This project explains the creation and the use of tweetTV, a system that collects and analyses tweets from the micro blogging site Twitter, with the aim of comparing them with TV ratings in order to identify meaningful relationships.
Using tweetTV, Twitter text messages were collected, processed and then classified, by applying some data mining techniques. By isolating tweets that referred to three specific popular TV shows, a significant number of tweets was analysed and compared with TV ratings available from the British Audience Research Board (BARB). Furthermore, sentiment analysis was implemented to investigate the temporal variability of positive, negative and informational tweets in conjunction with TV ratings and whether the awareness of tweet’s attitude could enhance the accuracy of the system.
The running of tweetTV was successful in collecting sufficient amount of tweets to carry out the analyses required, which included qualitative and quantitative comparisons. A strong correlation was found between the number of tweets and the number of viewers, confirming the existence of a link between the two values that could be promising in accurately estimating TV ratings by only using Twitter. The small numbers of tweets collected for one show revealed a potential limitation of linking Twitter to some types of TV shows with relative explanations being discussed. The use of sentiment analysis also proved to be useful in identifying trends related to TV shows such as a periodicity of positive tweets during the days prior to the show. This suggested that sentiment analysis
could be used to improve the accuracy of tweetTV by weighting people’s opinions before the shows and projecting estimations.
Overall, the performance of tweetTV was successful in collecting and filtering tweets and in conjunction with sentiment analysis; it has the potential to work as a real-time application that will provide TV ratings.
Click here to read my full thesis: “Are TV ratings possible with Twitter? (2013)”