Twitter sentiment analysis on IKEA. Gun found and England fans trashing a store.

Micro-blogging is a combination of blogging and instant messaging that allows users to create short messages to be shared online. The trend evolved from blogging, to allow users an easier and faster way to communicate with people online and to be informed with relevant shareable information in real time. Micro-blogging offers various benefits over traditional blogging, some of which are its speediness of sharing information online, less time needed to develop content, more direct way to communicate with followers and mobile convenience.

Sentiment analysis through tweets

Twitter is an online social network that allows users to post short messages of up to 280 characters. This short restriction is to encourage succinct, focused and timely updates. There are more than 300 million tweets posted each day, making Twitter as a data goldmine for data scientists. Twitter and social networking tools like Facebook have shown the potential to make a societal impact, for examples Egypt’s 2009 political uprising was first instigated from social media.

Sentiment analysis is a growing part of Natural Language Processing, under the Business Intelligence & Analytics umbrella and is used in many domains. These domains can mainly E-commerce, government, science & technology, health and public safety. E-commerce is the most direct and general use of sentiment analysis and this is where we will be focusing our work on.

Different stakeholders have different use cases of sentiment analysis

Customers

Product and service reviews on websites reflect users’ experience and can be in the form of a rating, score or written opinions. Think of Amazon and TripAdvisor reviews. Sentiment analysis can be used to analyze this volume of information, understand the reason why some customers are dissatisfied and address these promptly to convert them into future paying customers. Customer opinions can also be used for new product inventions and help identify both functional and non-function requirements like features and costs.

Competitors

Sentiment analysis can be used to track what customers are feeling about the products and services of competitors in real-time. Accurate and timely information on competitors can be used in gaining a competitive advantage and to guide future product development.

Brand reputation management

Opinions from customers and social mentions can damage or enhance one’s reputation in the market. Brand reputation management is focused on the company and product rather than the customer. Brand perception is not determined on a 1-way street through a company’s public relations team but also is an aggregation of the conversations about a company. Sentiment analysis helps in determining how a company’s brand, products and services are being perceived by observing the sum of all conversations online.

What makes up a tweet

A tweet is generally regarded as a sentence for sentiment analysis, void of complex grammatical constructs. The language used in Tweets is generally more casual, colloquial and filled with local slangs like Singlish and misspellings. Also, there is a common use of hashtags by users for easier search identification. Historical tweets are easily available, through the exposed API provided by Twitter. The domain of topics discussed on Twitter is huge, compassing almost everything topics available. For the employment of a generic classifier to track sentiment, Twitter is a good general social media platform to start with.

Tracking ‘Ikea’ twitter sentiment

Through APIs, we extract 300,000 tweets between 2018-05-13 to 2018-07-09 using ‘Ikea’ as the search query. We employ the Valence Aware Dictionary and sEntiment Reasoner (VADER) tool for sentiment analysis. It is specially designed to be attuned to sentiments expressed in social media. VADER’s main feature is to analyse polarity of a text (be it positive or negative sentiment) and also valence intensity (how strong different forms of words are).

Various scenarios for sentiment analysis are accounted for:

  1. Common negations (e.g., “not good”)
  2. Use of contractions as negations (e.g., “wasn’t very good”)
  3. Conventional use of punctuation to signal increased sentiment intensity (e.g., “Good!!!”)
  4. Conventional use of word-shape to signal emphasis (e.g., using ALL CAPS for words/phrases)
  5. Using degree modifiers to alter sentiment intensity (e.g., intensity boosters such as “very” and intensity dampeners such as “kind of”)
  6. Understanding many sentiment-laden slang words (e.g., ‘sux’)
  7. Understanding many sentiment-laden slang words as modifiers such as ‘uber’ or ‘friggin’ or ‘kinda’
  8. Understanding many sentiment-laden emoticons such as 🙂 and 😀
  9. Translating utf-8 encoded emojis such as 💘 and 💋 and 😁
  10. Understanding sentiment-laden initialisms and acronyms (for example: ‘lol’)

Sentiment analysis

We group the tweets into hourly bins and used 20 periods moving average to produce a smoothed series. We eliminate tweets with a polarity index of 0 as these could represent noise for certain tweets and doesn’t value add when we are more interested in the cases of positive or negative sentiment.

We first plot a histogram of all the sentiment scores with ‘IKEA’ as the twitter search query.

