A common challenge for marketers is understanding and engaging with people who have “thin” data profiles, such as donors with little demographic information or brief interaction histories that leave organizations without a clear picture of their interests.
Jiyeon Hong, assistant professor of marketing at the Costello College of Business at George Mason University, helped develop Attributed Graph Contrastive Learning (AGCL), a machine learning model that addresses this challenge, helping organizations make more of the data they already have in a way that respects privacy.
The research team consisted of Hong and co-authors Qing Liu of University of Wisconsin–Madison and Wenjun Zhou of University of Tennessee–Knoxville. Their working paper has been made available on the SSRN preprint server.
“Our framework enables more accurate inferences by synthesizing even minimal data, which is often all that’s available to smaller entities,” Hong says.
The team tested AGCL using data from DonorsChoose.org, a platform where public school teachers can post requests for classroom project funding. More than 70% of donors on the platform give only once, which limits the insights that traditional methods can generate. AGCL was used to “fill the gap” by connecting donors with similar profiles based on their brief interactions, creating a broader context for understanding each donor’s interests.
The idea behind AGCL’s unique graph-based approach is to bring together sparse data points, turning them into actionable insights. In the study, AGCL achieved a 34% success rate in recommending projects that donors in the test data-set would likely support next, compared to 24% using other methods.
The model combined three sources of information: donor-project interactions, donor connections, and project similarities. This method creates a “neighborhood” of attributes, even from sparse data, which provides a more comprehensive view of donor interests. “This layered approach aims to identify latent donor interests, providing deeper insights into diverse yet distinct areas a thin-profile donor might be drawn to,” Hong says.
“Traditional approaches might recommend projects based solely on a single past interaction, which often falls short,” she further notes. “By linking donors with like-minded users, AGCL enables a much more tailored recommendation, even from minimal initial data.”
This approach allows non-profits to maximize the value of each data point without relying on extensive tracking or additional data collection, addressing privacy concerns while enhancing engagement. For example, Hong describes how a donor who contributed to a math-themed project could be responsive to recommendations for a science or engineering project that aligns with their latent interests, which might not be immediately apparent from just one donation.
AGCL’s output offers practical insights that are easy for marketers and non-profits to interpret. Hong described how visualizations from AGCL can highlight “interest clusters” among donors, even within a limited data-set. This mapping helps non-profits understand the preferences of one-time donors, allowing them to build engagement strategies for those who may return in the future. “We’re looking at ways to use these insights not just for individual recommendations but to craft campaigns that resonate with larger groups sharing latent interests,” she added.
One of AGCL’s most promising features is its ability to operate within data privacy constraints, offering a way for organizations to derive meaningful insights without extensive data collection. With data privacy becoming paramount, AGCL presents a solution that respects these boundaries while still empowering nonprofits and small businesses to engage more effectively.
Looking ahead, Hong noted AGCL’s potential beyond donor insights, particularly in the realm of customer segmentation—a cornerstone of modern marketing. She envisions AGCL as a tool that could fundamentally reshape how organizations understand and group their audiences.
By refining segmentation techniques to uncover deeper connections and shared interests even in sparse data, AGCL has the potential to shift the entire approach to personalized marketing. As this technology evolves and becomes widely available, it may set a new standard for how organizations build lasting, meaningful connections with their audiences.
More information:
Jiyeon Hong et al, Gaining Deeper Insights into Thin Profile Donors via Attributed Graph Contrastive Learning, SSRN Electronic Journal (2024). DOI: 10.2139/ssrn.4959354
George Mason University
Citation:
Machine learning helps uncover hidden consumer motivations (2024, December 2)
retrieved 4 December 2024
from https://techxplore.com/news/2024-12-machine-uncover-hidden-consumer.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
