Characterizing Attitudes on Social MediaMisgivings from Misunderstandings
The frontier between biotechnology and the public acceptance is a frontier we have been losing lately. With major corporations such as Ben & Jerry's and Chipotle switching to GMO-free foods, the momentum of public opinion towards biotechnology is shrinking the ground underneath our feet. Stirring up the fear of synthetic biology is a misinformation, distributed faster than ever on the internet. If we do not begin tackling public perceptions, the inertia of this movement may defeat this field before it can reach its pinnacle.
This is not a trivial issue. Why do people share certain pieces of information over others? To study this question, we turn to the largest social media site on the web: Facebook. With 1.19 billion active users last quarter on this networking website, the trends in sharing and liking can give us insight into what draws people to certain information. Using a Facebook tool to computationally extract article titles, likes, shares, and comments from major news sources to try to identify patterns in the trait informally known as "click-bait", or content designed to attract visitors and attention to a particular source. The title of a link or post often the only major decider in whether someone will click.
With 24,462 New York Times posts, 13,758 Buzzfeed posts, and 14,069 Millions Against Monsanto (a leading anti-biotechnology group) posts, we sought to quantify linguistic metrics of these articles relative to "like" total. Using the NLTK and TextBlob modules of Python, we were able to analyze each title for their syntactically element and their sentiment and polarity. Additionally, we could crudely estimate emotional content by using the AFINN emotional lexicon, a dictionary of the most common emotional words with an associated magnitude. This gave us an array of metrics to try and estimate "likability" against.
What we found is exemplified below- reoccurring trends within individual news sources correlating very well for some news sources, and poorly in other cases. For example, title length positively correlates to "likability" for Millions against Monsanto and Buzzfeed, but is anti-correlative with the New York Times. In short, no systemic, cross-source patterns of metric correlation were found.
But these trends are superb within a news source! So we began to think these pattern arise from the followers of each news sources. People who follow Buzzfeed are slightly drawn towards articles with longer titles, whereas New York Timers like succinct information. This principle extends to the problem at hand- people who have a certain opinion follow news sources that strongly propagate that opinion and are likely subject to the same preferences in "click-bait". At this point, we performed a correlative analysis between each metric to identify the principle components of each news source. Each source had a unique signature as to what metrics strongly correlated to their "likability". For example, the New York Times had a strongly negative correlation to adjective content in titles and average title length as their first two principle components, followed by a positive correlation to proper noun usage. Indeed, we can also confirm this intuitively, as the New York Times is typically recognized as an objective news source that is frequently trafficked for breaking news stories, full of names and places with little adjective padding.
Finally, we return to biotechnology and how the scientists should talk to the public. Let's consider Millions Against Monsanto as our test group, which draws much of its following from those vividly fearful articles against GMOs. We can characterize the principle components of the news source, and design a title that others following this group would be more likely to click and possibly like. With the red bars in the figure below corresponding to negative correlation and blue corresponding to positive correlation, we can extrapolate this to decisions we would make when crafting articles representing our views. Lower adjective count and lower verb count in titles of articles are correlated to a higher number of likes. The power of this tool is limited only by our ability to target the correct demographics. With it, we hope to utilize what has proven effective in reaching the public in order to return science and reason to the evolving conversation about synthetic biology.