Privacy-Preserving User Profiling for Targeted Advertising via Homomorphic Encryption and Secure Multiparty Computation
This paper introduces H2Profile, a hybrid privacy-preserving framework that combines approximate homomorphic encryption and secure multiparty computation to enable targeted advertising with high utility and reduced latency while limiting confidentiality to a semi-honest, two-server model.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The modern internet runs on a simple, quiet exchange: you show interest in something, and an advertiser shows you something related. To make this work, platforms collect a trail of your clicks, views, and searches, stitching them together to build a profile of who you are and what you might want next. The problem is that this trail is often scattered. One company knows what you bought, another knows what you read, and a third knows what you searched. To build a complete picture, these companies usually have to send their data to a central hub. But that central hub becomes a treasure trove of private habits, and handing it over creates a risk that your most sensitive interests could be exposed or misused.
For years, computer scientists have searched for a way to let these companies work together without ever seeing each other's raw data. They have developed two main tools for this job. One tool, called homomorphic encryption, acts like a locked box that allows math to be performed on the contents without ever opening the lid. The other, known as secure multiparty computation, works like a group of people adding up their numbers by passing notes, where no single person ever sees the full total, only their own contribution. While both tools are powerful, using just one of them for this specific task has proven difficult. The locked-box method is slow and heavy when the math gets complicated, while the note-passing method can be inefficient when dealing with massive amounts of data. The question remained: could these two methods be combined to create a system that is both fast and private?
A researcher named Wenzeng Cui has proposed a new approach called H2Profile to answer that question. The system is designed to build a user profile from scattered data without ever revealing the raw details to the servers doing the work. Instead of forcing the entire process through one difficult method, H2Profile splits the job in half. It uses the "locked box" method to quickly add up the initial numbers from different sources, creating a rough draft of the user's interests. Then, just once, it converts that draft into a format that the "note-passing" method can handle. This second stage takes over to perform the trickier tasks, such as deciding which interests are strong enough to keep, normalizing the scores, and selecting the top ten items to show. By keeping the heavy lifting in the fast "locked box" stage and moving only the necessary, complex decisions to the second stage, the system avoids the bottlenecks that usually slow things down.
The researchers tested this idea using three different sets of real-world data, simulating a scenario where two companies tried to build a profile together. They compared their new system against the best existing methods that use only one tool. The results showed that H2Profile managed to keep 98.5% of the accuracy of a standard, non-private system. In other words, the ads it helped select were almost as relevant as if the data had been combined openly. More importantly, the system was significantly faster and required less data transfer than the alternatives. In a standard network setup, it took about 0.84 seconds to process a batch of 64 user profiles and moved only 34.7 megabytes of data. This was a 68% reduction in time compared to using the "locked box" method alone and a 53% reduction compared to using the "note-passing" method alone.
The study also looked closely at what information might still leak out. Even when the math is secure, the final list of top interests could theoretically reveal patterns about a user. The researchers found that their system leaked slightly less information than the other methods, with a score of 0.604 on a test measuring how well an attacker could guess a user's hidden traits. This suggests that by carefully controlling exactly what is released at the end, the system protects privacy better than simply encrypting the data and hoping for the best. However, the author is clear about the limits of their work. The system relies on a specific assumption: that the two computers doing the work will not conspire with each other. If those two servers were to collude, the privacy would break. The system also does not solve the problem of how to match a user's identity across different companies without revealing who they are, nor does it protect against a malicious actor trying to poison the data from the start.
Ultimately, H2Profile does not claim to be a magic shield that makes all data collection safe. Instead, it offers a practical engineering solution for a specific, difficult problem. It shows that by splitting a complex task between two different types of secure computation, it is possible to build a system that is both efficient enough for real-world use and private enough to protect user interests. The work demonstrates that we do not have to choose between speed and security; with the right design, we can have both, provided we accept the boundaries of the trust model and the specific rules of the game.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.