There is a negative and positive hump as the 0 values are eliminated. As seen from the histogram, there is a positive bias towards tweet sentiment. Good news for Ikea’s public relations team!

We observe the most recent representative sampling of the tweets for a sanity check.

Date, Text, Sentiment
2018-07-09 18:26 Hurting the country-brand... “England fans invaded an IKEA after beating Sweden in the World Cup” https://twitter.com/i/moments/1015856849764089856 … -0.6908
2018-07-09 18:26 IKEAだー!!✧(✪д✪)✧ 長久手のところですか? 私も去年行きましたがホットドッグとアイスめちゃ安いですよねꉂꉂ(ᵔᗜᵔ*) そして普通に美味しい(๑'ڡ'๑)♡ 私も最近忙しくてイベ消化がカツカツですがまったり頑張りましょ(´∀`*) 0
2018-07-09 18:26 IKEAレベルの棚上げ、なんちゃってー(*´ω`*)笑 0
2018-07-09 18:26 Strangely as an irishman there's a very small part of me rooting for the current @ England squad in this @ FIFAWorldCup - a bunch of young lads after all...but then their fans go and trash an @ IKEAUK. They make it so hard to root for them as an outsider # belgium2018 # bbcworldcup -0.4707
2018-07-09 18:26 i'm goingikea tomorrow 0 
2018-07-09 18:26 Neither of the boring heteros in the song actually saw the film, but they bought the big ugly Audrey Hepburn print from Ikea -0.6084

Note that tweets in Japanese language were assigned a 0 score as the VADER package is set to work on the English language. The negative tweets had negative scores assigned correctly. The caveat is that at times, there can be imperfections with VADER, like in high complexity cases of sarcasm. Even the human can classify sarcasm wrongly. From an overall aggregated basis, the sentiment analysis tool should do fine.

Interpretation

2 key patterns can be observed in the sentiment analysis.

The first is the dip to negative polarity between 25 June to 1 Jul. We dig out actual tweets from our database to see what happened.

Representative tweets:

Another child fired an unsecured gun. This time the gun came from a sofa at an IKEA
Man charged with criminal recklessness after shooting incident at Fishers IKEA
# BREAKING: The man who lost his gun inside the Fishers # IKEA (that was later picked up and fired by a child) has been...

Shockingly, there was a gun hidden in a sofa at an IKEA store! This is definitely bad for IKEA’s brand management. Public relations team should act quickly to resolve the public’s confidence that (1) this is an isolated case and (2) IKEA has taken steps to ensure that this will not happen in the future. Competitors of IKEA can then use this opportunity to catch the walk-away customers.

The second key pattern is the spike in tweet volumes from 7 July onwards and the corresponding dip in tweet sentiment. This rise in volume could be due to Zlatan Ibrahimović high profile tweet to David Beckham on 7 Jul, ahead of the England v Sweden Semi-finals clash raising initial awareness of Ikea.

Fortunately, Sweden won the game. Unfortunately, England’s hooligan fans started trashing on an Ikea store after.

Representative tweets:

England fans trashed an IKEA after beating Sweden in the World Cup
England fans causing criminal damage to an Ikea (and a bloody ambulance?!)
One reason why I hate England fans. Obviously this is just a minority but some people think just because they're celebrating they can do whatever they want. There was even a group of guys who went to Ikea and trashed a bunch of stuff.

This incident should be less brand damaging than the previous gun found in an IKEA incident, indirectly reflected by the less negative polarity value. The trashing was caused by the England fans rather than something that happened because of lax IKEA internal controls. What IKEA could possibly do is to be good sports, (1) create feel-good witty ads that take advantage of this increased publicity. Likewise, competitors can create similar witty campaigns.

Update:

Of which, IKEA actually played this out nicely with their tweets put out on Jul 8.

We are aware of a small group of fans celebrating the match result in one of our stores. Being both British and Swedish, we were on the edge of our seats during the game and we would like to say ‘grattis!’
(that’s congratulations in Swedish) to the England team for getting through to the semi-finals. In celebration, we’ve put one of the nation’s favourite dishes, fish and chips, on offer for just £1.


Conclusion

This is a working use case on using Twitter sentiment to track catalytic events and how companies can capitalize on the opportunity or mitigate the damage.

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *

Latent Analytics

35 Selegie Road #10-17, Singapore 188307
(65) 6610 3609 | hello@thelatentanalytics.com