2. Gender gap in political participation
● Strong evidence of gender gap in political decision
making positions
○ What about political participation more broadly?
● Literature suggests women in general participate less in
political activities and debates [Bode, 2017]
○ The higher the visibility of the actions, the stronger
the gap
● Women are more active on many social network sites
○ but what about online political debate?
4. Research questions
● Is there a gender difference in online political
engagement?
● Is there a gender difference in the influence on political
online debate?
● Are there different patterns among supporters of different
parties?
8. Dataset
● Tweets related to the 2017 UK general elections
● Posted between May 4th and June 9th, 2017
● All tweets mentioning or containing:
○ party names or acronyms;
○ names of candidates for Prime Minister;
○ broadly used keywords associated to the elections:
generalelection2017, generalelection17, ge2017, ge17
● Overall dataset: 4.5 million tweets
9. ● Retweets as a proxy for political endorsement [Conover 2011]
● Network of retweets with threshold 3
○ A→B: if A retweeted at least three tweets by B
Retweet network
● Louvain algorithm to detect communities (clusters) of users densely
connected to each other
○ Clusters in retweet network to represent political parties [Aragón
2016]
● Pagerank to measure relevance in the network
10. Gender detection
● Combined results from two established tools:
○ Gender detector, Sexmachine
○ Based on baby names from different countries’ census data and other
available sources
○ Accuracy over 97% for both genders, according to previous studies
[Karimi 2016, Knowles 2016]
● Procedure:
○ Extract first name from Twitter handle
○ take results where the tools indicate the same gender, or one
indicates no clear gender;
➡ Overall coverage: ~ 67% labelled users
Unlabelled users correspond in part to account with no gender
(i.e. parties, organizations, media...)
11. Results: gender differences
Which are men and women? (guess it!)
Size of each word is proportional to the difference in frequency with
respect to the overall dataset
12. Results: gender differences
Which are men and women? (guess it!)
Size of each word is proportional to the difference in frequency with
respect to the overall dataset
women
men ➡
13. Results
● After election surveys, broken by gender
○ Confirm higher support to Labour from women
14. Gender composition of (party) clusters
● 9 main clusters, mostly related to political parties:
○ percentage of women around 40% in most clusters
○ lowest proportion in the conservatives party (25%)
● Do women have lower pagerank?
○ Top 10, 100 and 1000 users for each cluster, according to pagerank
○ Mann Whitney U-test confirms the hypothesis for most clusters
Significance level of the Mann Whitney U-test: P-values < 0.05 (*), < 0.01 (**), < 0.001 (***)
16. ● Evaluate robustness of the gender detection method
○ Apply our method to already annotated Twitter datasets
● Look at different kinds of interactions (replies, mentions..)
○ Are there gender difference in the preference for different kinds of
interactions (e.g., are women less likely to engage in discussions
replying to other messages)?
● Study interaction within clusters and interactions that cross
ideological barriers
○ Which gender has higher tendency to create ideological echo
chambers?
Future work
17. ● Investigate gender homophily of (different kinds of)
interactions:
○ Are men more likely to interact with other men, and/or women with
other women?
○ Is homophily stronger in retweets, replies or mentions?
○ Is it stronger or weaker across ideological borders (inter-cluster
interactions)?
● Study of language style and emotions as in [Iosub, 2012]
○ Apply our method to already annotated Twitter datasets
Future work