ContentTECH Summit
March 20, 2018, Henderson, USA
ContentTECH Summit
Video
Integrating Chatbots into your Content Strategy
Available
In cart
Free
Free
Free
Free
Free
Free
Add to favorites
1.06 K
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speaker

Noz Urbina
Co-Founder at OmnichannelX

About the talk

This rush of enthusiasm and investment around AI and chatbots leads a lot of organizations to leap in with little attention to how their efforts wills scale or integrate with the rest of their content ecosystem. In this clip from Intelligent Content Conference 2018, Noz Urbina, founder & content strategist at Urbina Consulting, shares recommendations on how to integrate chatbots into your existing content strategy.

00:10 Conversational interfaces

01:34 Intents

03:54 Prompts

04:10 Context

06:33 Anaphora resolution

07:51 How to start?

10:28 Coverage “T”

14:04 Intelligent typed chunks for chatbots

Share

So I'm going to talk about what's a chat bot and I got some terminology there to walk you through. So first thing conversational interfaces so somebody I think I already asked earlier about chatbots like text chat bots in touch at Bots voice so we've got chat bots personal assistants voice control user interfaces and hybrids of voice control and chat bots. So conversational interfaces is kind of the CI's is the is the language based UI so you can have voice you can have text you can have a mixture of the two until it's the big umbrella over chatbots Siri

Cortana etcetera etcetera. We could be in voice mode. We can be in visual text chat mode or we can be in a hybrid. You can also check out Cheryl Platz. She is has super pedigree in chatbots. She did Amazon Amazon Echo and Microsoft Cortana and now she's kind of a chat bot specialist that I've been her workshops and they're very nice. Okay, so We go. Let's walk through an example of interaction of chatbot. So the user asks or enters. What's the temperature in could be by voice could be by

could be like text the bot replies with the temperature right now in Valencia is 23° And the bot can show something like that. So you've got a perfect day 20 degrees 5 day forecast and then prompts of bottom, you know of other questions, you might ask that are related. So that is representing or is math what we are calling intent so get weather. So the intent if you know, who does the Trump API most people through cell interphase. So how did the application speak at a certain commands which understands each of those commands has a name in this case something like get weather at

all. Those are called intense to ask a question, but they're all mapping to fix set of intense the football supports. So there's a grammar of the known command of the Bakken understand of unresponsive. within the reply we have parameter slots things like City and there a type that wishes in this type. It is a CD slot and the city in this if you talk about the whole reply is Dalton said it's the city's an entity within that reply. So it works it treats you start to treat language the same way we treat code is variable bits and then it's fixed bits 6 bits are the temperature right now in

that we have two variables the city and the temperature which is been going to vary depending on what the reality is. So this could be in the voice modality. And this is a text modality. So if you actually ask Google Assistant, it will reply invoice that the temperature right now the lights shut up. It's not going to try to read out all that extra information. But at the same time it's going to show Intex modality a perfect day and deliver we call payload. So that's content that is useful as a reply but you would never read that out

but still a good boss is out to attempt to read you a table of information, but it's very useful that I can actually retrieve that and show it. And then you have the prompt as I mentioned which is kind of leading the user to explore further and at the same time educating on the user of what other kind of questions is. This bought can understand and handle. And then contacts on the we talked about a lot here context is assumed can be assumed from the GPS. So if you would ask that that's why I said what is the temperature at assuming that I mean where I

am if I asked it specifically what is the temperature right now in Toronto Canada? It would it would force the context into a different situation. It would no longer assume that I'm where I am. So all these are important things to think about when you're planning on a bought interaction. These are the factors that you got to take in in mind and say what is the the the end of the house and what is it capable of doing? so Between request and response to body the recognizes us Pacific command from a predefined grammar or uses natural natural natural language processing

or an LP to parse the input intermittent at session. Just just before this one you would know all about MLP now and I don't have to go to bed. I'm so happy because this is not going to talk about and I'll pay as a doll said you got an LPN you got artificial intelligence and a bot can be your front end to that but not necessarily you can there are Bots which understand very fixed grammer's like you can say, what is the temperature? What is the weather? Like that's it. It cannot any understand natural language it no specific commands

