Shulou(Shulou.com)06/02 Report--

In today's ubiquitous era of intelligent recommendation, it is believed that each of us is suffering from privacy anxiety more or less. Usually when talking to friends, they feel paranoid that their phones are eavesdropping, so many people less and less like posting information on social media.

Don't say, don't turn, don't like, it seems like a very reasonable "privacy company"-if I don't leave any data traces, machine learning naturally won't be able to analyze my preferences.

But is this really the case?

To accurately predict your social media behavior, only 9 friends are needed.

Recently, in the journal Nature, data scientists from the University of Vermont published a study that does not need to look at individual user data on Twitter, but by analyzing the tweets of individual users who follow each other's friends. Accurate prediction of individual users' social media behavior can be achieved.

In other words, even if you delete the number and withdraw from the Internet, machine learning can still analyze your user profile through your friend list.

In the study, the researchers collected more than 13000 Twitter accounts, each with 150,200 friends. The top 9 friends with the highest interaction rate are found and grouped, and the behavior data of friends in the user's friend group are analyzed with time as a clue.

Data scientists have found that the average social media user usually uses no more than 5000 words on social media as long as it is not a KOL that focuses on a particular field. If you use a Chinese analogy, it is probably the usual words such as "ha", "not Chinese" and "my brother is so handsome".

Through the analysis of the common words of the user's friends, we can basically delineate the scope of interest of the user. Many users who focus on a particular field, such as politics, actually use only a few hundred words on social media, the researchers said. The entropy rate which represents the change of personal behavior is introduced into the vocabulary within the range of interest, and then combined with the text generation algorithm based on time series, we can predict personal social media behavior through social relationships.

The more friends there are in the friend group, the more accurate the prediction is. However, it is worth noting that once the number of friends exceeds 150, the accuracy of prediction decreases-because when there are too many friends, the relationship between the two parties tends to decrease, breaking away from the original circle of interest, and it is difficult to form a mapping relationship in behavior.

That is to say, if all your Weibo friends are Wolves Warriors, then you will fight for idol with your sisters late at night if your Weibo is a star-chasing position. A person's circle of netizens determines the scope of his cognition, which in turn makes his behavior predictable.

It seems that the essence of human beings is that the repeater is correct.

High walls are erected in the circle, and privacy collapses inside.

Although this study shows the progress in natural language processing, it also confirms two problems that we have been worried about for a long time.

First of all, to accurately speculate the words and deeds of users through the words and deeds of Twitter friends is undoubtedly a valid vote for the information cocoon house theory.

Since Sunstein's information cocoon house hypothesis was spread, the degree of approval of this hypothesis has been in a state of instability for a long time. Especially in recent years, with the popularity of social media and personalized information flow, many people often say the word "information cocoon room". Supporters believe that the algorithm recommendation will continue to recommend information in line with the interests of users to users, resulting in cognitive blockage of users, lack of a complete understanding of the world.

Its opponents argue that people are born to show an interest in the information they receive, and that even in the age of newspapers and books, people choose what they are interested in to read. This choice will not form the so-called "hindrance of information dissemination" and "cognitive alienation from the outside world", and the so-called cocoon house hypothesis can not be established.

But the University of Vermont made the point in this study that people to information are often not only interesting choices, but also social choices. The strong correlation between the interest, language style, active time of friends on social media and the social media behavior of individual users means that a person's information cocoon room is built through social circles, and this cocoon room has a direct effect on the expression ability of users, such as an interesting phenomenon in the study, the more interested users are in political topics, the less vocabulary they have to speak on social media. To some extent, this proves that the barrier of communication circle under the information cocoon room is becoming more and more obvious, and people who have been immersed in one circle for a long time have lost not only the opportunity to receive information from other circles, but also the ability to receive information from other circles.

What is even more frightening is that we have finally lost the last piece of personal privacy by finding out what users like about this technology through interrelated friends.

Austrian jurist Schernberg once published a book called "deletion: forgetting is a Virtue in the Digital Age", which is about the right of personal information to be forgotten on the Internet and the information about individual users. Individual users should have the right to require websites and platforms to be completely deleted from the server or restrict access.

At first, people's demands for the right to be forgotten only focused on personal negative information such as compulsory auctions and lists of broken promises, but soon the demand began to become more and more widespread. For example, around 2013, there was a popular tool to find information such as Weibo and Douban that had been deleted by users according to the links to their home pages. at that time, it was used by many people to conduct human flesh search. so that once again aroused a strong demand for the right to be forgotten.

Today, the right to be forgotten has been written into the GDPR of the European Union, canceling accounts and deleting everything has become the last line of defense for us to defend online privacy. But the emergence of the "friend algorithm" means that as long as the other person can find your friend relationship, even if you delete everything, you can still show yourself naked in front of the algorithm.

It also allows companies like Facebook, which makes money by probing users' information, to take advantage of ethical loopholes-if a user insists on not granting data rights to social media, but his social media friends are open to social media, social media platforms are likely to "reasonably and legally" bypass users themselves to obtain user information.

A copy of the self

The first physicist in the world to discover natural radioactivity came from Becquerel, France. When he was killed by radioactive material, he probably never thought of this technology that he had devoted his life to. The future will evolve into a nuclear bomb that has claimed countless lives.

This is always the case. The person who unwittingly opens Pandora's box often fails to see the appearance of the world changed by the box.

The same is true of this social media prediction algorithm, and under the same logic, we can almost imagine: is it possible to know how often her husband shaves through the analysis of a housewife's shopping data? Is it possible to know a person's ability to work and salary by simply analyzing it on workplace social software?

Under the joint action of digital survival and AI, the ownership of the "self" has been copied and infinitely divided, scattered in all the places that are related to us-the preferences of our friends, the behavior of our families, the actions of our loved ones. So that from these connections, we can also dig up fragments and restore our selves.

In the face of this possibility, it is difficult for us to decide whether it is good or bad right now. But this time, it may be a good choice to let laws and regulations stay ahead of technology.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.