1. Big Data Sources – Web, Social media and Text Analytics Social media and official statistics
Piet Daas, Ali Hürriyetoglu
THE CONTRACTOR IS ACTING UNDER A FRAMEWORK CONTRACT CONCLUDED WITH THE COMMISSION

2. Social media and official statistics
Overview
Social media
Potential uses
Risks when using social media

3. Social media platforms
In the Netherlands
1) Twitter
2) Facebook
3) Hyves

4. Dutch are very active on social media! Readily available
Around 70% according to a recent study
Readily available
Potential source of information on what goes on in ‘Dutch’
society (active on social media)
As an auxiliary source? In addition to survey and admin. data
Any other possibilities?
Explore! lt 4
Map by Eric Fischer (via Fast Company)

5. Examples of Social media studies performed at Statistics Netherlands
Compare Twitter topics and statistical publication theme's
Social media sentiment and consumer confidence
‘Measure’ other basic emotions in social media
Social cohesion and Twitter (for an area in NL)
Selectivity: background characteristics of Twitter users
Event detection on Dutch highways
Profiles of companies vs. persons (self-employed) …

6. 1. Twitter topics What are people talking about?
Decided to use Twitter
A. Collect as many Twitter usernames (ID’s) as possible
Expected between 150,000 – 400,000 (as indicated by studies of others)
B. Identify users from a particularr country
Location info provided (country, municipality)
C. Collect tweets from all those users (if publicly available)
Max 3600 per user, 200 per request
D. Identify topics discussed in the tweets

7. A. Collect usernames Started with a user with many followers
Breadth first algorithm / snowball sampling
Started with a user with many followers
A famous Dutch politician with 79,798 followers
Collect the followers of her followers etc.
By Twitter REST API, 12 user accounts and PHP-scripts
After 4 weeks we obtained
4,413,391 unique users (id’s) !
Collected user id, username, location and profile information

8. B) Identify ‘Dutch’ users
By using location information provided
A considerable number of users do this
Checked the location names provided
Inclusion and exclusion list
A total of 380,415 (~9%) users were identified as located in the Netherlands
38% of the users, 1,661,467, provided no location info

9. C) Collect Tweets
For the 380,415 users the 200 most recent tweets were collected
A total of 12,093,065 messages was obtained
39% of the users had no ‘tweets’
Some characteristics

10. D) Identify topics Hashtags (1,750,074 with 1 hashtag, 14.5%)
Used 2 approaches
Hashtags (1,750,074 with 1 hashtag, 14.5%)
Hashtag (#) identifies ‘keyword’
E.g. #ned, #fail, #wk2010
Manual and text-mining approach
Without hashtags (10,330,613 in total, 85.4%)
Manual (sample)
Text-mining approach failed here
Result of the large ‘Other’ group

11. Findings: Dutch Twitter topics
(3%)
(7%)
(3%)
(10%)
(7%)
(3%)
(5%)
(46%)
12 million messages

12. 4. Social cohesion
Can the relations between Twitter users be used to study other social phenomena
E.g. Social cohesion
Social cohesion is the set of characteristics that keep a group able to function as a unit. What constitutes group cohesion really depends on whom you ask. For instance, psychologists look at individuals' traits and similarities among the group members. Social psychologists treat cohesion as a trait that combines with others in order to influence the way the group does things. Sociologists tend to look at cohesion as a structural issue, measuring how the interlocking parts of the whole group interact to allow the group to function. Beyond all these disciplinary differences, there are some generalizations we can make about how groups function as a unit.
We studied a particular area in the Netherlands and collected the accounts of all active users on Twitter.
We subsequently looked at their relations (friends and followers)

13. 4. Network Visualization (2 Tweeps or more)
About 2000 active Twitter accounts
Tweeps: Bidirectional following
Colours indicate municipalities 13

14. 4. Network data: ‘Twitter friends’
About 2000 active Twitter accounts 14

15. 4. Risk of studying a small subset
Around 2000 Twitter users were found
Inhabitants in that area ~40.000
Is a small part of the population
Can be very noisy data (non-representative social media)
Difficult to generalize findings
Need more info for that

16. Its all about subpopulations
Twitter
Its all about subpopulations
The Netherlands
Village study

17. 5) Selectivity: Twitter user characteristics
Only a part of the Dutch are active on Twitter
If we want to use this source we need more info
By determining their ‘background’ characteristics
Such as gender, age, income, level of education etc.
What are the possibilities?
Feature extraction is the way to go
For gender

