We tend to think of social media as just another way to communicate, whether we’re exchanging stories with friends, giving our opinions on the latest issues, or watching viral videos. But what we consider to be communication, researchers see as data to be mined, which is used for predicting everything from stock prices to the future behavior of an individual.
If you trade or hold stocks, you’re directly affected by price movements in the stock market, but even people who don’t have a direct connection to the market may be touched by significant changes in it. The market’s movement affects the availability of jobs, credit for homes and businesses, and even the financial health of entire countries.
When researchers began analyzing Twitter to predict stock market changes in the mid-2000s, professional money managers dismissed these statistical wizards as nut jobs. But in 2010, researchers from Indiana University determined that the general mood on Twitter can be used to predict the direction of the stock market up to six days in advance.
After analyzing about 10 million tweets from 2008, these researchers found that calmness in the tweets, as measured by the Google Profile of Mood States algorithm, correlated with movements in the Dow Jones Industrial Average (DJIA). The calmness index predicted daily changes in the closing value of the DJIA between two and six days later with 88-percent accuracy.
That’s because markets react to emotions, especially fear and greed, which is reflected by Twitter in real time. Of course, most traders don’t rely solely on Twitter data, but it may be used as a contrarian indicator of sentiment. That means if the public is extremely optimistic, a contrarian stops buying stocks or maybe even sells them. Conversely, if the public is exceptionally pessimistic, a contrarian looks for an entry point to buy.
The predictive power of Twitter sparked massive interest from Wall Street money managers. Firms like MarketPsych began providing carefully filtered Twitter data to trading research firms and hedge funds. In 2012, MarketPsych’s managing director, Richard Peterson, claimed that his models would have returned 30 percent per year, and data from trading simulations backed him up.
Researchers have even identified a correlation between certain Google searches and impending market crashes. Using Google Trends, they found that spikes in searches for business and political keywords from 2004–2012 preceded declines in the S&P 500. However, the predictive power of these Google searches has waned as more money managers and professional traders have started using the data, so the researchers are working on more advanced Internet-related strategies to restore their trading edge.
In 2013, researchers showed that Twitter discussions were incredibly effective at predicting the winners of the 2010 US House of Representatives elections. Using a database of 543,000 tweets that mentioned a 2010 candidate, the researchers’ models chose the winner in about 93 percent of the races.
The analysis only measured the tweet share of each candidate—there was no attempt to determine the users’ political beliefs or even what they were saying about the candidate. This finding suggests that all publicity, positive or negative, can help a candidate. In one example, Republican Mike Turner received 65 percent of the tweet share in Ohio’s 3rd Congressional District and went on to win with 68 percent of the vote. The difference between his share of tweets and votes is within the margin of error of traditional polls, over which social media analysis offers many advantages.
But the methods for using social media to predict elections are still evolving. Their deficiencies became apparent on Super Tuesday, March 6, 2012. Out of 10 states, various social media predictions picked the wrong winner of the Republican primaries 40–60 percent of the time.
There are a number of explanations for the fiasco. It may have been because tweeting about a political candidate is easier than voting, many social media users may not be registered or otherwise eligible to vote, or the social media population in 2012 wasn’t a perfect sampling of the general population. (GOP primary voters are whiter and older than most Americans, while Twitter users are more likely to be young minorities.) As social media software and methods grow more sophisticated over time, though, detailed predictions in US presidential primaries may become more accurate.
Social media data is used by retailers to predict buying behavior. For example, Walmart has used Twitter data to uncover product trends in certain locations so that the store can stock inventory that’s more likely to be purchased by customers in that area. It can even be used to predict future purchases: For example, Facebook and Twitter users often post when they become engaged or pregnant. Marketers know that these individuals will soon be purchasing things like wedding invitations or cribs and advertise to them accordingly.
Although this may seem like creepy eavesdropping, it turns out that many of us can be bribed to give out our personal information. You can try disconnecting from social media and the Internet to protect your privacy, but you may still be tracked through credit card purchases, customer loyalty cards, and other sources of data that are sold to retailers. Even if you only pay with cash, technology is being developed that will make it almost impossible to escape inclusion in retail databases.
