Did Russia Win the 2016 Elections ?

DESARZENS Lionel, JOURNE Adrien, and LEFEVERE-LAOIDE Aindrias

November 8th, 2016

Against all voters survey, Donald Trump becomes the 45th US president. How could it happen ?
We live in a society where social media platforms have taken a great importance. Social networks such as Facebook, Twitter or LinkedIn have become very powerful tools to spread a message. They allow a direct connection between parties or companies and citizens, and recent events such as the Arab Spring have shown they can be used as a driving force. However, in such media which favor free expression, the control is weak or inexistant and thus fake information can spread quickly, against the social good.

The aim of this datastory is not to conclude by certainty that Russian have played a role during the US elections. We want to try to answer a few questions such as

Did Russia try to give more chance to Trump to become president?
Did Russia try to dishonor or demean Hillary Clinton ?

An overview of social media

What is the IRA ?

No, we are not talking about the Irish Republican Army, but about the Internet Research Agency, a russian compagny specialized in online propaganda. The IRA leads these compaingn to serve the Russian interests.
You might ask why we are talking about the IRA. In fact, many hypothesis might say that the IRA have been leading a disinformation campaign on Twitter against Hillary Clinton during the 2016 US elections.



And what about twitter ?

Twitter is an american social media used to communicate with each other by posting or interacting with messages called "Tweets". The interaction between two users can be explained as follow :

  1. When Caroline sends a tweet, all of his followers gets her tweet (with the red dots).
  2. If Emilie tweets, Caroline will not get the message, as the is not following her.
  3. As Caroline follows Manu, she will get his tweets.
  4. Caroline added an hashtag in her tweet (#joconde). An hashtag is a keyword that allows to group all the conversation talking about the same topic.



The 2016 US Elections

You may have heard during the elections, that a lot of scandales have been divulged. Just as a reminder, we explain some of them to you.

On Clinton's side

  1. Critisizing Trump's electorate: 09/09/2016 During a talk, Hillary Clinton sparks controversy by suggesting half of Trump's electorate belongs to "a basket of deplorables". Trump uses this to draw sympathy and to depict Clinton as someone who doesn't respect all Americans.
  2. Feeling unwell during commemoration ceremony: 11/09/2016 During the commemoration ceremony, Hillary Clinton is forced to leave as she is not feeling well. She is diagnosed with pneumonia. Trump uses this as an argument that Hillary Clinton is not strong enough to govern the USA.
  3. The Clinton emails: 28/10/2016 On March 2nd 2015, the New York Times reveals that Hillary Clinton used a private server to access her emails when she was Secretary of State, but the investigation suggest that she was merely careless and no suit is filed. However, FBI director James Comey announces 11 days before the elections that new emails have been found and so that the case will be reopened. Donald Trump will use this affair to discredit the democrat candidate.
  4. The pizzagate: 30/10/2016 The pizzagate conspiracy theory originates from a tweet by a laywer from New York asserting that there exists a pedophilia network around John Podesta, Hillary Clinton's compaign director. This tweet quickly became viral, and the hashtag Pizzagate appeared on November 7th on twitter (one day before the elections).

On Trump's side

  1. Reaction to slain US army captain: 30/07/2016 Trump held disrespectful word towards the father's of a US army captain who died in Irak in 2004. The public opinion reacted strongly to the interview and many Republicans attacked him for his behavior did not correspond to their values.
  2. Controversy over tax returns: 01/10/2016 On begininning of October, the New York Times reveals that Donald Trump may have declared in a doubtful way the bankruptcy of his casino, which may have led to him paying less taxes. Hillary Clinton attacked the republican on his responsability in this case.
  3. 'Access Hollywood' tape: 07/10/2016 The Washington Post published a tape from 2005 where Donald Trump describes how he attempted to seduce a woman. He holds sexist, and degrading comments about women. The action he formulates are described as sexual assault. This leads to the loss of the support from some of his voters.

Some Statistics

Over the past month preceeding the elections, many tweets have been posted using different hashtags. We can notice two peaks on this graph. The first one appeared on mid-july 2015, which seems to be correlated with the beginning of the electoral campaigns. The second appears a few weeks before the elections. It seems that the more we approached on the elections, and more tweets were posted by the IRA.




