Citizen scientists learn how algorithms affect their online shopping and help researchers break open the “black box” of price-personalization
By Chelsey Meyer Have you ever wondered whether you see the same online prices as other consumers? If not, you may want to after hearing about price personalization. While many Internet users may understand that algorithms affect their social media feeds, few realize that algorithms also personalize their online shopping experiences. Researchers at Northeastern University’s Volunteer Science, a platform for gamified scientific research, are studying how this personalization occurs and who it affects, and they’re tapping into the world of citizen science to do it. In 2014, Northeastern researchers published a study uncovering instances of personalized pricing on 16 e-commerce sites. Personalized pricing takes two forms: price discrimination (offering different prices to different users) and price steering (ordering ranked lists differently).
Jason Radford (left) and Luke Horgan (right) discuss maintenance of the price comparison tool at a weekly meeting. Photo: Chelsey Meyer The team found that personalized pricing could occur based on any of a number of variables, such as whether you’re logged in, your location, or your user history. Christo Wilson, assistant professor in the College of Computer and Information Science and a lead researcher on the project, characterizes the phenomenon as an “information black box” because consumers have no idea how a company’s algorithms use their data and ultimately affect them. “There’s an information asymmetry here,” Wilson explains. “A company has a lot of information about a lot of people and there’s nothing compelling them to share any of it or the data techniques they use.” The researchers teamed up with Volunteer Science, linking different prices to consumer data to learn which consumers are affected by price personalization and how it happens. “We expect the internet to look the same for us as it does for everybody else,” says Jason Radford, the project manager at Volunteer Science, “but that’s just not the case. We've developed a way to detect when different consumers see different things, but we don't know which consumers it happens to most often, or if it's unfair in any way.” Radford saw a unique opportunity to apply citizen science: Volunteer Science could tap into its volunteer base to obtain real-world data, and, at the same time, help expose price personalization to everyday consumers. By asking online shoppers to share data about themselves, the researchers could identify when price personalization occurs and whether or not it’s biased. With a grant from the Knight Prototype Fund, Volunteer Science started developing a tool for consumers. The first iteration of the price comparison tool was housed on the Volunteer Science website, where it ran a search on a user’s behalf and displayed a graphical comparison of the server price and the consumer’s price. But the format wasn’t scalable and had poor usability. Curious site volunteers had to take the initiative to visit the website. The team wanted to bring the tool to them. Luke Horgan, a developer at Volunteer Science and computer science undergraduate at Northeastern University, suggested they turn the tool into a browser extension. With a browser extension, not only could they detect personalization and collect data from participants, but they could also show participants their personalized prices while they were shopping, when it mattered most. The complex development process took place over six months, and Horgan finished the extension in December of 2016. “It’s a very resource-intensive application because it has to run an entire web browser for every single user that connects,” says Horgan. Now, when a volunteer runs a search on a supported website, the Google Chrome extension triggers Volunteer Science servers to make the same search. The extension sends the servers’ prices back to the volunteer and displays them in the consumer’s browser right next to the potentially personalized prices. All of this happens in the span of a few seconds. Since December, preliminary findings from the extension suggest some conditions in which companies personalize prices on three sites—The Home Depot, Priceline.com, and Google Flights. The Home Depot matches prices to those at the nearest physical store, and Priceline.com offers a discount to users who are logged in, but not to those who are not. Some price differences on Google Flights are caused by a load balancing algorithm, which directs what information is searched when sites experience more traffic. This primarily affects searches between smaller airports and searches for flights further in the future.
The price comparison tool reveals differences in prices offered by Priceline.com on hotels in Los Angeles. Image: Volunteer Science In the short term, says Radford, the extension is a tool to “make the personalization personal. We’re giving consumers a sense of how much it happens, how often, and when it happens specifically to them.” In the long run, extension users who choose to donate their data contribute to valuable algorithm auditing research. The researchers plan to investigate whether algorithmic price personalization can affect people based on characteristics like race, gender, or income. “It’s possible based on some sort of proxy,” Wilson explains. “Let’s say you change prices based on zipcode. That seems like a neutral thing, but it’s not because zipcode is so linked to socioeconomics and race in this country. The net effect is that you’d charge minorities different prices.” To learn more about the processes and effects of these algorithmic black boxes, Wilson says “we need real people with real cookies and real history to realistically probe the system.” Citizen scientists are crucial and the extension may be just the beginning of their involvement in data-gathering for algorithm auditing research. In the near future, Volunteer Science will explore other areas where algorithms can have unknown effects on Internet users. An extension investigating the search engine manipulation effect will debut on the site in early summer, and work is underway on a project that will examine volunteers’ social media data. To participate in the price personalization research, head over to the project page, download the price comparison extension, and be sure to donate your data in the management section. The tool is an ideal way to gain an understanding of how your online shopping experience is different from other people’s, while contributing to valuable new research on how price personalization affects Internet users on a larger scale. Chelsey is the Online Community Manager at Volunteer Science, where she oversees outward facing communications and media content creation. She has previously interned at the MIT Energy Initiative, where she developed an interest in promoting research, and is an aspiring science communicator and creator of short video sketches.
Want more citizen science? Check out SciStarter's Project Finder! With 1100+ citizen science projects spanning every field of research, task and age group, there's something for everyone!