Internet users reg­u­larly receive all kinds of per­son­al­ized con­tent, from Google search results to product rec­om­men­da­tions on Amazon. This is thanks to the com­plex algo­rithms that pro­duce results based on users’ pro­files and past activity. It’s Big Data at work, and it’s often advan­ta­geous for users. But such per­son­al­iza­tion can also be a dis­ad­van­tage to buyers, according to a team of North­eastern Uni­ver­sity researchers, when e-​​commerce web­sites manip­u­late search results or cus­tomize prices without the user’s knowledge—and which in some cases leads to some online shop­pers paying more than others for the same thing.

This trans­parency issue is at the core of a first-​​of-​​its-​​kind study co-​​authored by five North­eastern fac­ulty and stu­dents, including assis­tant pro­fes­sors Christo Wilson and Alan Mis­love of the Col­lege of Com­puter and Infor­ma­tion Sci­ence and pro­fessor David Lazer, who holds joint appoint­ments in CCIS and the Col­lege of Social Sci­ences and Human­i­ties.

In a new research paper, the team exam­ined 16 pop­ular e-​​commerce sites (10 gen­eral retailers and six hotel and car rental sites) to mea­sure two spe­cific forms of per­son­al­iza­tion: price dis­crim­i­na­tion, in which a product’s price is cus­tomized to the user; and price steering, in which the order of search results are cus­tomized to the user.

Overall, we find numerous instances of price steering and dis­crim­i­na­tion on a variety of top e-​​commerce sites,” the authors wrote.

Among their find­ings:
• The researchers found evi­dence of per­son­al­iza­tion on four gen­eral retailers and five travel sites, including cases where sites altered prices by hun­dreds of dol­lars. Overall, travel sites showed price incon­sis­ten­cies in a higher per­centage of cases, rel­a­tive to the con­trols.
• Cheaptickets and Orbitz imple­mented price dis­crim­i­na­tion by offering reduced prices on hotels to “mem­bers.”
• Expedia and Hotels​.com steered a subset of users toward more expen­sive hotels.
• Home Depot and Trav­e­locity per­son­al­ized search results for users on mobile devices.
• Price­line per­son­al­ized search results based on a user’s his­tory of clicks and pur­chases; users who clicked on or reserved low-​​price hotel rooms received slightly dif­ferent results in a dif­ferent order, com­pared to users who clicked on or reserved expen­sive hotel rooms or clicked on nothing. How­ever, because the dif­ferent orders did not cor­re­late to prices, this wasn’t con­sid­ered price steering.

Overall, most of the researchers’ exper­i­ments on the 16 e-​​commerce sites did not reveal evi­dence of price steering or price dis­crim­i­na­tion. But price dif­fer­ences were sig­nif­i­cant in some of the cases where they did find this evi­dence, and the researchers reported that they reached out to the six com­pa­nies iden­ti­fied in the study as imple­menting some form of personalization.

Their work—which will be pre­sented at the 2014 Internet Mea­sure­ment Con­fer­ence in Van­couver next month—represents the first com­pre­hen­sive study of e-​​commerce per­son­al­iza­tion that exam­ines price dis­crim­i­na­tion and price steering for hun­dreds of actual users as well as many more syn­thet­i­cally gen­er­ated fake accounts. The researchers selected what e-​​commerce sites to study based on informal rank­ings of the “top” sites. They noted that pop­ular sites such as Amazon and eBay were excluded because they func­tion as online mar­ket­places, while com­pa­nies like Apple were omitted from the study because they only sell their own products.

Wilson noted that he and his co-​​authors didn’t seek to judge whether these prac­tices are good or bad, stressing that price dis­crim­i­na­tion isn’t an inher­ently sin­ister ploy to take advan­tage of people. In fact, it hap­pens every day when someone gets a senior dis­count at the movies or a col­lege stu­dent gets a price break on books. Indeed, coupons are tech­ni­cally forms of price dis­crim­i­na­tion, he said. The key factor is whether these prac­tices are trans­parent. In most cases, dis­counts for select groups of people are clearly posted and widely under­stood, but the North­eastern researchers said such behavior is much harder to detect on e-​​commerce sites.

This unknown served as the pri­mary inspi­ra­tion for the team’s study, which was con­ducted in April and May. The team exam­ined each site’s activity for typ­i­cally over a two– to three-​​week period. The researchers devel­oped a sophis­ti­cated method­ology that set a range of con­trols to ensure that they could accu­rately iden­tify evi­dence of price dis­crim­i­na­tion and price steering.

Here’s how it worked: Let’s say you want to buy a hammer through Sears’ online site. Not only would you search for it using your per­sonal laptop or smart­phone, but you would also fire off iden­tical queries at the exact same time from clean accounts devoid of cookies and search and pur­chase his­tory. In theory, the results should be iden­tical. There might be what is referred to as “noise”—inconsistencies that aren’t due to per­son­al­iza­tion but rather other fac­tors such as changes in inven­tory or the geo­graphic diver­sity of the dat­a­cen­ters housing these e-​​commerce sites. But if the “noise” in your laptop search is greater than the “noise” in the syn­thetic accounts, then you’ve got price dis­crim­i­na­tion or steering.

The higher-​​level goal of the group’s research, Wilson said, is to study the effect of per­son­al­iza­tion algo­rithms on the Web, which goes hand-​​in-​​hand with the pro­lif­er­a­tion of Big Data.

I get this ques­tion from people all the time: ‘How do I get the best price?’ The truth is I don’t have a good answer,” Wilson said. “It changes depending on the site, and the algo­rithms they use change reg­u­larly. Good advice today might not be good advice tomorrow. The point is that as a con­sumer, you’re at a dis­ad­van­tage unless it’s transparent.”