Words have power. Some just happen to be more powerful than others, and researchers from Cornell University have uncovered what makes them so.
In a study reported in the Proceedings of the Association for Computational Linguistics, researchers detailed how the use of automated text analysis was able to predict what kind of Twitter post is likeliest to attract attention. They then came up with a website that could be used to provide iterations of a tweet to make a post more popular.
By using this website, researchers tested out Twitter posts and found that the computer algorithm they made was able to make more accurate predictions as to which version of a tweet would likelier be retweeted more compared to what human observers picked.
Tweets are just 140 characters long but the computer algorithm is also able to analyze longer texts such as essays.
"We're looking at persuasion everywhere," said Lilian Lee, one of the authors of the study and a computer science professor from Cornell.
Twitter gave the researchers access, allowing them to carry out a controlled experiment that eliminated the effects brought about by a poster or a topic's popularity.
Different posts use different words for the same topic. Researchers collected and analyzed thousands of tweets and found that regardless of how popular a poster or topic is, the choice of words still mattered.
The list of features a tweet must have to make it likelier to be retweeted include:
--Asking people for a retweet (saying please is always nice)
--Being informative
--Using the language of the community and being consistent in this language
--Imitating newspaper-style headlines
--Copying words that commonly appear in other retweets
--Using words that clearly express sentiments, whether positive or negative
--Referring to other people and using pronouns in the third person
--Using generalizations (statements that apply to different situations are easiest to repeat)
--Making the tweet easy to read
Lee added that the researchers are looking into capturing how clever or amusing tweets can be calculated, but they haven't discovered a way just yet. They also challenged social scientists to try and figure out why these features work in making a tweet attractive enough to share.
The study received funding support from Google and the National Science Foundation. It will be presented at a symposium at the American Association for the Advancement of Science annual meeting on Feb. 15 in San Jose, Calif. Other authors include Bo Pang and Chenhao Tan.
