As a part of the ongoing student fake news detection challenge, I have asked the groups to share some of their “ah hah” moments working with the data. We did not want the teams to give away their best (potentially contest-winning) insights. Rather, just a few field notes that might inspire others to take part or think about the challenge in a new way. Here is version one of our “Student Field Notes: The Hunt for Fake News”:
- We thought it was really cool how you could advance filter each archive for the specific keywords we needed for our research. It took us really long to figure out the buckets and how toggle with the settings, but once we got it, everything went really smoothly.
- For our group, our research is focused on what spread faster and more effectively, Fake news or the stories that clarified the fake news is indeed fake. We found it is easy to focus on this very specific aspect of fake news, and harder to try to both apply sentiment analysis while collected quantitative data.
- Our ah-ha moment was when we discovered how to use the time-period tracker to help us determine when the tweets were created from any given dataset. For our future research, we are planning to use this time-period tracker to help us identify major events that occurred during that period. Thus, this will allow us to understand how twitter users feel based on the events that occurred.
- So far we have noticed two significant trends within the metadata that seem to be correlated with supporting Breitbart. First, is that many of the user profiles are dedicated almost exclusively to supporting Trump or conservatism in general. A second important trend pertains to the user descriptions. We found that a significant majority of the profiles that retweet or support Breitbart in one way or another mention their political stance in their user description. For example, many explicitly state that they support Trump, are conservative, and so on. This seems to support a concept from our literature review called “belief perseverence” which pertains to how things such as political ideology or religious affiliation are interconnected with how people identify themselves. The fact that these people consider being a Trump supporter such an important aspect of their identity that their brief user description revolves around it seems to be an interesting trend that we plan to explore further.
- Deduplicated archive -> filter out tweets containing “polls” in the text of the tweet and create bucket with that -> from that bucket we then would insert the name of a swing state and filter that particular state from our “dedupe polls bucket”-> Now we have buckets that contain the name of a swing state (Florida, Ohio, etc) -> create another bucket that only shows tweets about the polls following the election in a particular state i.e- We have a bucket that contains tweets that contain “florida” and “polls” in the text, but all the tweets are from after the primary in that state so that we don’t have units containing data about the primary polls. Using three codes (trump leading the polls, the polls are tied, clinton is leading the polls) The result from 120 units coded yielded 43% of the tweets said trump was winning, 15% tied, and 42% said clinton was winning. The actual results of the florida election were 49% trump and 47.8% clinton. Of course this could easily change as the volume of tweets increases but this was something I found interesting so far during my coding
- This moment came when we found out that 547 out of the 880 tweets that we coded were in the category of “Trump/Believer”. This data will allow us to employ our reasons for why human processing is not the golden standard, specifically in regards to hardwired errors of reasoning and prejudice. This proves that there is a significant correlation between trump supporters and believers of fake news. The focus going forward: Trump believers: code for prejudice or hard wired of reasoning. Examine them qualitatively to determine why they put in each category. Trump non-believers to see if they use intuition or reasoning in their argument.
We hope to find the southern states being bias towards the construction of President Trump’s and support the decision to construct it. Most of the states that are in support of this wall would be labeled as “Red States” according to the US Election Polls. While those states registered as “Blue States” would be opposed to the construction of President Trump’s wall. We also believe that different media outlets are a factor in how individuals are seeing the wall. We believe that conservative news media outlets and the way they distribute their information factors into how individuals see the construction of the wall.