Miles Media @SWSWi

Recommendation Engine

#discotalk

Liz Gaines - All Things Digital

Ron - Google Hotpot

Gary Camp - Stumbleupon

Tom Conrad - Pandora

Hunter Walk - YouTube

SIGNALS

Stumbleupon

• Social
• Trending
• Interest graph (produce better recommendations)

Pandora

• music genome project (initial hypothesis)
• Everything we know about you
• Like vs  Dislike - can change what they serve
• Look @ all thumbs (8 billion data) & aggregate in context to guess

YouTube

• Context of where video is appearing outside of the site in embedded video
• idea: look @ top travel blogs to see what they are embedding & put into discovery paths
• What you view on the site

Google HotPot

• Place graph (the web)
• You
• Social dimension
• “like minded” - similar tastes based on what the user enters

SEARCH VS DISCOVERY

Search = context by keyword

Discovery = surprise, education, serendipity (explain/justify value to users)

YouTube

• 2nd largest search engine in the world
• Search results page with a narrative focus, a set of content that will help me learn about a topic » create an experience
• Unique: look at what you friends haven’t watched (first to show social karma)
• Do home visits
• Maintain content diversity
• Diversity in recommendations

Pandora

• “one click personal radio” translates to all aspects of the brand
• Musical mood - enter 1 time & recommend based on that
• Audiences help curate music & stations
• Context Specific
• Example: Christmas time, on a Christmas channel an indie rock band named Christmas kept playing on the channel, by the users hitting dislike, they helped curate the channel and phased out the band dyamically

Stumbleupon

• Only from friends - not surprised
• Social = trust but non-social aspect important

Google

• Freshness of social signals

DIFFERENCES IN CONTENT?

YouTube

• Only video content
• Trusting user to upload data
• Apply technology to understand technology
• What does perfect meta data look like?
• Trying to create an experience > tailor a start 2 finish experience for the user

Pandodra

• Only music content
• Keep playlist algorithm transparent - data for each person
• Democratizing affect - genome

Stumbleupon

• All forms of content, esp. photos
• Isn’t structured
• Interaction, surprise, user is not expecting one experience

THE SINGLE USER?

• Log ins / multiple users watching together > youtube
• Relevance of recommendations - you might like this b/c you’ve watched that
• Pandora - future of radio - environment changes - an work, home, weight sensor in car seats??
• Stumbleupon - mobile users

RANK - HOW?

YouTube

• Tuen algorithms
• Long view vs short view
• Session length
• Percentage watched
• Get false positives

Stumbleupon

• Thumbs down
• How long people spend on media types
• Thumbs up (80% give thumbs up)
• Time 80 secs to 20 secs avg spent

Pandora

• Playlist algorithm
• Metrics change dramatically and are hard to quantify
• 2 major goals: quality of playlist and always being available
• Launch based qualitative sometimes
• A&B testing everything you do … intuitive is not always right .. to sensitive content

Google Hotpot

• Coverage

OFF THE SHELF RECOMMENDATION ENGINES

Difficult, skeptical, commerce, open up signals, future development

TURN OFF RECOMMENDATIONS FROM CERTAIN SOURCES?

Not looking @ source, just your reaction - YouTube

Connecting you to recommenders whether you are friends or not

Social + similarity > combine signals, don’t treat as 1 signal

ADVERTISING

Promoted videos that advertisers have bid on - YouTube

Ads are content as well - same concepets - YouTube

Rate Ads - Stumbleupon

MOOD

Pandora

• Takes small gestures and react quickly to determine mood
• Do it with smallest gestures from users

YouTube

• Session awareness
• Implicit vs explicit collection of mood signals

Stumbleupon

• not there yet

CONTENT-LEVEL RECOMMENDATIONS ON THE WEB

Stumbleupon > concept of “people who bought also bought, do more”

How to address labor-intensivenessd

• “machine listing” - Pandora pulls out acoustical details
• People overestimate the work involved in analyzing content

sxsw

recommendation engine

recom

recommendations

Future of Collective Intelligence: Location, Location, Location

GIVE BACK …

… what will you give back to the consumer for their check-in that is meaningful?

The check-in is not the meaning … it’s a signal, that when looked at in aggregate, is useful. A history of check-ins can extract patterns — data being fed — and feed out recommendations.

