It is with tremendous excitement that we unveil today the new 2016 Campaign Television Emotions Tracker! Building on the extraordinary success of our Campaign Television Tracker, which tracks the number of times each candidate is mentioned across major American television stations each day, the Emotions Tracker explores the emotional undercurrents of all those mentions, assessing complex emotional dimensions like "anxiousness" "apathy" "discord" and "self depreciation." In all, nearly 2,500 emotions from 26 different sentiment packages are assessed each day from January 1, 2015 to present, with a 48 hour rolling delay using a portion of GDELT GCAM's massive pipeline.
These are based on scanning the closed captioning records of each broadcast, which are subject to a degree of error. At this time the list of words appearing beside each mention of the candidate's name are delivered for processing as a word histogram ngram (1-gram), meaning the format is a list of all words appearing within 10 seconds before/after each mention of the candidate's name and the number of times that word appeared on that station on that day. Sentences are therefore broken into single words and ordered by frequency of appearance. This means that at present all dictionaries and algorithms below are operated in "bag of words" mode, which reduces the accuracy of some of the more sophisticated algorithms that are otherwise capable of incorporating phrasal matches and sequencing logic, though the majority of tools are unaffected by this limitation. The words "trump" and "bush" are removed from consideration due to appearing in several dictionaries, but otherwise no modifications beyond those reported in the GCAM documentation have been performed. A minimum of 25 words must have been monitored about the candidate on the selected stations in a given day for it to be included below, in order to help reduce noise.
Remember that all emotional scores were calculated completely automatically using the specified dictionary operating in "bag of words" mode. All output is therefore completely algorithmic and automated and in no way reflects any editorial decisions about any given candidate. Remember that these graphs reflect the words spoken on television in the immediate vicinity of mentions of the candidate and thus reflect how television news is framing the candidates, consisting of editorial decisions made by each station including both the words spoken by its anchors and reporters and brief clips of candidates speaking, rather than the language used by the candidates themselves. The Y axis scale and range varies by dimension, please see the documentation for each specific dimension for further details.
As we explore the results this new system generates over the coming months we will be ramping up to deploy a version of this system for all world leaders, allowing an unprecedented look at the emotional undercurrents of global discussion of the world's heads of state.
As always, email kalev.leetaru5@gmail.com with any questions. Permission is granted for any and all use of these graphs in media reports, please cite "Analysis by the GDELT Project using data from the Internet Archive Television News Archive" and report the specific emotion appearing in the Emotion dropdown, such as "Emotion: Anxiety by the Regressive Imagery Dictionary."