First, there’s facial detection technology, which would be able to analyze your facial features to determine your gender, age, race, clothing size, and possibly even your emotional state as you enter a store. There’s also the possibility of facial recognition technology that would identify you by name and matches you to your individual profile in the store’s database. A retailer could then use that information, including that collected from social media, to push offers to your mobile phone and wearable devices like Google Glass as you shop.
Using social media to predict fashion trends is a blend of art and science. According to analysts at IBM, the “steampunk” look (a mixture of Victorian and industrial styles) would go mainstream by 2014. They predicted this in 2013 by studying over 500,000 public posts on social media sites, forums, blogs, and news sites. Sure enough, steampunk is featured as a fall 2014 trend in college clothes and accessories on The College Juice.
However, not all trends can be predicted the way IBM predicted steampunk. There’s really no way to measure the buzz of something that doesn’t exist yet, so designers still take risks with brand-new styles. One costly mistake may put a designer or brand out of business, so designers and retailers are using social media in a different way to involve consumers in the design process. New designs have debuted on social sites like Instagram, Twitter, Pinterest, and more.
By capturing immediate audience feedback, designers and retailers can determine what will hit, in what quantities, and where. Pieces that don’t work can be updated until they do, or they can be discarded entirely if they prove unpopular. Customers can even pre-order in some cases. At runway showings, buyers and magazine editors are being eclipsed in importance by social media celebrities. Pure analytics then reenters the picture to predict what styles, sizes, and colors to manufacture and ship to stores to meet consumer demand.
6Your Job Performance
It’s become commonplace for employers to perform Google searches and social media background checks on prospective employees. Employers may reject you for appearing drunk in photos, using racial slurs, gossiping, or even making grammatical mistakes online. Some employers may even reject you for what other people say about you.
But there’s a new way that employers may be using social media to make inferences about your personality and job performance. Employers aren’t supposed to make decisions based on race, gender, religion, or sexual preference, but even if you don’t tell them, your Facebook “likes” may reveal this information with surprising accuracy. It can even predict your intelligence, emotional stability and whether you abuse alcohol or drugs. Apparently, liking “Clark Griswold” makes you seem stupid, but liking “beerpong” makes you an extrovert. Liking “Kurt Donald Cobain” shows that you’re neurotic, but liking “serial killer” indicates spontaneity. To find out what your Facebook likes reveal about you, you can use tools like YouAreWhatYouLike.com for a personality profile.
University researchers have also devised a test that uses information from your Facebook profile to predict your job performance. They measure the “big five” personality traits: agreeableness, conscientiousness, emotional stability, extraversion, and openness. To get a high score on agreeableness, avoid public conflicts or angry debates with people on Facebook. For conscientiousness, you need a detailed profile, many detailed posts, and photos that show you working hard. To score well on emotional stability, avoid showing too much emotion in your posts and pictures. Lots of friends and photos of you in crowded social gatherings show extraversion, while pictures of international travel and posts about classic literature display openness. According to the research, if you score high in every category, you’re probably going to be an ideal employee.
Twitter can also be used to assess your personality with only three pieces of public information: how many people you follow, how many people follow you, and how many times you’re listed in other people’s reading lists.
5What You Need To Know Before A Meeting
Social media can simultaneously predict what you need to know before a meeting and your talking points. Mobile phone apps like Humin and LinkedIn’s Pulse and Connected automatically retrieve the latest business news as well as current information about the people you’ll be meeting. Whether you’re going to a job interview, a sales meeting, or a financial negotiation, these apps automatically push information to your mobile phone to help you break the ice, deepen a personal connection with a work associate, or give you the latest facts about a company or its products.
Connected and Humin also sync with your phone’s calendar to alert you to birthdays, personal and work anniversaries, business interests, job changes, news mentions, and other profile updates for the contacts you’ll be meeting. These apps will tell you how you’re connected to someone, including where you first met. Humin will even tell you which city the person is in and if they’re available to meet.
Apps like Pulse are professional news aggregators that let you customize the content you want to receive from major news organizations, blogs, and more. That way, you’ll always be up to date on the latest company and industry news.