Always needs to be a reward at the end, either immediate or long-term.

sxsw

check-ins

location

recommendations

Non-Visual Augmented Reality and the Future of the Interface

Quick notes from a great presentation on non-visual AR (in that it is text-based, invisible layers until you get there). Will fill in the gaps when I have time.

Presenters:

• Amber Case | @caseorganic
• Aaron Parecki | @aaronpk

#evaporation

@geoloqui

http://geoloqui.com

Cyborg Anthropology

Automatic production of space

Persistent paleontology

Devices & their discontents

Panic Architecture (too much)

Steven Mann - wearable computer apparatus - 1st cyborg

Diminished reality

• Adding your own layer onto reality
• replacing public messages with your own

Functional location-based reminder applications - 2006 (leave messages for yourself)

Location is just beginning to be feasible thanks to timing, hardware and affordability

Staying in one place - GPS noise

Visual of the everyday activities, movement - when plotted on a plane makes a visual map

Issues:

• Social punctuation - if rewards don’t increase, people don’t want to continue (FourSquare)
• Clumps of friends offline vs online friends
• real life relationships are temporal, not permanent (i.e. client meeting)
• Privacy - control sharing when, not just who
• Noise in GPS data from cell tower triangulation

Co-location negotiation (texting etc)

• Can eliminate this if you know where someone is…
• Get a notification of where someone is (proximal notification)

“A successful interface makes itself invisible” - like electricity (abstract use)

• Notification when you enter a quadrant
• Automatic check-ins
• It should get out of the way (technology)

“Phone becomes a remote control of reality.”

Geo notes

• location-based reminders
• send messages to yourself
• contextual notes

Transit Directions

Proximal address notifications

Relevant info should be pushed to users

Wayfinding with non-visual AR

• Haptic Compass

API and Library Examples (makes reality like a video game)

• Notify X when you arrive at Place Y
• Notify X if I’m not in place by time Y
• Send messages when X is going Y MPH, but not above Z MPH
• Send someone a reward if they arrive within place X within Y minutes
• Notify X when you’re .5 miles away from Place y

User Jabber for real-time location.

Ambient Location (other cool uses)

IRC channel

Android and iPhone App - Geoloqi

• Layer of geocoded data
• geocoded wikipedia articles
• text interface feature
• app architecture with leaderboard with trivia grame

Don’t eat that

Hear Near Events

Pac-Man game

Issues:

• Scalable
• license the technology
• where geofencing patent
• Patent issue with geolocation technology
• Virtual Graffiti and Privacy
• Create anonymous areas like your home “default closed”
• indoor positioning is an issue (big areas, install indoor beacons or indoor notification through sounds)
• spoofing location

Augmented Reality

SXSW

Mobile

geolocation

games

Left Brain Search = Google. Right Brain Search = X

Panel featuring:

• Barney Pell: Microsoft | Bing | barneypell.com
• Don Turnbull: Info Theory | donturn.com
• Lior Ron: Google Hotpot | google.com/hotpot | @lioron
• Nikhil Daftary: Moodfish.com
• Sep Kamvar: Stanford | kamvar.org

#bettersearch

OVERVIEW:

Looking at where search engines fall short. how limitations are giving rise to a new set of search experiences based on semantic understanding and recommendations that are personal, social and contextual. Look at emerging technologies and user experience that are creating the “next big thing” for the search industry …

Future of search engines - how do we go beyond keywords to behavior, mood, context and to allow serendipitous search.

Next evolution of search is about who you are, not what you type.

What are dimensions of future search?

• personal
• social
• local
• mobile
• contextual
• semantic
• recommendation system

• old search: connects data + people + places + events
• new search: social graph + interest graph = inferred interest graph

Why are innovations in search happening now?

• New content (data)
• New signals about users (real identities)
• Connecting new queries with new content

Where is search going?

• Think about what kind of new info will be out there — then offer recommendations based on that info.
• Think of different types of data
• Think of spectrum of “give to get” and the privacy concerns it raises
• Privacy + choice + consent + control = users start to sign away their results
• Transparency - should be up front about what you are collecting and allow users to take with them
• Future “got chas” of search - Privacy and AI

How does emotion affect search?

• Data you’re looking for has to match your intent
• “misery loves company”

Concerns brought up in the Panel:

• Are search systems overfit?
• Are we phasing out people-2-people recommendations?
• Are we doing away with serendipitous search?
• Are we doing away with natural discovery?

