Privacy-Aware Adaptive Diffusion-Propensity Mapping for E-Commerce Advertising Targeting

In the vast, noisy landscape of online advertising, a persistent question haunts marketers: who should they pay to speak to their customers? For years, the answer seemed obvious. The logic was simple: find the person with the most followers, the biggest crowd, and assume that a larger audience automatically means more attention. This approach treats popularity as a direct measure of influence, assuming that if millions of people are watching, millions are listening. However, the reality of human connection is far more complex. People do not exist in a vacuum; they belong to overlapping groups, share specific interests, and move between different communities. A person might have a massive following but speak only to one narrow group, while another person with fewer followers might sit at a unique crossroads, connecting distinct worlds that would otherwise never meet. This distinction between raw size and structural position is the heart of a new study that seeks to rewrite how we choose who delivers a message.

The researchers behind this work set out to test a specific idea: that a smart, adaptive system for choosing advertisers would outperform the traditional methods of simply counting followers or looking at past likes. They built a model that does not just look at how many people a creator knows, but how those people are connected to each other and to the specific product being sold. Imagine a map where creators are not just dots with numbers next to them, but bridges connecting different islands of interest. The study asks whether a system that understands these bridges, combined with the specific needs of a campaign and a rule to avoid repeating the same message to the same group, can find better results than the old ways.

To answer this, the team turned to a public dataset of 117 content creators. They did not use the actual follower lists, which can be messy and private. Instead, they created a new kind of map based on what these creators talked about and where their audiences lived. If two creators covered the same topic, like cinema or sports, and had audiences in the same country, the researchers drew a line between them. The strength of this line depended on how many of these things they shared. This created a network that represented the "opportunity space" of advertising—a map of who competes for the same attention and who connects different markets. On this map, they calculated how well each creator could act as a bridge between different groups, a concept known as brokerage. They also looked at how much people actually engaged with the content, measuring real attention rather than just the size of the crowd.

The researchers then ran a massive simulation. They created 1,000 different advertising scenarios, each with a specific product and a specific country in mind. For each scenario, they tested four different strategies for picking five creators to carry the message. The first strategy was the old way: pick the five creators with the most followers. The second was to pick the five with the highest engagement rates, regardless of the topic. The third was to pick the five who were the best bridges in the network, ignoring the specific campaign. The fourth, and most complex, was an adaptive strategy. This method picked creators one by one, constantly checking three things: did they have a history of good engagement? Did they fit the specific topic and country of the campaign? And, crucially, had we already picked someone from their community? If a creator was too similar to one already chosen, their score was lowered to force the system to look for someone who could reach a different group of people.

The results were clear and decisive. The strategy that simply chased the largest follower counts performed the worst, generating an average of about 6.27 million qualified engagements. The strategy that looked only at past engagement did better, reaching about 11.32 million, but it still missed the mark on relevance. The strategy that looked only at network structure was only slightly better than the follower count method, proving that being a bridge is not enough if the bridge leads to the wrong place. The adaptive strategy, which combined all the factors, generated an average of nearly 14 million qualified engagements. This was a 23.5 percent improvement over the best engagement-only strategy and more than double the performance of the follower-count strategy. The data showed that the old belief—that more followers always mean more influence—was incorrect. In fact, the study found a negative relationship between follower count and engagement; the biggest accounts were often the least engaged.

Beyond just finding the best creators, the study also asked a question about privacy. In a world where data protection is becoming stricter, can we still make smart decisions if we add a little bit of noise to the data? The researchers tested this by slightly scrambling the scores of the creators, simulating a situation where the exact numbers were hidden to protect privacy. They found that even with this added uncertainty, the adaptive system remained remarkably strong. When the data was disturbed to a moderate level, the system still retained 94.1 percent of its decision-making value. This suggests that a company does not need to know every single detail about a person to make a good choice; a slightly blurred picture is often enough to find the right path.

The study concludes that the future of advertising lies in a balanced approach. It is not enough to look at a single number, like a follower count, or to rely on a static map of connections. The most effective method is one that adapts to the specific moment, weighing the history of a creator's performance against the specific needs of the campaign and the diversity of the audience. By choosing creators who are not only popular but also structurally important and relevant, and by ensuring that the chosen group covers a wide range of communities, advertisers can reach more people with greater impact. This research offers a reproducible framework for doing exactly that, showing that when we move beyond simple popularity contests and embrace the complexity of how people are connected, we can make smarter, more efficient, and more privacy-conscious choices.