The idea behind this graph was to see the proportion of users that are really active on their Twitter account. It seems that we can learn more about it.
First, don't hesitate to go through the graph with the mouse in order to display the number of tweets for a given user.

Now, that you have been thinking about this graph, you may have noticed that only a small number of user published more than one tweet per day (about 400 users over 1800), and less that 300 users published more than two tweets a day. This learn us that most of the account in the campaign of the IRA were not really active.



The idea behind this graph was to look for the impact of each category. The impact takes into account the number of follower of the user, the number of like, retweet and quote of the tweet. To give an example, if a tweet has been posted with one follower, and later, one like, one retweet and one quote have been posted, its impact score would be 4.
The aim of looking through the mean impact score of each category was to be help us to conclude about any campaign of disinformation about the elections.
We can notice that the category with the most impact correspond to the news feed. Followed by right troll.
A right troll is a propaganda tweet that is here to promote the right. In this case, most of the right trolls were here to support Donald Trump during the elections. But as the score of the left troll is of 90, we cannot conclude anything about it.

Sentiment analysis

The aim of this part was to be able to see in which categories belongs the negative and the positive tweets. For this purpose, one can look for the mean of the sentiment for all tweets belonging to each category. We wanted to see if we are able to conclude anything about the nature of the tweet and its category.

When looking to the interactive graph, nothing can be concluded about the mean of the sentiment. But what if we look on the sentiment evlution over time?
To make your own conclusion, you can first look to the graph below.

Now that you may have made your conclusion, we can compare. You may have noticed that the most oscillating negative sentiment comes from left and right troll. This is logical, as it increases as the elections come closer. Unfortunately, this graph does not allow us to make any conclusion about the two main questions. Indeed, it seems that the sentiment of the tweets oscillate in the same way for left and right trolls.

Scandale research

The numerous scandals which marked the last presidential campaign on both sides created a climate tension and uncertainty in the country. One would expect the IRA to target those events to try ad make an impact on the public opinion and spread dissension. Therefore, we had a look at the main topics of the tweets of the IRA during the day of the scandal. Surprisingly, the users seemed more interested in discussing PokemonGo than Trump’s attack on the parents of a soldier slain in Iraq! They did seem to react to important scandals such as the Hollywood Access tape, however.


We wanted to go further than this global view on the reactions of the IRA twitters by seeing if some politically-oriented troll categories, such as Right troll or Left troll, reacted more intensely to specific events. We added other milestones of the campaign, which aren’t scandals to say so, but had a significant impact during the campaign. For example, we could expect trolls to react to Trump reaching the delegate threshold or to a thumping win from Bernie Sanders against Hillary Clinton in the primaries. Looking at the activity per category (ie the total number of tweets), we see that the Newsfeed trolls are very active on most events, as expected. We also see that Right trolls react very strongly to scandals linked to Clinton’s campaign, such as the FBI email scandal or the Podesta pedophilia affair. Left trolls reacted strongly to the Pope being doubtful of Trump’s Christian faith. However, this activity does not show what feelings were expressed by the trolls in their tweets: Did they attack the opposite party? Did they commiserate, or stay neutral?


We did a sentiment analysis of the tweets for each category and for each milestone. We see that it is hard to see a difference on the means of the tweet categories, which probably means that the trolls do not adopt a common strategy that involves being very critical or very praiseful. However, we can see differences for some events, such as when Obama endorses Clinton or when Clinton insulted Trump’s electorate, which created positive sentiment in the Left troll tweets.


To come back to our first questions :

Did Russia try to give more chance to Trump to become president?
Did Russia try to dishonor or demean Hillary Clinton ?

we can say that indeed the IRA has tried to influence the US elections by having a tendency to give less impact to the left trolls and the scandales relative to Donald Trump. Given this, we can also conclude that the IRA does not only speak about politics, as many other categories of user exists. It seems that the IRA follows the general tendency of Twitter in order to create more impact when trying to propagate a propaganda message.