The researchers from the University of Rochester have developed an algorithm that helps in detecting the tweets done by a person while drunk. This experiment will help in examining the drinking patterns among the people and will also be helpful in studying various other human behaviors.
The researchers began with selecting the tweets from New York City and rural New York, and sorted them based on various keywords such as “drunk”, “get wasted”, “vodka”, “getting high” etc. These keywords were then used in building an algorithm so that these keywords can be identified by it.
Further, they filtered the tweets selected and sorted by the algorithm through various stringent questions to select the specific tweets, which not only referenced about drinking but also indicated that whether the person was drinking while sending the tweets. This way the algorithm was able to determine whether the person was actually drinking while sending the tweets or was just sending the tweets containing such keywords. Upon the successful completion of a reliable database, the algorithm was fine-tuned so that it could recognize the words and locations that more likely to prove that the people were actually drinking while sending the tweet.
In order to precisely locate the position of the tweeter, the only geo-tagged tweets were used. Further, the users’ home location was determined by checking their location while they sent the tweets during the evening or the tweets containing the keywords like “bed’ or “home”. This addition helped the researchers to determine whether the users preferred to drink at the bars or at the restaurants.
The researchers then combined the both datasets to get an extensive idea of the number of peoples drinking in an area at a particular time. As result of this, they found a correlation among the number of bars and the number of people drinking.
According to the researchers this is just the beginning. Their machine-learning algorithm can track a wide range of behaviors using the actions that people record on twitter such as exercising, playing, shopping and eating etc. Theoretically, anything containing a hash tag on twitter can be tracked.
But there are certain drawbacks associated with this algorithm. Since twitter is more actively used by certain group of people as compared to the others, the data used by the algorithm for retrieving any kind of behavioral information will represent those specific groups. Nevertheless, this algorithm can be very promising in gathering the information about various habits and behaviors in the society.
Chanwoo's Weekly Tech 