, a soccer game or a football game and actually write the story about it. Is that true? >> Yeah, so imagine all these technologies working together. So at the top, you have the video technology recognizing the players. The players are wearing uniforms that have numbers on them. You know the roster. So that makes it easy to track them. You know they're two different teams, so they've got two different color uniforms on. You can look for various events inside of the game, like a yellow card, or somebody got hurt and ended up on the ground, or somebody scored a goal. These are all events that are relatively, they're anomalies, and so, they're easy for the machine to pick out. And so that allows somebody who wants to write a story about a two hour event to collapse it down to three minutes of, okay, here's the interesting stuff. And when you combine that with, say, speech to text, you can understand what the announcers are saying and collapse that down as well, and have all that analysis presented to you so that you can then put your own verbiage and spin on it. >> That's really amazing. One of the things I really appreciate is Lexalytics also has a product called Semantria that essentially takes the big information you've talked about as extracting and actually turns it into big insights for a business. And one of the things I love is that you're going to provide that to our participants. Can you tell us a little bit about Semantria? >> Lexalytics focuses entirely on text. We feel that there needs to, there's enough complexity there for us to be solving those problems for a long time. So what we've done is we've produced a couple data sets based on the hotel industry. And so, hotel A, hotel B. You run hotel A. And so you wanna know what's going well and what's not going well inside of your hotel and if there's certain people on staff that happen to be doing a better job or worse job. But you wanna move in on Hotel B's territory and start taking business away from Hotel B so you can look to see what are they doing well and what are people complaining about there, so that you can offer the services and advertise the things that will actually bring business from them to you. >> That's a fantastic tool for our participants to learn how to do social monitoring. I want to thank Seth for these great insights into what can be done with the photos and videos, as well as access to Semantria. [MUSIC] [SOUND]
