beancounter.io - Social Web user profiling as a service #semtechbiz

beancounter.ioa Social Web User Profiling as a Service

Davide Palmisano @dpalmisano Wednesday, September 19, 2012, London

table of contents

the Social Web

the illusion of content personalisation

beancounter.io: user profiling as a service

a scenario for Social TV

the Social Web

“the Social Web is currently used to describe

how people socialise or interact with each other throughout the World Wide Web”

december 2007**from webarchive.org

today** http://www.readwriteweb.com/archives/alternate_reality_games_viral_marketing.php

semantic markup technologies and authorisation protocols blurred the borders

between contents and users’ social graph

the Social Web is not only

about socialising or

interacting with others

the Social Web is the place

where the users project their

identity though consuming

contents

your app, your

contents

your app, your

contents

your app, your

contents

engagement,content syndication

your app, your

contents

separated analytics, content

recommendations

engagement,content syndication

the illusion of content personalisation

“are analytics the most you can get from your audience?”

insights, analytics and statistics are essentially

quantitative measures of your audience

but there’s a lot more to be

discovered from your users

what are your users

interests?

what are their

preferences?

are there valuable

patterns between their interest?

crunching the Social Web, in real-time.

Beancounterformerly known as

each activity done on the Social

Web, carries some implicit knowledge which could be

considered as a fraction of a

user’s identity

how we can make it explicit?

how we can represent it?

how to follow its evolution over time?

anatomy of an activity

subject verb object context

subject verb object context

anatomy of an activity

subject verb object context

anatomy of an activity

subject verb object context

anatomy of an activity

subject verb object context

anatomy of an activity

every Web page text contains

entities potentially representative

of a user’ interest

and those named entities are represented as Linked Open

Data identifiers.

Natural Language Processing technologies are used to extract

named entities from textual objects

Linked Data as Palette

picture by @danbri http://www.flickr.com/photos/danbri/3478830059/

http://dbpedia.org/page/Mario_Monti

http://dbpedia.org/page/Italy

http://dbpedia.org/page/Spain

http://dbpedia.org/page/2007-2012_global_financial_crisis

named entities extraction, text categorisation

named entities extraction, text categorisation

record linkage

profile updateold profile

named entities extraction, text categorisation

record linkage

old profile

* for each incoming activity

profile update

named entities extraction, text categorisation

record linkage

record linkage

follow-your-nose

record linkage

*

*owl:sameAs

follow-your-nose

record linkage

profile updateold profile

*

*owl:sameAs

follow-your-nose

record linkage

profile updateold profile

* for each incoming activity

*owl:sameAs

*

activities

Web identifiers

profile weighting

your app, your

contents

your app, your

contents

your app, your

contents

your app, your

contents

engagement,content

syndication

your app, your

contents

separated analytics, content

recommendations

engagement,content

syndication

your app, your

contents

engagement,content

syndication

separated analytics, content

recommendations

real-time profiles

interest mining (batch processes)

Now, think about having stored

all the snapshots of your

users’ profiles in terms of theirs weighted interests

interest mining, is that process which allows you to

discover patterns and relationships between di!erent

users’ interests

a Social TV scenario

“60% of Americans use the Web simultaneously while

watching TV”http://blog.nielsen.com/nielsenwire/online_mobile/three-screen-report-q409/“

TV broadcaster

curated contents

login, comments, sharing contents

TV broadcaster

curated contents

TV broadcaster

curated contents

login, comments, sharing contents

TV broadcaster

curated contents

real-time profiles

interest mining (batch processes)

TV broadcaster

curated contents

login, comments, sharing contents

TV broadcaster

curated contents

personal recommendations

real-time profiles

TV archives

interest mining (batch processes)

advertising, audience tracking and identification

40K new users/week expected

2nd screen iOS/android launch foreseen for October 2012, backed by beancounter.io

a user watched something from my archive

a user shared something on Facebook

a user watched something from my archive

a user shared something on Facebook

generic interests layer

a user watched something from my archive

a user shared something on Facebook

custom profiling rules

generic interests layer

a user watched something from my archive

a user shared something on Facebook

custom profiling rules

generic interests layer

application-specific interests layer

a user profile

a user watched something from my archive

a user shared something on Facebook

custom profiling rules

generic interests layer

application-specific interests layer

a user profile

beancounter.io in few words

Open Linked Data profiles, for interoperability

real-time computation, to closely follow your users

available SaaS, in-house deployment

fully customisable, to tail it on your domain

baked by top-class open source products, lambda-architecture

N. Marz, “Big Data”, Manning, 9781617290343*

*

crunching the Social Web, in real-time.

http://launch.beancounter.io

@dpalmisanoDavide Palmisano