characterizing brand advertising strategies on twitter
play

Characterizing Brand Advertising Strategies on Twitter Shana - PowerPoint PPT Presentation

Characterizing Brand Advertising Strategies on Twitter Shana Dacres, Hamed Haddadi, Matthew Purver Queen Mary University of London http:/ /www.eecs.qmul.ac.uk/~hamed/ @realhamed July 2013, MSN-NGN Friday, 12 July 13 Advertising on social


  1. Characterizing Brand Advertising Strategies on Twitter Shana Dacres, Hamed Haddadi, Matthew Purver Queen Mary University of London http:/ /www.eecs.qmul.ac.uk/~hamed/ @realhamed July 2013, MSN-NGN Friday, 12 July 13

  2. Advertising on social media Social media have changed brand advertising scene. The volume of content and user interaction is unknown But, what do people actually say abotu a brand? Twitter has two advertising modes: Promoted tweets, and Promoted Trends Friday, 12 July 13

  3. Data collection Between December 17th 2012 to January 7th 2013 180k tweets from ~120k users directly engaged with the brands and hashtags. Spam removed (8k) .. BUT, TOPIC RELEVANCE? Friday, 12 July 13

  4. Friday, 12 July 13

  5. Topic classification Used a variety of Naive Bayes, J48 and other simple ML algorithms After, 80,000 tweets from ~50,000 different Twitter users We restricted the feature space to be based on the most common 100 words. Ten-fold cross-validation in order to simulate performance on unseen data. Best performance (78% overall accuracy) was obtained using NB only bag-of-words text features, with stop- words removed and a TF-IDF weighting, after manual filtering. Friday, 12 July 13

  6. Sentiment analysis What are the sentiments in tweets? Used the Chatterbox Sentiment Analysis API (http:/ /mashape.com/ chatterbox-co/sentiment- analysis- for- social- media Based on statistical machine learning over large, distantly labelled datasets (to avoid large manual labelling) Chatterbox report 83.4% accuracy in an independent study ( http:/ /content.chatterbox.co/Sentiment\~Analysis\%20Case\ %20Study\%20-\%20Chatterbox\ %20and\%20IDL.pdf ) Also tried SnetiStrength, however that didn’ t perform as well on slang and hashtags (Chatterbox is Twitter-optimized!) Friday, 12 July 13

  7. Sentiment analysis Example Friday, 12 July 13

  8. Advertising is effective Friday, 12 July 13

  9. Observations Promoted trends lead to higher engagement volumes than promoted tweets. Although promoted tweets obtain less engagement than promoted trends, their engagement forms are often more brand inclusive (more direct mentions) Although the volume of tweets is highest in promoted trends, they do not lead to the same level of positive sentiment that promoted tweets do Friday, 12 July 13

Recommend


More recommend