Suggested solution to concerns:

• Purposely inject randomness & moments of serendipity into search experience
• “leveling up serendipity”
• User and computer evolve together through matching

How do you balance friend vs stranger recommendations?

• Balance signals
• Make personal adjustments
• Evaluation metric can provide answer
• Give reasons of why you received a recommendation — say why algorithm recommended something

There’s a double standard of people giving away individual data vs search engines not giving up their algorithms, etc.

Future: software on your phone will read your mood to help provide better recommendations for you

GOOGLE HOTPOT - RON:

• google.com/hotpot
• Google looking at signals you give it and recommends accommodations to you with maps.
• Build a graph of connections based on what Google knows about a user.
• Hot Pot Friends - adds a layer of friend’s recommendations on top of Google’s recommendations - allows you to choose which friends to give recommendations
• A discovery tool

SIMON ALPHA PROJECT - PELL:

• Makes recommendations
• Scan through choices
• Can choose, specify mood (semantic properties)
• Rate, feedback system more robust than thumbs up or down

PROJECT EMPORIA - PELL:

• Personal newspaper with channels
• projectemporia.com

MATCHBOX:

• Matches Xbox players
• Best guess but the system or device should understand the limitations
• Matches that would be the most fun

MOODFISH - DAFTARY:

• Mood has significant impact on impulse decisions
• Traditional search models lead to information overload
• Search content + intent = effective search
• Will eventually be able to search by fractions of a mood
• Mood doesn’t use history or past info, focuses on the now.

search

sxsw

search recommendations

mobile

recommendadtions

Ordering Disorder: Grid Design for the New World

Khoi Vinh | @khoi | #odogrids | desk@subtraction.com |Book | subtraction.com

“Thanks for coming to hear me talk about lining things up,” is how Khoi Vinh started his grid design presentation. I was hesitant to go to this session, because I’m not a designer, but it ended up being very interesting (architectural tie in) and straight-forward (by that, I mean very little math required to follow the principles outlined).

Whether you believe in following grids to the letter or just use them as a base for lining up elements on a page, it was worth hearing Khoi talk about his passion for ordering information:

“Design is a form of tyranny…”

GRID BACKGROUND INFO:

Any two marks on a single plane create a grid design order. Interesting fact: the brick is responsible for the first grids. The brick is shaped around the formation of the human hand. As a building material, it allowed for collaboration of workers building a structure in parallel. A brick wall was the first grid design. Multiple walls lead to extension of the grid, and so on …

WHY GRIDS? POWER OF GRIDS FOR DESIGN:

• Add order, continuity and harmony to information
• Allow an audience to understand into better and help them to predict where to find info
• Make it easy to add new content consistent with existing content
• Facilitate collaboration on a single solution

STANDARD PRACTICES:

• Derive beauty out of the browser
• Design as coders
• Every site should be usable, accessible and conform to open standards

USING GRIDS ON THE WEB: CORE PRINCIPLES:

Khoi talked about how it’s tempting (actually, he used the word seductive) to dive in, but first remember:

• Problem solve before aesthetics
• Serve the user experience
• The simpler, the more effective

STEPS:

Research

• Understand the problem
• Catalog, inventory and prioritize content
• Understand your constraints (such as ad placement) – this is a must

Wireframes

• Very basic layout of the page - for prioritization - more about what is on the page (content requirements) than how the page will be laid out

Preparatory designs

• Sketches - key is to generate ideas quickly & at low cost so you can close out avenues that won’t be fruitful right away; also, sketch throughout the process not just at the beginning
• Calculations - break down the page into units (base of page layout), regions (groups of units) and fields (sets or regions and units); use mathematics where appropriate (look at golden ratio, rule of thirds, 1.618 etc)

Visual design

• Go back to your sketches
• Place the logo
• What’s important in visual design is to meet users’ expectations for: access, speed, usefulness, social filtering, great content

Code

EXECUTION - KEY TAKEAWAYS:

Khoi walked us through how to build a grid using the steps above. Rather than recreate that here, I’m going to share some key notes that I took during the presentation:

