Good Samaritan Index Identifies Web Users Who Help Others Most (MIT Technology Review, 11/28/12)
Online health communities play an increasingly important role in many people’s lives. Here, people ask for and get for information, help, and advice about illnesses affecting themselves and their loved ones.
In addition, people receive vital emotional support. For many, this has a significant impact on their ability to cope at a time that may be the most traumatic of their lives.
But while it’s straightforward to measure the amount of information a community provides by counting visitors, posts, and replies, it’s much harder to judge the value of emotional support.
It’s even harder to identify users who provide this support most effectively. These people are hugely important because of the vital role they play in influencing the emotional state of other community users.
Which is why the work of Kang Zhao at the University of Iowa and a few pals is impressive. Today, these guys reveal how they have automatically identified influential members of an online health network called the Cancer Survivors Network.
Kang and co studied 500,000 anonymised posts, organised into 50,000 threaded discussions that took place between 2000 and 2010. Their hypothesis is that conversations can change the emotional state of the person who started the discussion and that this ought to be reflected in the emotional content of this person’s posts.
So they set out to examine the emotional content of the initial post in the discussion and see how later comments by the same person changed.
They began by manually tagging 300 randomly selected posts as either positive or negative. An example of a post with negative sentiment is: “My mom became resistant to chemo after 7 treatments and now the trial drug is no longer working :(, ” An example of a post with positive sentiment is: “Hooray! The tumor is gone, according to my doctor!”
They then extracted features useful in identifying the sentiment, such as the presence of smileys, exclamation marks, and words with positive or negative connotations. They used these features to train software classifiers to automatically spot positive or negative posts.
They then ran the classifier on the entire set of threads that had at least one reply and then one self-reply from the originator.
The results make for interesting reading. Kang and co found that the sentiment in the first self-reply is generally significantly more positive than the original post. After that, self-replies become gradually more positive.
That seems to confirm the hypothesis that other users can significantly alter the emotional state of the originator.
This also allowed Kang and co to see which users triggered the change. They looked only at the replies published before the self-replies, and they also concentrated on quick replies to rule out the possibility that other factors might be responsible for the change, such as a change in health.
Kang and co then used this data to work out which users were the most influential: in other words, which people provided the biggest and most consistent emotional boosts.
Of course, it’s possible to use traditional methods such as PageRank to find important users in a network. But interestingly, Kang and co say these methods do not necessarily find the individuals who provide the most significant emotional support.
In effect, what they have found is an entirely new way to rank members of a community based on their ability to help others—a kind of good Samaritan index.
That could turn out to be very important for these kinds of networks. Knowing who the best Samaritans are and when they leave the community (perhaps through death) is important for judging the utility of the community and how it is changing. As is the ability to spot when new Samaritans that are becoming influential.
Clearly these networks provide an important social service. So any way that this service can be maintained and even improved is surely of great value.
College Football Chain Reactions (11/1/12)
Fans of the Kansas State Wildcats probably don’t realize it, but their BCS hopes will be affected at least a little by this weekend’s University of Northern Iowa game against Western Illinois.
K-State, currently second in the BCS rankings, hasn’t played UNI this season, nor will it. But because of the way the computer generated portion of the BCS rankings work, the outcome of the Panthers game will have an impact on the Wildcats’ rankings. It works like this: K-State played Iowa State on Oct. 13 and won, 27-21. Iowa State played Iowa on Sept. 8 and won, 9-6. And Iowa played UNI on Sept. 15 and won, 27-16.
Because the computer rankings take into account not only each team’s wins and losses but the wins and losses of all the teams they played, it creates a kind of college football butterfly effect (it should be noted that the computers do not take into account margins of victory or on-field statistics, only wins and losses). So if UNI beats WIU this weekend, that will give UNI a stronger profile, which in turns gives Iowa a stronger profile, which gives Iowa State a stronger profile, and which finally improves K-State’s profile and the likelihood that they’ll play for the BCS national championship.
