This Thanksgiving, while the Turkey was cooking in the oven and the television was warming up for all-day football, we decided to fire up the Twitter API to see what the Twittersphere was saying about the first game of the day: the beat-up Packers vs. the 10-year Thanksgiving-losing Lions. Going into the game, the Lions were 6-point favorites, three points of which were granted due to home field advantage; so from Vegas’ perspective this should have been a fairly well balanced game.
We were curious to see how contextual sentiment (positive/negative terms related to context-specific terms such as “Lions”, “Packers”, and “First Down”) fluctuated throughout the game. Measuring contextual sentiment is a strategy that many brands and marketers employ to understand consumer perceptions regarding brands, events and even sports teams themselves.
To start, tweets containing the words “Lions” or “Packers” were captured. From there, a custom positive versus negative linguistic dictionary was applied to the raw tweets, coding each instance of positive and negative words for each tweet. For a first exercise, we looked to see what the sentiment was for tweets containing the words “win” and “lions” (or “packers”).
Much like the Vegas odds, the sentiment was essentially the same for both teams an hour before kickoff; 82.5% of tweets containing the word “win” and “Lions” were positive, compared to 82.9% for the Packers. The twitter sentiment changed once the game was over, as expected due to the 40-10 Lions victory. From the end of the game to an hour later, the sentiment for the Packers shifted from 82.9% to 69.6% positive compared to an increase to 84.0% positive for the Lions. Many of the tweets (15%) discussed either the absence of Aaron Rodgers in the game or how “one guy” could make a huge difference in how the Packers performed.
Staying in the vein of quarterbacks, we were curious what else we could parse from the tweets by using quarterback names as a contextual filter. For this next example, we looked at the coded sentiment of tweets between 12:30 and 4:30 on November 28 (during the Thanksgiving Day game) containing either “Rodgers” or “Stafford”, as well as at least one of the following words:
Stud, Star, Winner, Beast, Awesome, Amazing, Wonderful, Incredible, Champ
Tweets using the above small contextual dictionary for Matthew Stafford were positive at a rate of 81% compared to tweets for Aaron Rodgers, which were positive at a 37% clip. This suggests that for tweets containing the word “Stafford” and at least one of the aforementioned words, 4 out of 5 were contextually positive in nature, referencing Matthew Stafford positively and associating him with one of those dictionary words (for example, “I’m a packers fan but Matt Stafford is a beast”, or, “I’m thankful for Matt Stafford being such a stud for the Detroit Lions”). For fun, an additional dictionary related to mustaches was thrown in for Aaron Rodgers, which accounted for 25% of his positive tweets and 15% of his overall tweets in the previous example, meaning that 25% of all tweets that contextually fit the above example were about Aaron Rodgers’ mustache, not the Packers team.
While the above exercises were somewhat silly and done for fun, they show how quickly sentiment can be coded based upon mined conversational text and how important contextual clues can be parsed from the coded data. Combining various contextual dictionaries can provide interesting and insightful information related to trends in conversation surrounding a team, company, or competitors.
via Business 2 Community http://www.business2community.com/twitter/fun-twitter-football-fan-sentiment-0703740?utm_source=rss&utm_medium=rss&utm_campaign=fun-twitter-football-fan-sentiment
Aucun commentaire:
Enregistrer un commentaire