§14.2
Case Study: Goose Island Acquisition Sentiment
Goose Island is a better sentiment case than it first appears because the obvious question is too simple. In March 2011, Anheuser-Busch acquired the Chicago craft brewery. A naive social-listening report would ask, "Did sentiment go negative?" The better NLP question is: which part of the tweet stream is product opinion, which part is acquisition news, and which part is just noisy social-media plumbing?
The file contains 19,687 tweets from January 1, 2011 through June 30, 2011. The source already labels each tweet as Pre, Acquisition, or Post. The event window is short and loud: March 28-29 is the acquisition-news window, and March 30 starts the post-acquisition period.
This case uses a transparent lexicon rather than a black-box sentiment model. That is deliberate. Students can see the measurement choices, audit the cue words, and understand why a single sentiment line would be misleading.
The Executive Question
Did the Goose Island acquisition change public sentiment, or did it mostly change the composition of social chatter?
The answer is mixed. The acquisition window is visibly more negative, but it is also dominated by news links and acquisition vocabulary. Post-event conversation returns toward product talk, with acquisition language lingering in the background.
Scope, Measures, and Caveats
Before scoring sentiment, define what the row can and cannot represent. The dataset is not a panel of consumers, purchases, or brand loyalists. It is a public tweet stream with lots of URLs, check-ins, auto-posts, beer-app activity, and ambiguous mentions.
| Item | Case choice | Why it matters |
|---|---|---|
| Document | One public tweet mentioning Goose Island | The row is social chatter. It can include product opinion, check-ins, news links, jokes, and false positives. |
| Event window | The source labels March 28-29 as the acquisition-news window and March 30-June 30 as post-acquisition. | The source labels create a pre/event/post structure; the analysis treats that structure as descriptive. |
| Sentiment | Transparent teaching lexicon: positive product words minus negative/friction words, plus a separate acquisition-cue dictionary. | The score is transparent enough for teaching, but it is not a human-coded sentiment label. |
| Noise controls | URL share, check-in source, promotional/feed source, and acquisition cue share | A social-listening dashboard must separate opinion from automated links and location posts. |
| Period | Tweets | Positive | Negative | Acq. cues | URL share |
|---|---|---|---|---|---|
| Pre | 5,575 | 19% | 5% | 2% | 46% |
| Acquisition | 6,090 | 9% | 14% | 74% | 64% |
| Post | 8,022 | 17% | 6% | 14% | 49% |
The Event Spike Is Real, but It Is Not Pure Opinion
The event window has 6,090 tweets in two days. That is a major spike. It is also a measurement warning: 64% of acquisition-window tweets contain URLs, and 74% contain acquisition cues.
The acquisition first shows up as a chatter spike, not a clean sentiment series
The negative share rises from 5% pre-event to 14% during the acquisition window, then settles at 6% post-event. That is a real directional signal. But because the event window is also a news-link burst, the correct interpretation is "acquisition talk created a more negative event stream," not "all Goose Island drinkers became negative."
Read Volume, Tone, and Noise Separately
An event-study chart needs separate views. Volume tells us whether the conversation changed. Sentiment shares tell us how the tone changed. URL and acquisition-cue shares tell us whether the measurement target changed.
The event study needs separate lines for volume, tone, and measurement noise
The most important visual habit is not to over-read the peak. A news spike can make sentiment look worse because negative words appear in stories about loss of craft independence, not because everyday product experiences collapsed. The spike is still useful, but it answers a different question: what did people talk about when the acquisition became news?
The Period Explorer Shows What Changed
The acquisition period is dominated by anheuser, busch, bought, sold, and craft. The pre and post periods return more strongly to product and venue vocabulary: ipa, matilda, 312, stout, sofie, honkers, and clybourn. The source mix changes too. Check-in and beer-app posts are common before and after the event, but nearly disappear during the acquisition-news window.
A sentiment readout is only useful after separating event talk from product talk
2011-03-28 - twitterfeedUS: Anheuser-Busch InBev to buy Goose Island in US: Beverage news ...: The US arm of Anheuser-Busch InBev has lined up the purchase o...
2011-03-28 - Twitter for BlackBerry®How can Anheuser-Busch buys Goose Island but InBev owns Anheuser-Busch?!?!?
2011-03-28 - GoogleGoose Island Bought By Anheuser-Busch For $38.8 Million: Anheuser-Busch announced Monday that it was spending nearly $40 million to buy
This is why a social sentiment dashboard should show term evidence beside the score. A line labelled "negative sentiment" is not enough. The analyst needs to know whether the score came from product disappointment, anti-corporate acquisition language, a news headline, or noisy false positives such as "Grey Goose" and "Long Island."
What the Case Teaches
Three NLP lessons carry the case:
- Sentiment needs a measurement target. Product satisfaction, acquisition anxiety, and news amplification are different constructs.
- Source mix is part of the model. Foursquare, Untappd, feeds, links, and native tweets do not mean the same thing.
- Transparent baselines are useful. A simple lexicon exposes the core measurement problem before a more complex model hides it.