Probably we have one of the best marketing technologies in the future.
Quite a scary experiment conducted by scientists from the University of Michigan (USA) Steven R. Wilson and Rada Mihalcea – they were able to teach the neural network to predict the actions of social network users in real life (not online!) on the basis of information about their past actions.
Despite the fact that the experiment was difficult and time-consuming, the researchers were able to ensure that the results of neural network predictions come true more often than simply guessing.
Using the public Twitter API, scientists have collected data about users who wrote in their accounts things like “eat pizza”, “watched a documentary”, “a walk with the dog”. It assembled a huge set of data based on the posts 200 thousand users. It was able to identify that they make about 30,000 daily activities. They were split into “activity clusters” in groups and began to train the network to distinguish between a high-level group. For example, “caring for Pets and playing with them” or “consumption culture through concerts, exhibitions, cinema” etc.
Additionally, a survey was conducted among 1000 U.S. Twitter users about their activities in real life for one week in order to clarify the classification of activities.
Only the phrases with an exact match to structure type, “I went to the gym.” were taken into account. All messages that could be interpreted in different ways, were cut off. Each user profile has managed to gather information about 3 200 posts – so much to explore Twitter allows third-party companies and scientists.
Then the neural network for user profiles was to predict their further actions in the near future. And that some problems were implemented. Not all prediction models that were built, scientists were able to overcome the indicators of “random guessing”. But in some groups, activity classes of the neural network was the best predictor, then “at random”.
What does this mean for marketers and users of social networks
The work of the Michigan researchers foreshadows the technology “targeting the predicted activity.” If the development of “predictive neural networks” will continue, soon we will be able to see in social networks a function of targeting audiences like “people who next week will go to the dentist” or “people who gathered at a football match”.
But users of such social networks may not like, if the knowledge that information about their online activity used by advertisers, Internet users started to get used to tracking their movements in the offline and the bombardment is in accordance with future actions and planscan scare them.
It all seems fantastic at first glance, however, ongoing research makes reality not so. Maybe today it is worth considering – and whether to report to Twitter, Instagram or Facebook that yesterday you “went on night and day hobo-bikers” or “let the orange clouds on the window sill”?
It is curious that described the study was funded by including money DARPA – the Management of perspective research projects of the Ministry of defence. As you can see, the military is also experiencing some interest in machine prediction of the actions of people in real life based on data from social networks.