that you use them if I can come back and I can be useful if any of you were making devices or your car or something and it's like a playlist of content. You only need very particular commands. And the only reason that you're using voice is to avoid the physical interaction but useful, but that's not as open as some of the okay. I like this a lot. This is something I think we don't think about often cuz we kind of are looking at the question-and-answer pair. We have a

situation like this user and asks, what is an elephant the boss says an elephant large land mammal body body blah and then the user asks how much do they weigh and an assistant like Cortana or Google can now say elephants typically weigh between and continue the conversation that's called anaphora resolution. So you've got this kind of ability for about to remember what what what was the previous topic and therefore add that to the context of the conversation and make more intelligent reply so dumb

boss can't do that. So I think that's the calendar the all the info you need for the rest of the talk. Take away to your main things chat bot responses are based on their ability to recognize specific support intense very few chatbots that you're going to see in the wild. Are you are actually going to be able to hold up any kind of real conversation. They have abilities. They have a scope that they're good at even though the Watson base ones that have huge corpuses behind them. They still only know how to handle what they've been trained on. If you're doing use

Mandrill natural language processing, you can go beyond the limited grammar and accept the water Ranger questions and push your techies and Play Tell Him by about Anna for resolution. And if you want to keep that kind of flowing Behavior, you have to tell them make it happen. So how do you get started the chatbot creation process starts with the basics the fundamentals of task analysis? So you're going to prioritize what you were going to actually support. build a comprehensive list of user task choose, which you will support with a Blog

I got some links here, which you can pick up in the slides afterwards methodologies for doing a task analysis, which is a whole thing obvious simple hack go to your search engine see what people are typing in go. Do you support department? See what people are asking that you know, that's a that's a quick and dirty way to do it. If you're going to be more formal and try to be more then you want to do a proper task analysis to see what what you should be doing. My workshop in the first day you saw the description or I got to see some people were actually there was all about a journey

mapping where we walked through of achieving of an object of an objective for user as a series of questions. Like I want to access my mortgage options or choose a car or go to university or whatever. I got something I want to do and we step through that overtime saying what questions would they have at each point and so we create as part of walking through the Journey question-and-answer pairs. Well with me then pair with answers either from the content that we do have or we have requirements for Content that we need to make That's a very useful process

on a budget looking more customer Journey mapping for preparing your map of what about us going to do? Stage by stage analysis of what questions use of maybe asking when they're teething objectives? Vital exercise for bought channel handoffs the best experiences a very narrow in scope. So if I bought fails which we call coming to a cliff with a real bat charm, so it's a cliff and it's like I don't know what to do anymore is that but should be able to hand off so you want to fail early and honestly with people so you want this is what I bought can do and we

think we can help people this far along with a bought in their journey. And now we we don't think I can handle that your boss going to put up a tens and hand off and say I'm going to call my colleagues or I'm afraid that something I can help you with. Here's a link to where you can get Solutions are here the phone number or whatever it however it handled that failure. You want it to fail well and fail honestly and clearly Okay, so be clear on your limitations. We call this a coverage T. You could have a very Broad. But

shallow breath of knowledge. So your your boss knows your entire glossary of terms. It knows all the terminology and all that kind of stuff until you had a chatbot in the corner of your window. You can ask it what like terminology or was it means or related topics questions and I can come back with that. But that's all it does or you could have a boat can carry you through as he saw a couple with the couple of seconds ago Pub possible Salesforce. It can take you through an entire registration process or password reset or very well a particular kind of

