Sentiment Analysis & Newspapers: Why So Negative?

A long time ago (in a galaxy far far away) my friend and I decided that for a month (30 days) we would harvest the news stories from a few of the most popular news sources we knew. We figured that there was going to be a fairly negative sentiment trend given that news sources generally cover negative stories. This was just from-the-hip thinking so we decided that twice a day, one in the morning when I got to work, and once in the evening when he returned from school we would grab the mean sentiment of the news outlets. We chose the Bing sentiment analysis method which we chose because in other cases it tended to be used for shorter texts and given that we would be mostly grabbing headlines/snippets we thought why not. Here are a couple news sources we looked at:

  • Wall Street Journal
  • USA Today
  • Fox News
  • Chicago Tribune
  • Denver Post
  • CNN
We included sites like the Blaze because, in reading their website, one could see they would be a nice reference point for if the negative sentiment calculation was actually working. Some things became evident within a few days. Most of the news outlets¬†hovered around a fairly negative mean sentiment most of the time. Again this confirmed our thoughts but a few of the outlets were very negative. During these thirty or so days a few world events happened that we thought were of note and wanted to see the sentiment by outlet on those days. The first was Hillary Clinton’s announcement of her candidacy for president. The other was the Baltimore Riots and Nepal Earthquake which were reported by most major news outlets around the same time as each other.

My opinions on these results are a little mixed. News is naturally negative in word use, which over time would strengthen the negative sentiment mean score. So the relatively dramatic jumps from the points showing specific days need to be taken with a grain of salt. As we progressed through the days leading up to the end of this little project we were thinking to ourselves…

Is emotion in the news good? Does it bias readers? Is there a correlation between the political leanings of the news outlet and their mean sentiment…



These are left for someone else to answer and decide. For now it was just fun to make this chart with
the Plot.ly API in R.

Below is the code to generate this Plotly chart in R.