The government can track your movements and predict where you’ll go in the future by mining social media data with software developed by Raytheon, a defense contractor, called Rapid Information Overlay Technology (RIOT). With just a few clicks of the mouse, RIOT can create a map of where a person has been as well as the locations of their friends and associates.
The software tracks a person across Facebook and Twitter and uses time and location information from Foursquare, a mobile phone app that tells your friends where you are. If available, RIOT also extracts latitude and longitude information from photos posted online. RIOT not only knows what your pictures look like, it knows where they were taken.
This type of social media data mining raises serious privacy concerns, especially because innocent people may be swept up in a government investigation for sharing harmless pictures of their lives. According to Ginger McCall of the Electronic Privacy Information Centre, “Users may be posting information [on social networking sites] that they believe will be viewed only by their friends, but instead, it is being viewed by government officials or pulled in by data collection services like the RIOT search.”
Researchers at Brigham Young University found that Twitter can be used as an early warning system to predict—and more importantly, prevent—suicides and suicide attempts. After sifting through millions of tweets throughout the United States over a three-month period, these researchers discovered that the ratio of suicidal tweets to number of unique users closely matched the actual suicide rate in each state. The troubling tweets contained statements about suicide or keywords associated with bullying and other risk factors for suicide.
Since most tweets are public, the researchers believe that suicide prevention groups may be able to provide some services through Twitter. About 15 percent of tweets are identifiable by state location, so state health departments may also be able to respond. The researchers hope to develop an app for other social media sites to evaluate posts for suicide risk and notify school counselors when appropriate. Of course, schools may need permission to monitor the students’ content first.
2Relationship Conflict And Divorce
A recent study from Boston University found that heavy usage of social media sites is a good predictor of relationship conflict and divorce in the United States. From 2008–2010, a 20-percent rise in a state’s Facebook users was associated with a 2-percent increase in the divorce rate in that state. When these researchers analyzed the effect of social media usage on marriage at the individual level, they found that nonusers were 11 percent happier with their relationships than heavy users. Among nonusers of social media, about 16 percent considered deserting their spouse at one time or another. For heavy social media users, that number doubled to 32 percent.
The researchers emphasize that social media usage appears to be a strong predictor, not a direct cause, of relationship problems and divorce. It may be that unhappy spouses simply spend more time on social media, and of course, not all heavy users of social sites have marital problems.
A different study from the University of Missouri also found that heavy Facebook usage was a strong predictor of relationship conflict and divorce, but this trend was much greater for couples in relationships of three years or less. This study also examined the effect of Twitter usage on marital satisfaction and found similar results. Twitter-related conflicts actually occurred in relationships of all lengths.
Despite privacy concerns, the concept of “predictive policing” is gaining momentum in local law enforcement. According to Matthew Gerber of the University of Virginia’s Predictive Technology Lab, even seemingly innocent tweets may help police officers predict where certain types of crime are most likely to occur.
“What people are tweeting about are their routine activities,” Gerber explained. “Those routine activities take them into environments where crime is likely to happen. So if I tweet about getting drunk tonight, and a lot of people are talking about getting drunk, we know there are certain crimes associated with those things that produce crimes. It’s indirect.”
Indirect but effective, at least for 19 of the 25 types of crime studied in Chicago. Twitter-based statistical analysis is especially good at predicting stalking, criminal damage, and gambling, though it’s not so helpful with arson, kidnapping, and intimidation.
Gerber’s technique analyzes historical crime data based on type of crime and neighborhood location, compares it to tweets tagged with the tweeter’s location from the same time period, and then predicts the location of future crimes by matching keywords in tweets that are correlated with certain crimes. From a map showing the predicted crime areas, police can deploy their resources as needed.
For criminal damage cases in Chicago, some of the tweets most predictive of imminent crime contained keywords such as “Blackhawks,” “Bulls,” “center,” and “united” and came from around the basketball arena. Those words seem to make sense, but the keywords associated with prostitution (“continental,” “lounge,” “studios,” “village,” and “Ukrainian”) are harder to fathom.
The New York Police Department has contacted Gerber about using this type of analysis in Queens and the Bronx, but Gerber doesn’t believe there will be widespread acceptance of this technique until it’s proven to reduce crime rates.