Do you want to learn how to use the low-hanging fruit of knowledge graphs — schema.org and JSON-LD — to annotate content and improve your SEO with semantics and entities? This hands-on workshop with one of the leading Semantic SEO practitioners will help you get started.

1. From Knowledge Graphs to AI-powered SEO:
Using taxonomies, schemas and knowledge
graphs to improve search engine rankings
and web publishing workflows
wordlift.io for Connected Data - October 2019

2. IN
TRO
Andrea Volpini
Cofounder and CEO of
WordLift
David Riccitelli
Cofounder and CTO of
WordLift

4. We help website owners expand traffic with an
organic sustainable growth
...using semantic web technologies
Source: SEOZOOM data from an Italian blog on photography

5. We’ll dive into SEO and structured data.
Our goal is to build a knowledge graph that search engines can use to understand the content of a
website.
We’ill run a “semantic audit” on a reference website using Python (code available in Google Colab) to
extract the most common entities.
Starting from these entities we will enrich them using OpenRefine with queries against DBpedia and
Wikidata (code included in this presentation).
At last we will import this data on our website and we will re-use it to markup the articles, to add internal
links and to gain more insights on search traffic.
What is the plan for today?

7. “Make Your Website Talk” using Google
featured snippets and structured data

8. ● Developed by and for the search engines
● Stable, reliable and extensible
● Has become the de-facto standard for Linked
Data development
● Strikes the right balance from complexity and
expressiveness
● Open and community-driven

9. ...then one day, in 2010, we tell
our biggest client that we have
a product that uses schema
markup to bost SEO!!

10. ROI of Semantic Web technologies - in the context
of Search Optimisation - is now easy to prove
● +12% avg. rankings
growth (from 28.7 to
25.6)
● +22.22% CTR
increase (from 1.8% to
2.2%)
Source: GSC data from a travel brrand in Canada

11. “Open data needs to be
taken as serious as open
source software.”
Chris Taggart
Semantic
Open Data is
an essential
building block
of Modern
SEO
LEARN
IN
G

12. Danny Sullivan and
John Mueller on
how to enable
snippets,
thumbnails and
rich results
following EU
reform of online
copyright law
https://twitter.com/cyberandy/status/1176942111767388160?s=12

13. GOOGLE MAKES BROADER USE
OF SCHEMA TYPES THAN WHAT
IS APPARENT THROUGH
EVIDENCE LIKE RICH SNIPPETS.

14. Query augmentation patent - 3/2018
“In addition to actual queries submitted
by users, augmentation queries can also
include syntetic queries that are
machine generated [...] A way of
identifying an augmentation query is
mining structured data, e.g., business
telephone listings, and identifying
queries that include terms of the
structured data, e.g., business names.”
Ok Google,
find the best
syntetic query to
answer my request
“Google may decide to add
results from an augmentation
query to the results for the
query searched for to improve
the overall search results.”
Bill Slawski - SEO by the sea
PATENT

16. Isn’t this
supposed to be
Wikipedia?

17. MainEntityOfPage
http://data.wordlift.io/wl0737/entity/andrea_volpini

19. P6363
http://data.wordlift.io/wl0737/entity/andrea_volpini

20. What Content
Shall be spoken
aloud?
Using speakable schema
markup we can tell Google
what sections within an article
or a webpage are best suited
for audio playback using
text-to-speech (TTS).
BETA

21. Make Content easy to be
found by Assistant Users
● Optimised for Search
and for the Google
Assistant
● Semantically enriched
content
● 5 stars linked open
data
● Personalise the listing
in the Google Actions
Directory
● Let Google find your
action with implicit
invocation
● Add links to your intent

22. Can I book this
bungalow for
you?
Using Schema Actions we
can tell Google and personal
digital assistants what actions
can be trigger for a given
entity.+7.5% +43.0%
Source: GSC data from a travel brrand in Netherland

23. <script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "WebSite",
"url": "https://www.example.com/",
"potentialAction": {
"@type": "SearchAction",
"target":
"https://query.example.com/search?q=
{search_term_string}",
"query-input": "required
name=search_term_string"
}
}
</script>
Actions in Action

