Individual actions and remarks may be easily predicted from others by quitting Twitter or Facebook

by Blogtrepreneur

The flow of information in modern society depends on social media, and many people send enormous information to Twitter, Facebook and so on. Also, even if you are active by anonymous or handle name, it is common to bring social connections such as family members, friends, company colleagues directly to social media. Meanwhile, research results have been reported that "it is possible to predict personal behavior and remarks by analyzing past records left in social media".

Information flow reveals prediction limits in online social activity | Nature Human Behavior
https://www.nature.com/articles/s41562-018-0510-5

Social media can predict what you'll say, even if you do not participate | Ars Technica
https://arstechnica.com/science/2019/01/social-media-can-predict-what-youll-say-even-if-you-dont-participate/

A research team at the University of Vermont wondered "what sort of limits are there to predict personal activities and benefits by using social media data" and said, "In a conventional way including machine learning, How accurate can you predict behavior? " We created a language model from more than 30 million tweets generated from approximately 14,000 Twitter users, and entered " entropy " which shows the average of probabilistic information amount and " par " representing the average number of branches of succeeding words Plexiti ", we evaluated the number of candidate words that follow and their predictability.

According to the research team, 927 active users were selected out of 14,000 users, 15 of whom were talking frequently online. The research team measured the predictability of words used by this 927 users and found that it is possible to predict remarks and behavior sufficiently for most users.

For example, the number of candidate words obtained from most of 927 users is about 45 to 256 words, and it seems that it was about 64 to 4096 words when only 15 people who were talking particularly actively on Twitter . Predictability seems to drop much if it expands to 4000 words, but prediction becomes possible with considerable precision when the lower limit of 64 words is reached. In conclusion, the research team said that 40 to 70% predictability could be derived from users' past tweets.

by Alan O'Rourke

Also, when talking online, there are many cases where words used mutually are common, so we say that word predictability will be higher than calculated numerical value. In addition, it has been found that people who regularly perform more than 8 tweets a day tend to predict more easily, and users with stronger social connections have also become more predictable. In conclusion, "If 8 to 9 social connections are made for one individual, sufficient predictability is obtained".

ArsTechnica of overseas media argues that this research has a clear influence on privacy. Even if I quit social media, if the remarks of the past remain or the social connection remains on the net, it is because someone who has nothing to do with it can analyze and predict that person's actions and remarks . The research team says, "As personal information is strongly carved in social media, even if you stop social media, it is possible in principle to identify and analyze individuals from social connections".