18. 4) Picture
3) Messages
content
1)Name
2) Short bio

19. Studied a Twitter sample
From a list of Dutch Twitter users (~ )
A random sample of 1000 unique ids was drawn
Of the sample:
844 profiles still existed
844 had a name
583 provided a short bio
473 created ‘tweets’
804 had a ‘non-default’ picture
409 Men (49%)
282 Women (33%)
153 ‘Others’ (18%)
companies, organizations, dogs, cats, ‘bots’..
Default Twitter picture

20. Gender findings: 1) First name
Used Dutch ‘Voornamenbank’ website (First name database)
Score between 0 and 1 (female – male); 676 of 844 (80%) names were registered
Unknown names scored -1 (usually companies/organizations)

21. Gender findings: 2) Short bio
If a short bio is provided
Quite a number of people mention their ‘position’ in the family
Mother, father, papa, mama, ‘son of’, etc.
Sometimes also occupations are mentioned that reflect the gender (‘studente’)
155 of 583 (27%) indicated their gender in short bio
Need to check both English and Dutch texts

22. Gender findings: 3) Tweets content
In cooperation with University of Twente (Dong Nguyen)
Machine learning approach that determines gender specific writing style
Language specific: Messages limited to one language (but extendible)
437 of 473 (92%) persons that created tweets could be classified

23. Gender findings: 4) Profile picture1 3 2
Use OpenCV to process pictures
1) Face recognition
2) Standardisation of faces (resize & rotate)
3) Classify faces according to gender
- 603 of 804 (75%) profile pictures had 1 or more faces on it

24. Gender findings: overall results (1)
Diagnostic Odds Ratio (log)
First name
4.33
Short bio
2.70
Tweet content
1.96
Picture (faces)
0.57
Diagnostic Odds Ratio =
(TP/FN) / (FP/TN)
Random guessing
log(DOR) = 0
Multi-agent findings
Need ‘clever’ ways to combine these
Take processing efficiency of the ‘agent’ into consideration 24

25. Gender findings: overall results (2)
Combine findings in the best possible way
Unassigned (%) Approach used
844 (100%) 1. Use short bio scores (very precise for females)
689 (82%) 2. Use first name scores
153 (18%) 3. Use Tweet content
29 (3.4%) 4. Use picture
20 (2.4%) 5. Assign male gender
Final log(DOR) is 7.02, an accuracy of 96.5%!
What about other characteristics?

26. Concluding remarks Social media is a difficult source to study
Contains a lot of ‘noise’
Social media is a secondary data source
Produced for a ´reason´ not identical to way we want to use it
A paradigm shift is needed (for primary oriented people)
Try to improve quality (reduce noise; apply filter)
Benefit from the large volume of data available
Analysing texts and pictures is different/difficult
Learn by doing and by cooperating with experts
Interesting results but
It is a relatively new area for official statistics, so a lot needs to be checked (this takes time)
There are possibilities for official statistics but
Is everybody in the office ready?

27. Risks when using social media
Platforms : There are several platforms
Populations: Mixture of several target populations (e.g. persons, companies, others)
Selective: Only measuring the part of the population active on it
Dynamic: Users may hop on and off
Skewed: Not every account is equally active over time
Language: Has its own vocabulary and abbrev. 27

28. Introduction to Twitter access and API
Exercises

29. Various ways of accessing Twitter

30. What is API access?
In computer programming, an application programming interface (API) is a set of routine definitions, protocols, and tools for building software and applications.
An API expresses a software component in terms of its operations, inputs, outputs, and underlying types, defining functionalities that are independent of their respective implementations, which allows definitions and implementations to vary without compromising the interface. A good API makes it easier to develop a program by providing all the building blocks, which are then put together by the programmer.
Its an interface by which you can interact with an application

31. Twitter Rest API object
In R:
via twitteR package
Link:
web/packages/twitteR/
In Python:
via tweepy
Link:
Code examples will be given!
More on: 31

32. Rate limits !!

33. Exercises 1) Connect with Twitter API (need the keys)
2) Find a famous Twitter user from your country and extract some basic info
id, name, nr of followers, followers list, description, location (etc)
3) Download n tweets of this user (n = 200)
check for duplicates and retweets
4) From tweets get most frequent used
Hashtags (#word), links ( mentions
5) Create a wordcloud of words in tweets
With and without stopwords, and without hashtags 33