Agenda
• Financial Times?• Universal Publishing• Semantic Publishing• Ontology aware NLP and semantic disambiguation• API always first• Responsive semantic http://next.ft.com (ALPHA)• Recommendation, engagement and reach
“The old “page editor” is dead; long live today’s content editor”
Global network of over 500 journalists
Universal Publishing Platform
Fundamentals of a Universal Publishing strategy…
Creating the right tools
Making the right things
Making the process efficient
Organised assets
Customer data
Good Metadata
Stop binding to platforms
API for services
Access and usage
Making Managing Exposing
This is what we are investing in…
Universal Publishing Platform
Stop being obsessed by individual platforms…
The game is much bigger than this…
Does my CSS look big in this?
Universal Publishing Platform
Think universally to avoid this…
Back end stuff
Database Stuff
Publishing stuff User stuff
Legacy stuff
Worrying stuff
Expensive stuff
Annoying stuff
Product
API First
API
Instead start your project like this…
Product
Back end stuff
Database Stuff
Publishing stuff User stuff
Legacy stuff
Worrying stuff
Expensive stuff
Annoying stuff
Dynamic Semantic publishing
USER EXPERIENCE
ONTOLOGY
TRIPLE STORE
Graph of Data
Ontology
Presentation
Ontology aware NLP and Semantic Disambiguation
RDF
Generic Analysis…
KB Gazetteer…………
Disambiguation………
Relevance Ranking
Some words about talking Apple the company. Also mentioning a person named Steve Jobs says a tweet on twitter
? Apple: Organization? Steve Jobs: CEO? ……….
V Steve Jobs: CEO- Steve Jobs:
football player- ……….
V Steve Jobs- …….- ……….
V Apple- …….- ……….
1. Apple (78%)2. Steve Jobs (69%)3. Twitter (58%)4. …
Curate
Update
Retrain & Adapt
Entity relevance
• Rank entities by their relatedness to the article
• Accuracy 80%+
• We consider various frequencies of entity mentions in the article and in the entire set of articles
• Positions in the article fields or in the first paragraphs of the body boost the relevance
Confidence and Relevance. How?
The relevance of an entity in arbitrary document may depend on:
Text context and the vicinity of an entity/concept within the text.
(Confidence)
Ontological graph context and the vicinity of an entity/concept within
the graphs knowledge model
The frequencies of entities in the corpus and document. (Relevance)
API, API, API……API
• APIs first
• APIs for everything (Content, Concept, Data)
• API a product?• Content Syndication to B2B partners
• Build multiple products on a single set of APIs• Flipboard• Google News Stand• http://next.ft.com• Etc…
http://next.ft.com [ALPHA] - Responsive + Semantic
• Stream/Page per Concept, Page per company [2million+], Page per Person[1million+], Page per location etc..etc..
A lot of output Complex navigation
• Far too many pages/streams for far too few journalists
• UPP semantic annotation architecture automates content aggregation (articles, images, video, comment etc..)
• Time coded, metadata annotated, on demand video
Recommendations
Serve relevant articles
to increase user engagement
and improve usability
Primary Objective
How Can We Increase Engagement?
By augmenting 1st party data with semantic data and applying related content we can increase engagement
1st Party Data
User’s Click Data
Engagement
Semantic Data Related Content Increased Engagement
How Do We Achieve This?
By using semantic data we achieve a more detailed view than FT’s 1st party data. This is due to exposing relationships between organisations and people. If an FT user exists in the semantic database we can utilise the additional data
FT 1st party data
Series1
1st Data Party
PeersCompetitorsColleaguesOrganisations
Sub-OrganisationsIndustryJob Role
Related Content
Series1
Increasing Reach
By increasing engagement we have a greater set of related content from which to build profiles. Also the semantic data provides a more granular view. This together with the relationships which semantic data surfaces allows for better content aggregation.
PeersCompetitorsColleaguesOrganisations
Sub-OrganisationsIndustryJob Role
Related Content
User’s Click DataCEOs in
Automotive Industry
User A Click Data
User C Click Data
User D Click Data
User B Click Data
Using this data we can create profiles e.g. CEOs in insurance. By recommending commonly consumed content to those users who share the same profile we can increase our reach.
Increased Engagement
User Behaviour Recommendation
Active Registered Users New Subscribers
Look for common patterns
of site usage
Target other registered users with
similar behaviours
Automated process that delivers ~15% of new subs per week with minimal CPA
Modelling subscription propensity
It’s just about listening to your audience
Data: User and Content metadata are crucial
Use of data…is absolutely crucialFocus on Data comes from…
First 120 years of FT little user data
Now we are 125 years old…
…we know better
Thank you. Questions?
[email protected] @jemrayfield
Top Related