24. 1 2 3 4
Recap
Using schema
markup I opt-in for
Google’s snippets
thumbnails and rich
results. Google
considers structured
data as free-to-use
open data.
The ROI of rich
results is now easy to
demonstrate.
Google makes
broader use of
structured linked
data that we can
benefit from.
Findability is
improved by helping
Google augment user
queries.
Branding is also
improved/controlled
by helping Google
enrich its Knowledge
Graph (I can claim kg
panels and connect
data).
We can also claim a
directory pages for
the Google Assistant.
Using schema actions
I can improve the
user engagement on
Google SERP.
This results in new
entry points for the
funnel (ie. deep links
to apps) and support
for conversational
UIs.

25. Hands-On
Agentive SEO WordLift
Your
Content

26. A Knowledge Graph is a programmatic way to model a
knowledge domain with the help of subject-matter
experts, data interlinking, and machine learning
algorithms.
The Knowledge Graph (for SEO) is built on top of
existing databases such as Wikidata and DBpedia to link
all data together at web-scale combining both structured
information (i.e. the list of destinations on a travel
website) or unstructured (the articles on the website).
“
”

27. Building a
SEO-friendly
Knowledge
Graph

28. Website
analysis and
URLs selection
Named
Entity
Extraction
Data Linking &
Enrichment
Data
Publishing
Data
Exploitation
& SEO
WordPress/WordLift (or your favorite CMS)OpenRefineCrawler Google Colab

29. From URLs to a Knowledge Graph
data expansion for travel destinations
CRAW
LER
1. Crawl and analyze urls
that can benefit from
structured data
2. Collect data from LOD
to enrich each page
3. Improve Impressions
and Clicks on Google

30. Named
Entity
Extraction
using Google
Colab !
👉 http://bit.ly/semantic-audit

31. From URLs to most common Entities
Semantic Audit - We extract named entities from the URLs using Python
BETA
all the
urls that
matter
the most
common
entites

32. How to clean
data

33. MISSING DATA
WRONG VALUES
ENTITY RESOLUTION
TYPE CONVERSION
DATA INTEGRATION
MISSED MEASUREMENTS,
INCOMPLETE FIELDS, ETC.
MISSPELLINGS, OUTLIERS,
“SPURIOUS INTEGRITY”, ETC.
DIFFERENT VALUES,
ABBREVS., 2+ ENTRIES FOR
THE SAME THING?
E.G., ZIP CODE OR PLACE
NAME TO LAT-LON
MISMATCHES AND
INCONSISTENCIES WHEN
COMBINING DATA

34. We refine data by isolating rows and
by applying changes

35. FILTER TRANSFORM

36. How to link &
enrich data

37. Data Enrichment Pipeline
1. Reconciliation/Match against Wikidata
2. Get Wikidata URI
3. Load Wikidata JSON
4. Get Wikipedia Address
5. Add synonyms from Wikidata JSON
6. Get DBpedia URI
7. Load DBpedia JSON
8. Add images from DBpedia JSON
9. Add schema entity types from DBpedia JSON
10. Add sameAs from DBpedia JSON
11. Add comment from DBpedia JSON
12. Add abstract from DBpedia JSON
13. (Add every other property from Wikidata)
W
alkthrough

38. Entity Reconciliation on Wikidata
Here is a video to watch

39. Step-by-Step 1/6
1. Reconciliation/Match against Wikidata in the required language
2. Get Wikidata URI
id = cells["entity"].recon.match.id
wikidata_uri = "http://www.wikidata.org/entity/" + id
return wikidata_uri
3. Load Wikidata JSON
import urllib
q = cells['entity'].recon.match.id
url = "https://www.wikidata.org/wiki/Special:EntityData/" + q + ".json"
response = urllib.urlopen(url)
return response.read()
4. Get Wikipedia Address
import json
dict = json.loads(value)['entities']
q = dict.keys()[0]
return dict.get(q)['sitelinks']['enwiki']['url']
Python/Jython