Sam Burer thinks this is kind of nuts. Burer, a management sciences professor in the University of Iowa’s Tippie College of Business, doesn’t think such inconsequential games should play such a prominent role in deciding the best teams in college football.
“Should games played between teams at the bottom of the heap have that much of an effect on what’s happening at the top?” says Burer. “The top rankings provided by computer systems should be more robust against the outcomes of inconsequential games.”
The idea that such an inconsequential game impacts the rankings has actually been borne out. In 2010, one of the computer programs used to determine the BCS rankings—the Colley Matrix—forgot to include in its final ranking the result of a football championship subdivision game between Western Illinois and Appalachian State. The game clearly had no direct impact on the BCS rankings because as FCS teams, neither qualify for the BCS. But thanks to the butterfly effect, forgetting to include that game had enough changes to actually have an impact.
The final BCS rankings had LSU ranked 10th and Boise State 11th. However, if the missing game had been added, the two schools would have switched places.
This, Burer points out, was a single game missing from just one of the six computer rankings the BCS uses, and that makes up just 6 percent of the BCS ranking (the other five computer rankings and two human polls make up the remaining 94 percent).
The issue with the Colley Matrix, Burer says, is that it gives equal weight to games between weaker teams, giving the UNI-WIU game an importance it does not deserve. His own formula works by also emphasizing each team’s wins and losses, but minimizes the butterfly effect by systematically changing the results of five games between weaker teams, comparing the results of each change, and devising a balanced ranking based on the comparisons.
“Our goal is to devise a ranking system whose top rankings are stable even if a ‘far away,’ inconsequential game happens to have a different outcome,” he says. “Our approach asks, ‘suppose the outcomes of just a few of the inconsequential games switched, but we do not necessarily know which ones.’”
Burer is compiling a ranking this season using his own matrix and comparing his results to the weekly BCS rankings. The BCS top five this week are Alabama, Kansas State, Notre Dame, Oregon, and LSU. The Burer top five, however, are Notre Dame, Kansas State, Alabama, Oregon, and Florida. (His ranking also includes Ohio State, which the BCS does not include because of its bowl ineligibility.)
He says his concept could eventually lead to improved rankings in other areas where head-to-head competition is not possible, such as chess, college and university rankings, or online searches.
Burer’s study, “Robust Rankings for College Football,” is published in the current issue of the Journal of Quantitative Analysis in Sports.
Saving Money While Making Your Vote Count (10/10/12)
On Election Day, when voters step to the polls and cast their ballots, not many are likely to wonder how the voting machines got there in the first place.
But someone had to deliver them, and that’s not as easy as you might think. State and federal laws require optical scan voting machines that are larger, heavier, and more difficult to transport than the collapsible ballot box of old. In metropolitan areas, thousands of these machines have to be loaded onto trucks and transported to hundreds of precinct locations. Something as seemingly insignificant as an unforeseen traffic jam or too many red lights can result in a missed delivery time, possibly leading to delays and longer voting lines on Election Day.
But getting the machines to the polling places too early can also increase the risk of voter fraud, because the more time voting machines sit around before the polls open increases the possibility they could be tampered with.
“There are numerous factors that play into distributing optical scan voting machines to polling places,” says Jeffrey Ohlmann, an associate professor of management sciences in the University of Iowa Tippie College of Business and an expert in operations research. “The number of voting machines, the number of polling locations requiring machine deliveries, and the number of trucks available to distribute them in a cost effective way are just the start of the analysis.”
That was the situation facing Hamilton County in Ohio in 2006, the county’s first election using the new, heavier machines. The county’s Board of Elections contracted Ohlmann and Michael Fry, a colleague from the University of Cincinnati, to develop the most cost-efficient model to distribute thousands of optical scan voting machines to 531 polling places in and around Cincinnati in the days leading up to the May 2006 primary election.
Ohlmann and Fry had to overcome several unknowns in their work. For a given set of polling locations, they had to determine how to deliver the voting machines using the fewest number of trucks driving as few miles as possible to save money. One complicating factor was that each delivery had to be made at a time when the poll worker would be available to accept and secure the delivery at each location.
