Companies do anything we say, and have made billions of dollars by avoiding watching online in the consumer profile experiment. Recently, it is enough for some users to reduce enough use of social media or delete their accounts altogether. But according to a new study, it is no guarantee of privacy. If you can connect with other users, their activity can also expose you. Now, computer scientists have shown that your 10 nearby Twitter streams can better predict your future tweets than your own stream.
David García, a scientist at the Vienna Medical University in Austria, says that "it seems a little different than it looks," to understand the person's character with such other hand supervision, which was not involved in the study.
Rather than predicting anyone's actual tweets, researchers at the University of Vermont at Burlington speculated how the future words of a person would be used by using a known measurement as an autopsy. More antony means more randomness and less repetition. They saw 927 users' Twitter streams, each containing 50 to 500 followers, and 15 users tweeted each one of them. In each person's stream, he calculated how many antrophrosis are in sequence of words. (On average, tweets had more antropos than James Joyce, less than Ernest Hemingway.) He then plugged that number with information theory called Fenso's inequality, so that a person's stream could count the first word in its next tweet. . . At the correctness, it was 53% high bound average. But the prediction of each subsequent word is a little less accurate.
Next, he calculated the upper bound for prediction based on the user's stream, as well as computed the flow of close to 15 contacts. Accuracy increased to 60%. When they removed the user's stream from the equation, that figure went up to about 57%. This means that watching the streams of users' contacts is as good as inserting a user – and better than the single-user surveillance, researchers have reported this today. Nature Human Behavior. It takes only 10 contacts' streams to cross the predicted accuracy of a person's own Twitter stream. For comparison, predicting the accuracy of 51% of someone who writes on the basis of random classification of strangers, gives the maximum accuracy. (About 53% of the person using his own tweets is hypothetical because the English language has many regularities and people tweet about it.)
"We are asking to use some very interesting mathematics from information theory: how well do you do if you have a full machine learning method?" Chief writer James Bagro says, Data Scientist of the University of Vermont in Burlington. Joan Hinds, a University of the University of the United Kingdom, agrees. The new approach is that "this is a unique method that goes beyond most of the current work in this field."
The results show that, in theory, someone has guessed that a person who is not on Twitter will tweet, crores say. Actually, that means someone's friends were offline and then find friends' feeds on Twitter. But many apps ask for access to contact lists and some are known to share them. Facebook, for example, announces the contact list of users to create "Shadow Profiles" of people who are not on the network. Researchers have already used people's own tweets to predict personality, depression and political orientation. Hypothetical tweets based on tweets of friends can allow equal integration.
One of the practical limitations of this work is that it takes all those words as informative, but some people can tell you more about it, says Bégro. If your friends, gay gestures, or just follow Republican politicians, tweet about a lot, which can specifically inform you of your sexual orientation or political orientation. García has found that contacts on Friendster can predict the status and status of one's sexuality, and Twitter's contacts can predict somebody's location. Hinds say, "We rarely scrub on the surface of what kind of information can be disclosed in this way."
Bagro says, "What concerns me with regards to privacy is that there are many bigger roads that I think people feel that people can not understand." Another thing that people do not take into account, says: "When they leave their data, they are also giving data to their friends. What we think is personal choice in the social network is not really that."