A Digital Nose For News:

Re-Inventing Journalistic
Sourcing?
Dr Neil Thurman
n.j.thurman@city.ac.uk

@neilthurman

Probability of Computerisation

We dont think it is desirable

that journalism is done with
algorithms Email to Konstantin Drr

Tools for computational news

detection & verification

Social and professional contexts

How they work.
Biases?
Changes in journalistic work and
outputs

SocialSensor is a single tool that

quickly surfaces trusted news
stories from social media with
context.
A Single Tool: one platform, one interface
Quickly: in real time
Surfaces: automatically discovers and clusters
Trusted: automatic support in verification process
Material: any material (text, image, audio, video =
multimedia)
Social Media: across relevant social media platforms
With Context: location, time, sentiment, influence

1. To what extent does social media

break news first, and how much
news does it carry thats not
covered elsewhere?
2. And, in addition to surfacing news,
what else can tools like Social
Sensor do?

etc

Broke
1st or
1st=

Source: Osborn, M. and Dredze, M (2014) Facebook, Twitter and Google Plus for Breaking News: Is
there a winner? Proceedings of the Eight International AAAI Conference on Weblogs and Social Media.

Where Twitter was first

Source: Petrovic, S. Osborne, M, et al, 2013 Can Twitter Replace Newswire for Breaking
News? Proceedings of the Seventh International AAAI Conference on Weblogs and Social
Media

Only carried on Twitter

Death of Canadian Ice Hockey player

Source: Petrovic, S. Osborne, M, et al, 2013 Can Twitter Replace Newswire for Breaking
News? Proceedings of the Seventh International AAAI Conference on Weblogs and Social
Media

Only carried on Twitter

Identification of looters car,

London riots 2011

Source: Petrovic, S. Osborne, M, et al, 2013 Can Twitter Replace Newswire for Breaking
News? Proceedings of the Seventh International AAAI Conference on Weblogs and Social
Media

Newspaper articles / Broadcast news packages

quoting social media (per outlet per day)
4

0
8 UK & Dutch Quality &
7 Elite US newspapers
Tabloid Newspapers
and TV stations (2010-11)
(2011)
Source: Broersma and Graham (2013)

Source: Soo Jung Moon and Patrick

Hadley (2014)

2 quality Flemish
newspapers (2013)
Source: Paulussen and Harder (2014)

 Providing editors with information on

trends in popularity and sentiment
Alerting newsrooms to ongoing
developments in running stories and
providing contacts and content
Giving journalists information on the
reliability of contributors and the
veracity of content

Source: Broersma and Graham (2013)

Types of story using Tweets as source (%)

40
35
30
25
20
15
10
5
0

Source: Broersma and Graham (2013)

Types of social media contributor quoted in UK & Dutch newspapers

Source: Broersma and Graham (2013)

Types of social media contributor quoted in UK & Dutch newspapers

Source: Broersma and Graham (2013)

the biggest problem is how to exploit the vast

amount of content in social media with a small
team MSN journalist (pers. Comm.)

we need algorithms to take more onus off

human being, to pick and understand the best
elements New York Times Social Media
Team member (pers. Comm.)
Current tools arent powerful enough CNN
social media expert (pers. Comm.)

Representative non-events:
Jaime pas Bieber, 1D le rap et plein dautres
conneries. Vous pouvez mamener 500 haters
je changerai pas davis.
This wine is going down a lil to smoothly. Here
comes trouble.
LIMA HARI BULAN LIMA ! KEK SEBESAR
GUNUNG ! kena belajar buat
kek ni, tinggal 2 bulan jea lagi -.-
RT ZorianRamone: Happy Bday
Less than 5% of Tweets carry newsrelated
content

Event detection in Twitter

Source: Osborne &

Benjamin Van
Durme in CallisonBurch

Running traditional First Story Detection

systems on Twitter produces a mass of false
positives
Less than 1% of events detected in
Twitter are news related

Examples of false positives

Juicy Couture, Ed Hardy, Coach, Kate Spade and many

more! Stay tuned for more brands coming in http://. . .
i lovee my nephew hair :D
Going to look at houses tomorrow. One of them
is & right behind Sonic Taco Casa. If I live there, I
might weigh 400 lbs within a year.
Hope a bad morning doesnt turn into a bad day...

Source: Osborne & Benjamin Van Durme in Callison-Burch

Example list used to seed News Hound database

Lyse Doucet, Chief International Correspondent: @BBCLyseDoucet

Gavin Hewitt, Europe Editor: @BBCGavinHewitt
Lucy Williamson, Paris Correspondent: @LucyWilliamson
Fergal Keane, World Affairs Correspondent: @FergalKeane47
Chris Morris, Correspondent: @BBCChrisMorris
Christian Fraser, Correspondent: @ChristianFraser
Damian Grammaticas, Correspondent: @DNGBBC
Simon Wilson, Europe Bureau Editor: @Siwilso
Piers Schofield, Senior Europe Producer: @Inglesi
Natalie Morton, Senior Producer: @NatalieMortonTV
Imelda Flattery, Senior Producer: @ImeldaFlattery
Frank Gardner, Security Correspondent: @FrankRGardner
Gordon Corera, Security Correspondent: @GordonCorera

Scoring newshounds
Criteria

Score

On seed list?

150

Each seed that follows them

Each seed they follow

Send at least 10 tweets per

day

50

Verified with Twitters blue

tick

25

Presence on at least 50 Twitter 25

lists

Dispersion of news on social media

Who are the news hounds?

60

52

50

Males

40
30
20
10
0

Female
23

Institutional
Accounts

Who are the news hounds?

News hounds scoring system

Criteria

Score

On seed list?

150

Each seed that follows them

Each seed they follow

Send at least 10 tweets per

day

50

Verified with Twitters blue

tick

25

Presence on at least 50 Twitter 25

lists

Computational journalism
tools
Tuned to: Most mentioned, Most
followed & Most Vocal?or

Agents for change?

Research
Potential

The Personal Brand

The Personal Brand

@wblau

@Le_Figaro
#6758

#6924

#43

Burstiness

News values

Tony Harcup & Deirdre O'Neill, 2010

Alternative News Values?

....political, structural and natural root
causes and contexts
the accounts of the people involved
rather than third interpretations by a
third party.

Source: NGO-EC Liaison Committee, 1989

Verification

Principles for social media verification

Content
Contributor
Context

Content

Contributor

Context

Tweets per min

3500
3000

2500
2000
1500
1000
500
0

Bostom Marathon
Bombings 2013

US tornadoes 2010

And test=

If training=

Computing Contributor Credibility

No. of Tweets

HISTORY

Frequency

HISTORY

No. of followers

POPULARITY

No. of follows

POPULARITY

Retweets

INFLUENCE

Human vs. algorithmic evaluation of

social media contributors
Grade 0-9

Standard
Deviation

Journalists
Evaluation

5.67

2.10

Truthmeter
Evaluation

5.71

2.45

Inactive!

Yeah, but
check out his
followers!

5/10

What? Shes
Deputy Leader of
the Labour Party!

Weight
No. of Tweets

HISTORY

Frequency

HISTORY

Digital Nose for News

The social and professional contexts

How they work
Biases?
Agents of change?

Digital Nose for News

Rely on journalistic input

Success measured against
journalistic ground truth
Created in our own image

Digital Nose for News

Its biases are ours:

short-termism
Personalization
Demography

To enjoy the privilege

of making stockings for
everyone is too
important to grant to
any individual

Thank you!
Dr Neil Thurman

@neilthurman
n.j.thurman@city.ac.uk