40. Step-by-Step 2/6
5. Get DBpedia URI
return value.replace('https://en.wikipedia.org/wiki/', 'http://dbpedia.org/resource/')
6. Load DBpedia JSON
import urllib
url = value.replace( 'http://dbpedia.org/resource/', 'http://dbpedia.org/data/' ) + '.json'
response = urllib.urlopen( url )
return response.read()
Python/Jython

41. Step-by-Step 3/6
7. Add synonyms from Wikidata JSON
import json
def get_value( synonym ):
return synonym['value']
dict = json.loads(value)['entities']
q = dict.keys()[0]
synonyms = dict.get(q)['aliases']['en']
values = map( get_value, synonyms )
return 't'.join( values )
8. Add images from DBpedia JSON
import json
def get_value( item ):
return item['value']
dbpedia_json = cells["DBpedia JSON"].value
dbpedia_uri = cells["DBpedia URI"].value
dict = json.loads( dbpedia_json )
images = dict.get( dbpedia_uri )['http://xmlns.com/foaf/0.1/depiction']
values = map( get_value, images )
return ', '.join( values )
Python/Jython

42. Step-by-Step 4/6
Python/Jython
9. Add schema entity types from DBpedia JSON
import json
def get_value( item ):
return item['value']
def starts_with_schema_org( value ):
return value.startswith( "http://schema.org/" )
dbpedia_uri = cells["DBpedia URI"].value
dict = json.loads(value)
same_as = dict.get( dbpedia_uri )[ "http://www.w3.org/1999/02/22-rdf-syntax-ns#type" ]
values = filter( starts_with_schema_org, map( get_value, same_as ) )
return ', '.join( values )
10. Add sameAs from DBpedia JSON
import json
def get_value( item ):
return item['value']
dbpedia_uri = cells["DBpedia URI"].value
dict = json.loads(value)
same_as = dict.get( dbpedia_uri )[ "http://www.w3.org/2002/07/owl#sameAs" ]
values = map( get_value, same_as )
return ', '.join( values )

43. Step-by-Step 5/6
11. Add comment from DBpedia JSON
import json
def by_language( lang ):
def filter_by_language( item ):
return lang == item['lang']
return filter_by_language
def to_value( item ):
return item['value']
dict = json.loads( value )
dbpedia_uri = cells["DBpedia URI"].value
items = filter( by_language('en'), dict[dbpedia_uri]['http://www.w3.org/2000/01/rdf-schema#comment'] )
return ', '.join( map( to_value, items ) )
12. Add abstract from DBpedia JSON
import json
def by_language( lang ):
def filter_by_language( item ):
return lang == item['lang']
return filter_by_language
def to_value( item ):
return item['value']
dict = json.loads( value )
dbpedia_uri = cells["DBpedia URI"].value
items = filter( by_language('en'), dict[dbpedia_uri]['http://dbpedia.org/ontology/abstract'] )
return ', '.join( map( to_value, items ) )
Python/Jython

44. Step-by-Step 6/6
13. Add every other property from Wikidata (all at once) either using the property name
(instance of) or the identifier (P31)
Python/Jython

45. Data Publishing
N
extSteps
export
From
OpenRefine
custom
WP Plugin
WordPress
&
WordLift
(WP or Cloud)
Webpages
Knowledge Graph
data.wordlift.io
TSV

