2. Introduction
Firetoss and David Eccles School Business at the University of Utah
collaborated in a study to understand the relationship between
social media (specifically Twitter) and commercial advertising
during the recent 2017 Super Bowl.
The price tag for a 30-second commercial during Super Bowl LI
inched up to $5 million for 2017, a 4% increase from 2016 at $4.8
million. 56 brands made the investment and a total of 66 different
commercials aired during the almost four-hour nail-biting football
game between the New England Patriots and the Atlanta Falcons
on February 5th, 2017.
1,063,236 tweets were collected between 6 to 11 pm EST on
February 5th, 2017 with the help of 200 virtual servers using
Twitter Streaming API . After a thorough data filtering process,
259,279 of those tweets were related to the 66 different
commercials. Specific keywords, mentions, and hashtags used to
distinguish the commercial tweets from all other Super Bowl
related tweets were collected. These tweets were then parsed into
two categories, positive sentiment and negative sentiment,
relating to their specific brand/commercial.
The top performing commercial on Twitter for the Super Bowl LI
was from T-Mobile and their #UnlimitedMoves commercial
featuring Justin Bieber, Rob Gronkowski, and Terrell Owens.
58,616 tweets related to the ad were collected and 84.07% of
them were positive, while only 15.93% were negative. The
trending hashtags pertaining to this commercial were
#UnlimitedMoves – used 30,998 times, #BagOfUnlimited – used
23,169 times, and #TMobile – used 2,238 times.
Project overview
Understanding true effects of advertising on a consumer is
difficult with so many new media outlets. This project seeks to
understand the effects, positive or negative, that Super Bowl
advertising has on viewers by collecting data from user-generated
tweets about the brands and their ads.
3. Twitter allows for a connection to their Streaming API in
order to monitor activity and collect data on Twitter. 200
virtual servers with Python scripts and MongoDB were
deployed for this project to monitor tweets with hashtags
and tweets mentioning the advertiser’s accounts. Lists of
target accounts and keywords were prepared prior to the
Super Bowl however, many trends that arise can not be
pre-planned. This required constant interaction and
monitoring of live television by the team managing the
project. This collection was done between 6 to 11 pm EST on
February 5.
The collected data was then cleaned and analyzed in order
to calculate the number of tweets relating to the Super Bowl
ads. Using Python NLTK (Natural Language Toolkit) Text
classification packages, these tweets were classified to
determine if the content contained positive or negative
sentiment.
Firetoss teamed up with a group of graduate students led by
a professor from the Eccles School of Business. The students
and the professor are affiliated with the M.S. in Information
Systems program and the newly created M.S. in Business
Analytics program at the University of Utah. This is the
second year that data has been collected around Super Bowl
advertising in the Eccles School of Business. This year’s data
collection was broader and more comprehensive, compared
to previous year’s data, due to an increased number of
servers and a more advanced technology stack.
Findings
4. T-Mobile and their #UnlimitedMoves ad was the front-
runner commercial this year. 58,616 tweets were collected
that either mentioned T-Mobile’s accounts or one of the two
hashtags -- #UnlimitedMoves and #BagOfUnlimited. AirBnB
was a close second place with 47,446 tweet counted.
The least talked about brand from Super Bowl LI was King’s
Hawaiian. Amongst the 1,063,236 tweets collected only 2
tweets related to the brand. Fiji Water, Turkish Airlines,
Michelin and Weather Tech weren’t far behind.
Activity on twitter was geared more toward the content of
the ad as opposed to the brand name itself. Both the
#UnlimitedMoves and #BagOfUnlimited hashtags were used
more frequently compared to the T-Mobile brand hashtag
(#tmobile). The same was true for Airbnb’s #weaccept
campaign versus #airbnb.
“There is no such thing as bad publicity.”
In addition to monitoring brand performance, this project
examined sentiment for each tweet. Using Python NLTK
text mining packages, sentiment analysis was performed to
determine whether the tone of each tweet was positive or
negative. The text classifiers first tokenize each tweet into
groups of words (known as features) with irrelevant words
(stop words) removed. Stop words are commonly used words
that are generally considered useless for sentiment analysis
(e.g. he, she, the, etc.). Then the team used the Python
TextBlob package (a supplement to NLTK) to label each word
in the tweet based on a set lexicon (dictionary) of
pre-determined positive-negative terms. Finally an overall
sentiment index was generated for the tweet which was
aggregated for each brand or commercial.
Once again T-Mobile not only led the group in volume but
the majority of tweets related to their brand were found to
be positive (84%). AirBnB was a very close second on the
sentiment scale (83%). It is notable that most of the tweets
this year were positive in sentiment across the board. The
volume of tweets spiked in the 2nd quarter when the top
commercials #weaccept, #cocacola and #unlimitedmoves
were broadcasted. In the 4th quarter, the gap favoring
positive index increased as tweet volume spiked. The peaks
and valleys were more volatile with the positive tweets
overall as compared to the negative tweets.
Trending, positive, and negative word clouds provided a
clearer picture of the level of attention given to different
brands and ads during the game. Although the overall result
is consistent with the volume of tweets for each brand, these
5. word clouds show vividly the spread of positive-negative
sentiment. For example, both of the ads #weaccept and
#avosecrets appeared in both the positive and negative
word clouds implying a presence in the positive as well as
(perhaps not desirable) negative spaces. Ads like
#unlimitedmoves, on the other hand, were mostly positive.
The chart above shows distribution of tweet volume (5-
minute interval) between 5 to 11 pm EST. These tweets
counts include those from the halftime show and the game.
We note a close correlation between broadcast times of top
commercials (and half-time show) with volume of tweets
generated implying a close relationship between social
media and Super Bowl commercials.
The project also examined the relationship between time
zone and sentiment of the tweets collected. The
percentage of negative tweets to positive tweets held
pretty consistent across all United States time zones with
the Eastern Time Zone being slightly more negative than
the average and the Pacific Time Zone being slightly more
positive than the average.
Other factors such as time of day, and affinity towards
twitter as a social platform may have played a part in this
finding.
6. Summary
Firetoss and David Eccles School of Business, University of Utah
collaborated in this research in exploring the relationship
between social media (specifically Twitter) and 2017 Super Bowl
commercial advertising. A framework comprising of Python
scripts and MongoDB was built to collect tweets from Twitter
Streaming API during the game between 6 to 11 pm EST. Over 1
million tweets were collected and later processed to extract
volume and sentiment for each commercial and brand. We
found T-Mobile was the most talked-about brand with over
58,616 tweets. We also noted that certain brands have a wider
distribution of positive-negative spread than others implying a
higher negative public response than other brands. This project
adds to the relevance of social media (specifically Twitter) as a
brand monitoring tool in today’s well-connected Web 2.0
environment.
ADVISOR
Chong Oh
Assistant Professor
Operations & Information Systems
University of Utah
GRADUATE STUDENTS
Vikal Gupta
Shambavi Khare
Venkatesh Sharma
Andrew Vernon
Savan Kumar
SPONSOR
Tony Passey
Assistant Professor
Department of Marketing, University of Utah
Founder + CEO, Firetoss
Silvia Potempa
Social Media Coordinator
Firetoss