• Prioritize content on wireframes
• Use constraints (non-negotiable element such as an ad) to help you build your grid. Build your grid around this element.
• 1024X768 is reasonable mid point | 960x650 is live area where you build your grid
• Sketches should always be done very quick and rough
• Start hacking by units to determine how many “columns” you will have. The audience will feel tension when they don’t know which side of the page is dominant. Multiples of 4 units are usually good. Keep subdividing units until you have the basis for your grid.
• Next, look @ typography. Look @ widest body of text of an article and how wide that would be on the page. Run samples. 60-80 characters/line is considered comfortable for readers. Set text @ various widths across columns. Font size lets us determine line spacing, which establishes our baseline grid. The heavier or wider your characters, the more line spacing you need. Set your H1, H2 and H3 text.
• Establish fields using your mathematics to set your horizontal grid.
• Right-column blindness occurs when users tend to ignore the right column of a page expecting advertisements.
• “Nudging” can be used to account for all states of an element, for example navigation usually can’t line up exactly on the grid b/c it’s active state may be off
• Grid helps you determine what you should leave blank, too

FLUID GRIDS - RESPONSIVE DESIGN:

Can still use a grid to guide the responsiveness of the designs.

Best practices:

• Users crave consistency
• Users want a shared experience
• Designs should not be to erratic
• Behaviors should be consistent with each device
• Content needs to change too, there should be more or less of it depending on the design
• Is it worth the expense of creating the complex? Not always yes

FINAL THOUGHT:

Grids were a created as a documentary form. Today’s media is about conversation. Audience was in receiving mode, and now in conversational mode — grids must adapt to fit this new model.

sxsw

grid design

design

expert

Your Mom Has an iPad: Designing for Boomers

John Mcree, Lead User Experience Architect at EffectiveUI

@johnmcree | #yourmom

Your Mom Has an iPad View more presentations from EffectiveUI

MYTH: Boomers are scared and intimidated by technology.
FACT: Boomers <3 technology. Boomers represent 1/3 of online population and 47% of 50-54 year-olds use social media.

MYTH: Boomers are not confidant in their ability to use technology.
FACT: Boomers self-perception = consider themselves more savvy than other age groups.

MYTH: Boomers don’t invest in technology.
FACT: Boomers spend the most on technology. 80 million Americans control 50% of the spending. Boomer women spend more in all channels than women from all other age groups (Forrester Research). Boomers outspend younger adults by $1 trillion.

WHO ARE BOOMERS?
Oldest members turn 65 this year (Ages 46-64)

Too young for personal memory of WWII but old enough to remember the postwar American high.

79 million – [  ] % of the US population (need to look up!)

HOW TO DESIGN FOR BOOMERS

Don’t – all are different

Design for GOALS | BEHAVIORS | APTITUDE | ATTITUDE | (NOT GENERATION)

Look at patterns

EXAMPLES / CASE STUDIES/STUDIES:

Nintendo Wii

Boomers & Technology study by Microsoft and AARP called “An Extended Conversation”

KEY TAKEAWAYS:

Boomers like to learn new technologies and share their knowledge

• Make use of things obvious and efficient
• Provide tutorials
• Overlay instructions. Help boomers be successful beginning at point of entry | what should they do next? Example: Things app
• The Kick Ass Curve (see Kick Ass graph here)

Boomers want technology to be safe

• Feedback (progress bars, etc)
• Retraceable steps (show how to back up; breadcrumbs)
• Strong visual design to establish trust (prominent logos, etc)

Boomers want ease of use

• Consistent navigation & behaviors sitewide
• Consistent nomenclature
• “Choose your horse and ride it.”

Boomers see technology as a tool

• Can’t keep adding tools to make a site better | increasing features doesn’t make it more compelling
• Give Boomers useful tools that mean something to them, do something unique

Boomers  expect technology to adapt to them

• Augmented reality (bring the tool to me)
• Voice commands – must be useful
• Any tool you provide must be useful!

Boomers expect accessibility

• 20% of the workforce has a significant disability
• 42% of the workforce with a significant disability are nearing retirement
• 60% of the workforce is currently benefiting from Accessibility tech (examples include: closed captioning, voice on GPS command)
• The word “disability” is a HUGE turn-off – boomers self perception of age is that they aren’t “old” no matter their age.
• Limit on-click events
• Use external labels on forms (not in the field)
• Provide meaningful links (for example: all links that say learn more on one page are not useful to Boomers)
• Watch contrast (example: button design, depends on size of text)

CONCLUSION:
Boomers want to learn and share as long as they see it as a tool that’s useful.

kickass

swsw

boomers

design

technology