46. DO I REALLY NEED
SEMANTIC ANNOTATIONS AND
LINKED DATA?

47. 8. SITE LINKS
9. SITE IMAGE
CAROUSEL
10. TOP
STORIES FOR
NEWS
11. AMP
12. GOOGLE
FLIGHTS
13. PEOPLE
ALSO ASK
14. CATEGORY
15. IMAGES
16. VIDEO /
TRAILERS
17. LIVE
18. TOP SIGHTS
19. REVIEWS
20. BLOGS
21. KNOWLEDGE
PANEL
22. CAROUSEL
23. APPS
24. GOOGLE FOR
JOBS
25. RECIPES
26. SCHOLARLY
RESEARCH
27. WEATHER
28. GAME
SCORES
29. TWEETS
30. DISCOVER
MORE PLACES
31. SEND TO
GOOGLE HOME
32. PEOPLE
ALSO SEARCH
FOR
33. SEE
RESULTS ABOUT
34. WIDGETS
35. FOUND IN
RELATED SEARCH
36. QUOTES
37. EVENTS
38. DATASETS
SEARCH
1. STANDARD
2. TALLER
ORGANIC CARD
3. LOCAL
3-PACK
4. HOWTO
5. SHOPPING
6. RICH
SNIPPET
7. SITE
CAROUSEL
39. MOVIE
CAROUSEL
40. PODCAST
41. COURSE

48. 8. SITE LINKS
9. SITE IMAGE
CAROUSEL
10. TOP
STORIES FOR
NEWS
11. AMP
12. GOOGLE
FLIGHTS
13. PEOPLE
ALSO ASK
14. CATEGORY
15. IMAGES
16. VIDEO /
TRAILERS
17. LIVE
18. TOP SIGHTS
19. REVIEWS
20. BLOGS
21. KNOWLEDGE
PANEL
22. CAROUSEL
23. APPS
24. GOOGLE FOR
JOBS
25. RECIPES
26. SCHOLARLY
RESEARCH
27. WEATHER
28. GAME
SCORES
29. TWEETS
30. DISCOVER
MORE PLACES
31. SEND TO
GOOGLE HOME
32. PEOPLE
ALSO SEARCH
FOR
33. SEE
RESULTS ABOUT
34. WIDGETS
35. FOUND IN
RELATED SEARCH
36. QUOTES
37. EVENTS
38. DATASETS
SEARCH
1. STANDARD
2. TALLER
ORGANIC CARDS
3. LOCAL
3-PACK
4. HOWTOs
5. SHOPPING
6. RICH
SNIPPETS
7. SITE
CAROUSEL
??. ...
38. DATASETS
SEARCH
39. MOVIE
CAROUSEL
??. ...40. PODCAST
41. COURSE

49. Can I simply add structured data with my CMS?
Average
5.23 89 2.39% 35.32
2.15 57 1.60% 37.93
243.09% 156.56% 149.06% 6.88%
Clicks Impressions CTR Position
Semantically
Enriched
Non-Semantically
Enriched
W
O
RD
LIFT
Source: GSC data from a content publisher in Germany

50. Semi-automate Structured
Linked Data using NLP
1. Annotate/Link Named
Entities in posts and
pages
2. Build a Knowledge
Graph optimised for
SEO
3. Improve Impressions
and Clicks on Google
● 148.166 new users
from organic search
in 6 months
● +92.64% growth when
compared to similar
websites in Austria
W
O
RD
LIFT
Source: Google Analytics data from a travel brand in Austria

51. DRIVING FORCES
BEHIND AI
TRANSFORMATION
IS DATA FOR
GOOGLE ONLY?

52. Google Analytics goes Semantic
Traffic by Topic Topics mentioned in an article
W
O
RD
LIFT

53. “We want
raw data NOW.”
Tim Berners-Lee
Always link
data with other
data - data
becomes the
center of your
new
product/service
LEARN
IN
G

54. GRAZIE!!
Here is your
SPECIAL OFFER!!
https://wordlift.io/connecteddata

55. Resources
● More information about data cleaning:
https://www.aviz.fr/wiki/uploads/Teaching2018/2018-11-22-CleaningIntroTutor
ial.pdf
● Entity reconciliation with Openrefine:
https://www.youtube.com/watch?v=q8ffvdeyuNQ
● Openrefine tutorial 1:
https://www.youtube.com/watch?time_continue=8&v=cO8NVCs_Ba0
● Openrefine tutorial 2:
https://www.youtube.com/watch?time_continue=167&v=B70J_H_zAWM
● Wikidata for begginers:
https://upload.wikimedia.org/wikipedia/commons/6/64/Wikidata_-_A_Gentle
_Introduction_for_Complete_Beginners_%28WMF_February_2017%29.pdf