For the May election, poll workers for each precinct specified the day and time (typically a one- to four-hour time window) during which they would be available. To ensure delivery during these specified time windows, Ohlmann and Fry had to appropriately factor in traffic to accurately estimate travel times.
The poll workers’ delivery day and time specifications greatly complicated the development of a voting machine distribution plan, though, so Ohlmann and Fry worked with election officials to solicit more flexible time windows from the poll workers for the general election in November 2006.
While dictating delivery times and days would result in the least costly distribution plan, Ohlmann says that this wasn’t possible because some polling locations cannot easily accept delivery at certain times, and some poll workers’ schedules are inflexible.
“In the end, we compromised with the poll workers,” Ohlmann says, and a new protocol with more flexible delivery windows was developed. Ohlmann conservatively estimates that the new protocol resulted in one fewer truck and five percent fewer miles traveled for the general election.
In addition to the distribution of voting machines, Ohlmann is using his operations research expertise to try to improve other facets of election administration so that voters can more easily cast ballots. For instance, by using operations research to more accurately project voter turnout and to account for voter arrival patterns, election officials can better allocate the limited number of voting machines to precincts with greater anticipated need. This shortens lines and increases voter participation.
Ohlmann also says that operations research can potentially increase the equity and openness of elections by reducing the opportunity for election officials to distribute machines in a way that favors certain parties or candidates.
“If you use an analytical formula based on accepted facts to develop an allocation and distribution plan, and this objective plan varies significantly from one actually devised and implemented, it definitely raises questions about why,” he says.
Some of Ohlmann’s work with the Hamilton County Board of Elections can be found in the paper "Route Design for Delivery of Voting Machines in Hamilton County, Ohio," that was published in the journal Interfaces in 2009.
Study Finds Cut-Price Merchants Less Reliable (6/8/12)
Remember that scene in “Guys and Dolls” where the band girl’s grandfather gets a screaming deal on a street watch? And then it turns out to be fake? Well, online merchants who use dollar signs and advertise low prices in search engine results may be the equivalent of that street vendor. Loud voice, low price, but if it breaks, you’re on your own.
Researchers in a recent University of Iowa study performed 243 internet searches on Google, Yahoo and Microsoft for digital cameras made by major manufacturers and sold online. They looked at how search engine results on the first few pages correlated to merchants’ BBB business reviews.
The study found that merchants who used dollar signs in their Google results earned markedly lower grades from the BBB. In addition, many of the merchants who appeared only in the paid results received low marks from the BBB. Iowa’s Professor Gautam Pant stresses that this doesn’t mean all retailers in the paid results are less reliable. But it does mean that merchants who can’t compete on reliability can get more visibility by paying for results.
Pant also found a correlation between lower price and lower reliability. “Merchants who can’t compete on service attract customers using the lure of a good deal,” he says.
Remember, reliability is an important consideration when making a major purchase. You can always Start With Trust by checking the company’s track record at www.bbb.org. Selecting a purchase on price alone can lead to trouble.
The study “Can visible cues in search results indicate vendors’ reliability?” was published recently in the journal Decision Support Systems.
- MIS Students Win 2012 Student Innovation Award (4/11/12)
- Tippie Professors Appear on WorldCanvass (2/29/12)
- Tippie Professor Finds Online Merchants Who Emphasize Price Are Less Reliable (2/9/12)
- Anstreicher Named INFORMS Fellow (11/14/11)
- Anstreicher Named to CIC Leadership Program (10/27/11)
- Study Says Low Taxes Just One of Many Factors Needed for Global Economic Development (10/11/11)
- UI Professor Designs App to Help Fantasy Football Players Draft Their Teams (8/16/11)
- Ohlmann Interviewed on Iowa Public Radio (Iowa Public Radio, 6/23/11)
- NFL Locks Out Fantasy Football, Too (6/15/11)