If you analyze Twitter's tweets by machine learning, you can identify whether it is "drunk"

BySprogz

When tweeting with drinking alcohol and having a good feeling, there is a possibility that it may be better not to talk about it better than anyone who is drunk and may send out strange e-mails to someone drunk. It is enormous. An algorithm that automatically finds such "drunk tweets" was developed by American scientists.

Inferring Fine-grained Details on User Activities and Home Location from Social Media: Detecting Drinking-While-Tweeting Patterns in Communities
(PDF file)http://arxiv.org/pdf/1603.03181v1.pdf

Machine learning algorithm can identify drunken tweeting | Ars Technica UK
http://arstechnica.co.uk/science/2016/03/drunk-tweeting-computer-algorithm/

Research teams such as Mr. Nabir Hossein of the University of Rochester School of Computer Science started up the idea of ​​combining Twitter and machine learning to track alcohol consumption in certain areas based on information obtained from Twitter . The research team gathers tweets with position information in New York State between July 2014 and July 2015 and collects tweets including keywords related to alcohol such as "beer barrel" "drunk" After sifting, we gathered tweets about 11,000 sake.

Crowdsourcing tool to provide more human labor "Amazon Mechanical Turk"By extracting 11 thousand tweets from 3 questions, I extracted the tweets posted in a drunk state. The three questions are as follows.

Q1: Does this tweet mention something about alcoholic beverages?
Q2: If so, is the person who tweeted it alcohol drinking oneself?
Q3: If so, is it likely that you were drinking alcohol at the time the tweet was posted?

Based on these results, Mr. HosseinSupport vector machineWe created an algorithm to classify tweets by a method called. As a result of the learning, the algorithm was able to perform 92% agreement behavior with the result of Amazon Mechanical Turk, and a tool to distinguish "drunk tweet" as a person judges by machine learning was completed.


After that, the research team shifted to the step of identifying where drunk Twitter users drunk, such as homes and different places, where they drunk and tweet as a next step. In addition to the location information data, we incorporated thousands of tweets with algorithms incorporating whereabouts such as "baths", "sofas", "televisions" that are likely to be used at home, for example.

I re-checked the algorithm classification result again by Amazon Mechanical Turk and incorporated other judgment factors such as the position information tweeted at the end of the day and strengthened the accuracy of the algorithm. The completed algorithm is now able to specify whether the user's location is at home or not with an accuracy of 80%.

Analysis of the algorithm thus completed shows that there are different trends in where alcohol is consumed depending on the place of residence, so people living in New York City, unlike those in the suburbs, either at home or at home Tend to drink in a place close to. This is believed to be due to the presence of clubs and bars in the city every block.

ByLeo Hidalgo

The research team is planning to conduct comprehensive research on alcohol consumption in social media in the future, for example, when it becomes apparent that "how exchange and Twitter relations and human relations affect drinking amount" It is.