Journey and you have to choose because the reason we do a t and not a square because we can't cover everything. It's not going to work. So this kind of thing where you say? Hi, I'm your virtual assistant. I can tell you about our products and help with unfamiliar terms, but I don't do frog support. I don't do after sales and you have any kind of Depot technical question. That's not me be upfront. What I think it's the best way to think about boss think about it like yourself. Would you ever tell anyone that you knew everything know so you can go say I am

this this is what I know. It's looking at things we can talk about if you want to talk about about particle physics or or dolphin reproduction, that's not me. So there's nothing wrong with that just fail fast fail on Italy. Do not try to fool people. I think if you see any other sessions already, you've got this message. It's not a human. Don't try to pass the Turing test it it's not we're not ready for that at all. So the question that I was asked to answer in this talk by Robert was how this fits in with our other content. So what I want to talk about

is the fact that intelligent type chunks of content the type of intelligent content that we talked about this conference are very good for bots. So when we have contact which is targeted and topic-based we can do things like when a user asks, how do I get a banking Safe digital wallet number and we've got a topic that says determine a wallet number and it says you can receive a banking safe Vault number when a customer does not remember what it is if you've got enough information there to figure out from the question that the seeking a task.

And you got tasks and the nicely marked up and categorize in your repository your CMS as tasks, and they're consistently structured. So you can also interpret that they're looking for an entity banking safes. You determine the product context. You can use a little bit of a little bit of natural language processing to go from how do I get to how do I determine and how do I retrieve up there? So you can map from the key words that are in your in your head or content to the terminology with using your question? Once you got all that you got an intent get task steps and then you

pass it parameters that is looking for the banking Safe products and you narrowed it down to what pasta actually want. What you do want to do there is not past the entire page not asking Tire module cuz that's not appropriate or useful in the chat bot context were using this nice intelligent content. So where they grabbed it just the test steps and list those out. It was a really big complicated tasks. You might even pass them one step at a time and ask for some kind of confirmation. But if you don't have that intelligent type chunks of content, then that's not going to work because you can't

go out and get just the steps and you can't use just the top bit and the metadata and tagging to help you determine which piece of which unit of pain. Did you want so intelligent content and bosko very nicely together. This is example. I like a lot. I think it's too much. Jargon in a lot of a lot of content that we create and idcc. There's so many words that people just don't know the brand knows them cuz that's all they do all day, but the the users don't so if you have an example not a task. This is a concept you can go in find out you're seeking a concept for

a glossary of reference and then pull out just a short description. So what is omnichannel and then you returned just a definition unlike the tasks that you went in and got the guts and gave the steps here. We do the opposite we're going to give a short description because that's all we want to do in a chat if they want more than we could pass them a link or allow them to drill drill down. But this time we just want that little bit like they kind of got in search results. And someone it someone you can have your very specialized types. This one's a

case study. So someone someone asks what is omnichannel and then. With can you show me an example of omni-channel then you can pull out your case study. And then in this case it knows it's a case study because of the content type and it's tagged as being relevant omni-channel. So it provides it up as an example in Fallout question. And then you might pull out two bits depending on what you would determine the appropriate payload is but it's kind of question. So the more intelligent content the

more sophisticated you can be in your question answer pairs, which I hope I don't know what president served by this point so that it is what is the answer the crappy way to do it is to rewrite all these answers for your bought the good way to do. It is pair up the knowledge that you have with questions and be able to update the questions in one place and publish them to your butt. Do you have intelligence that's possible? If you don't then it's not you have to spend a lot more time massaging and fixing and rewriting your content for the body.

Cackle comments for the website

Buy this talk

Access to the talk “Integrating Chatbots into your Content Strategy”
Available
In cart
Free
Free
Free
Free
Free
Free

Access to all the recordings of the event

Get access to all videos “ContentTECH Summit”
Available
In cart
Free
Free
Free
Free
Free
Free
Ticket

Interested in topic “Marketing and advertisement”?

You might be interested in videos from this event

August 12 - 13, 2020
Online
26
302
ai, b2b, b2c, ecommerce, enterprise tech, financial services, gaming, growth, healthcare, innovation, iot, marketing, marketplace, media, mobile apps, product, telehealth

Similar talks

Val Swisher
Founder & CEO at Content Rules
Available
In cart
Free
Free
Free
Free
Free
Free
Tony Byrne
President at Real Story Group
Available
In cart
Free
Free
Free
Free
Free
Free
Heather Whaling
Founder/president at Geben Communication
Available
In cart
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “Integrating Chatbots into your Content Strategy”
Available
In cart
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
558 conferences
22059 speakers
8190 hours of content