Conclusion 1 See also: danah boyd, Kate Crawford, “Six Provocations for Big Data”, September 2011: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1926431http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1926431
“relatively high amount of hype” “even when the predictions were better than chance, they were not competent compared to the trivial method of predicting through incumbency.” “We simply tried to repeat the (reportedly successful) methods that others have used in the past, and we found that the results were not repeatable.” “Hoping that the errors in sentiment analysis ‘somehow’ cancel themselves out is not defensible.” “Spammers and propagandists write programs that create lots of fake accounts and use them to tweet intensively, amplifying their message.” “Predicting elections with accuracy should not be supported without some clear understanding of why it works”. “Learn from the professional pollsters … identify likely voters and get an unbiased representative sample of them”
Examines relationship between emotional reactions and public opinion Seeks to offer insight into how public opinion is formed Based on analysis of posts from Usenet online forum Evaluation of emotional content is based on counting of words in ANEW – Affective Norm for English Words Nevertheless, this still begs the question of sample bias How typical are Usenet users of the general population?
SPARQL query language use cases: “Give me a stream of locations where my product is being mentioned right now.” “Give me all people that have said negative things about my product.” “Give me all URLs that people recommend with relation to my product.” “What competitors are being mentioned with my product.” 511,147 tweets about iPad (June 3 rd – June 8 th 2010): http://wiki.knoesis.org/index.php/Twarql
Use of agent-based prediction market Each agent extracts users sentiments from a different social medium Reflects it beliefs by trading in the marked Belief-Desire-Intentions paradigm Agent will intend to do what it believes and will achieve its goals given its beliefs about the world Avoids problems with human agents Poor estimation at either end of probability spectrum Agents do not manipulate the market Do not require recruitment and incentives Bothos et al, IEEE Intelligent Systems, November/December 2010
Presents methodology for predicting individual retweets in Twitter Input to the model is the tweeter, a retweeter and the content of the tweet Output of model is the probability of a retweet of a tweet by the retweeter Probabilistic collaborative filtering prediction models used, called Matchbox Crawled twitter from June 10 th 2010 to July 29 th 2010, finding 20,000,000 retweets
Understanding of specific groups useful for commercial and political organisations Four key tasks: Discover or extract the group itself Develop a profile from group descriptors and defining group characteristics Understand group’s sentiment and ability to influence other individuals or groups Study group composition Privacy and security are important concerns
END OF SLIDES One reason for low concordance is use of U for you or rly for really. Also, frequent typos, and use of Internet acronyms such as rly for really. Sentence fragments, and pronoun drops such as busy now instead of I’m busy now.