A Focused Analysis of Twitter-based Disinformation from Foreign Influence Operations

Abstract

We present a focused dataset of misinformation in the US 2016 election. Our dataset is made up of two different parts: set (1) was collected between November 9th 2016 and March 31st 2017 using the following keywords: #MyVote2016, #ElectionDay, #electionnight, @realDonaldTrump, @HillaryClinton to tweets related to the election campaign. This collection yielded a total of 57,379,672 tweets. Set (2) was retrieved from [1] and consists of 2,946,220 tweets ranging from June 19th, 2015 to December 31st 2017. To ensure tweets corresponded only to the presidential campaign in the United States, we restricted tweets in set (2) to those before March 31st 2017, yielding a total of 1,244,480. Of these, we only retain original tweets (i.e., we purge `retweets' or duplicate mentions). It is important to note that the set (2) corresponds to accounts identified by the FBI as belonging to a foreign influence campaign. For more details see [1].

For the negative samples (samples which are not disinformation), we remove all tweets that have any author level content that corresponds to accounts in (1). We also use tweets only in English. To ensure tweets in the sample are relevant, we restrict the tweets to those that belonged to the US as the geographical location in the metadata. Specifically, we restricted our sample to tweets that have geolocation coordinates to be within the US. We used Twitter's API to ensure that tweets we considered were coming from users whose accounts have not been suspended by Twitter four years after the events and consider this to be a proxy for valid accounts.