TechTheDay

Bad news for those who’ve always wanted to hide their home location on social media – Researchers at IBM have developed an algorithm capable of determining a Twitter user’s home location. Yes, Twitter can reveal your address without you clearly posting it. The process is not even a long and meticulous one. With only the last 200 tweets posted, it is possible to locate your address even as you try your best to avoid posting any information related to your address.

Twitter has a feature that allows tweets to be tagged with a user’s location information. Obviously, this is intended to broadcast a user’s whereabouts. It is very helpful if you really want people to know where you are.  Fortunately, there’s an option to turn it off or to prevent your location from getting associated with your tweets.

Preventing location information tagging, however, is not a guarantee that your location will no longer be identified. Jalal Mahmud, a researcher from IBM Research in Almaden California, along with other IBM researchers, announced that they have come up with an algorithm that can identify home city location. The algorithm is referred to as Naive Bayes Multinomial.

The Method

For the algorithm to work, there should be at least 200 tweets posted by a user. These tweets should be public. The system will reject private postings.

Mahmud and his team selected 100 Twitter users from 100 different different locations to evaluate the effectiveness of the algorithm they developed. They extracted 200 tweets from the selected users. The compiled tweets were then divided into two groups. The first group, composed of 90% of the obtained sample tweets, was used to train the algorithm. The remaining 10% was used to test the algorithm.

The Results

After the testing, accuracy has been estimated to be near the 70% level. Well, this is not a high number and this would probably be a good thing for many. The algorithm is not capable of exactly pointing out a user’s home address in many cases. Still, the algorithm can be considered functional and good enough considering that it derives the location information not from exact details posted in or associated with the tweets being analyzed.

How It Works

The algorithm works (at least to some extent) because there are still many ways for obtaining location information aside from geotagging. For instance, in the study conducted by the IBM researchers, they found that more than 100,000 of the tweets were posted by Foursquare, a location-based social networking site. These tweets even had links to exact locations. Also, nearly 300,000 of the tweets they examined included the name of the cities listed in the US Geological Service gazetteer. Tweets may also include details that can offer hints on a possible location. Expressions of support or cheers to particular sports teams, for example, can give a clue on where a Twitter user is based.

Demonstration

The research team used Naive Bayes Multinomial on actual tweets. It was necessary to examine the Twitter users for the possibility of them traveling or regularly moving from one location to another. Obviously, tweets about travel are very unlikely to provide reliable home location information.

The results after running the algorithm are promising. As tested, the algorithm has a 68% accuracy for predicting home cities, 70% accuracy for predicting home states, and 80% accuracy in identifying the right time zones.

Practical Uses of the Algorithm

Apparently, the algorithm was not developed to be used to track an individual. With the level of accuracy it currently has, it’s mostly useful only in narrowing down location details. It can be more useful, however, when it comes to tracking prospective customers or target markets. For marketers, it is a great tool that can facilitate marketing campaigns. It can also be used to assess current customers, especially in identifying product popularity and customer satisfaction. It is likewise very useful for politicians or those who are seeking elective posts in the government.

Taking everything into consideration, it is safe to say that the algorithm is unlikely going to be a privacy concern. If you observe a good amount of privacy mindfulness, details about your location are not going to be divulged through Twitter or social media in general. The possible issue here is far from the rising risks of smartphone hacking. Discretion is still going to be the most potent threat against your own privacy.