from the Journal of Agriculture, Food Systems, and Community Development |
|
|
JAFSCD is the world’s only community-supported journal. JAFSCD content is open access (free) thanks to the generous support of our shareholders: the JAFSCD Shareholder Consortium, Library Shareholders, a growing number of Individual Shareholders, and our six JAFSCD Partners: |
|
|
From surveys to smart data: AI-powered data harvesting for farmers markets JAFSCD peer-reviewed article by Huy Pham and Yue Cui (both at Michigan State U)
Researchers from Michigan State University have developed an artificial intelligence system that automatically harvests data from farmers markets’ digital footprints in real time. This new approach provides policymakers and market managers with immediate, accurate insights at a fraction of the cost and time of traditional surveys, transforming how we understand and support local food economies. |
|
|
Farmers markets serve as vital nodes in local food systems, supporting economic resilience, community engagement, and equitable access to fresh produce. However, effective management and evidence-based policy-making for these markets has long relied on resource-intensive survey methods that are costly, time-consuming, and often produce outdated results, with reports sometimes released years after data collection and suffering from recall bias and incomplete records. Paradoxically, while farmers markets increasingly operate across multiple digital platforms creating rich, untapped data streams, much of the factual information traditionally gathered through expensive surveys already exists in publicly accessible digital sources like websites, social media posts, photographs, and government documents.
In a new article, Beyond self-report surveys: Leveraging multimodal large language models (MLLMs) for farmers market data harvesting from public digital resources, authors Huy Pham and Yue Cui from Michigan State University demonstrate a novel solution. Their research presents a method using multimodal large language models (MLLMs) to automatically harvest and synthesize accurate market data from these digital sources. This AI-driven approach offers near-real-time data updates at a fraction of the cost of traditional surveys. It is positioned as a powerful complement to survey methods, freeing up resources so that surveys can be reserved for capturing the confidential and subjective information they excel at gathering.
Corresponding author Yue Cui can be contacted at cuiyue@msu.edu.
KEY FINDINGS
By harvesting the digital footprint of Michigan's farmers markets, this research compiled publicly available data from websites, social media, local news, and public records. The result is a unique, large-scale dataset encompassing 424 market locations and 348 organizers for the 2024–2025 period.
When benchmarked against the primary reference dataset, this AI system successfully captured 76% of the data points traditionally gathered by surveys with complete coverage on essential operational details like schedules, locations, and vendor information.
The system demonstrated a high degree of reliability, achieving accuracy scores (F1: 0.935-0.998) across diverse tasks, from reading schedules to identifying products in images, with only minimal instances of hallucination observed.
RECOMMENDATIONS FOR POLICY, PRACTICE AND RESEARCH -
For policymakers and market managers: Adopt a hybrid data collection approach: Use AI to automate the collection of factual data (slashing costs from over US$50 to under US$1 per market) and reserve surveys for subjective insights. Establish data-sharing partnerships to enable real-time monitoring and dynamic decision-making.
- For practitioners: Implement scalable, automated data collection using standardized protocols. This allows for cost-effective, longitudinal tracking and a more holistic view of market ecosystems by integrating diverse data sources.
-
For researchers: Leverage this transferable taxonomy-driven framework across disciplines by prioritizing domain expertise over technical complexity through iterative prompt engineering; refocus survey efforts solely on capturing data that is unavailable from public digital sources.
SHARE ON YOUR SOCIALS
Researchers at Michigan State University have developed a new AI tool that can automatically gather crucial data on farmers markets from public online sources, such as websites and public records, with a high degree of accuracy. This slashes the cost of data collection from over US$50 to under US$1 per market, freeing up resources for markets to thrive and making real-time insights possible for the first time.
This isn’t about replacing surveys; it’s about working smarter. By letting AI handle the facts (like schedules and vendor info), market managers and policymakers get real-time insights and can save surveys for what they do best: capturing the personal stories and private experiences that make farmers markets so vital. The future of local food is data-smart and community-strong.
Read the full @JAFSCD article for free: https://doi.org/10.5304/jafscd.2025.144.025
#FarmersMarkets #LocalFood #AIResearch #FoodSystems #MLLM #DataScience #MichiganState
Image above: From surveys to smart data: A framework for AI-powered data harvesting for farmers markets using a multimodal large language model. Image provided by the authors. |
|
|
FALL 2025 ISSUE IS COMPLETE |
|
|
We appreciate the cover photo provided by Timothy Willms, owner of Talus Wind Ranch! |
|
The fall 2025 issue shares open call papers on a wide range of food systems topics, all aimed at making food systems “fundamentally better”—to quote columnist John Ikerd. They’re joined by viewpoints, a commentary, and reviews of five books.
On the issue cover, sheep graze peacefully on a golden fall afternoon at Talus Wind Ranch, which overlooks the Galisteo Basin in northern New Mexico, USA. The ranch participated in the New Mexico Grown Meat Pilot Program reported in the article in this issue, Farm-to-institution in the Southwest: An evaluation of the New Mexico Grown Meat Pilot Program. Learn more about Talus Wind Ranch.
You can read or download any of the articles or the entire issue for free, as always. |
|
|
Join the 5th webinar in the Sustainable Diets series, Setting the Table, on Wednesday, October 22, 12:00-1:00 PM EDT. Learn and chat about the connections between trade, economics, design, and sustainable food systems. |
This webinar series is organized by JAFSCD Shareholder Dalhousie University. |
|
|
SUPPORT JAFSCD THROUGH YOUR LIBRARY! |
|
|
|
If you are affiliated with a college or university and want to support JAFSCD, now is the time of year when libraries order new journal subscriptions. |
|
|
How do I know if my library already contributes? Check the list here (scroll down on the page).
Isn't JAFSCD open access and free? It is! But we are a community-supported journal. Libraries contribute to become JAFSCD Library Shareholders instead of buying a subscription — keeping JAFSCD free to all instead of buying access just for their faculty and students. Think of us as the CSA of journals!
Considering submitting a manuscript to JAFSCD? If your institution is a JAFSCD Shareholder, you benefit by having our US$750 APC waived automatically.
How do I recommend JAFSCD to my library? Tell your librarian why JAFSCD is important to your work, and send them this link to learn more!
Feel free to contact Amy Christian, managing editor, for details. |
|
|
NORTH AMERICAN FOOD SYSTEMS NETWORK (NAFSN) |
|
|
This email is sent to you as a notification of the newest JAFSCD articles and other occasional JAFSCD news. |
|
JAFSCD is an open access, community-supported journal! Your library, program, or organization can become a shareholder to help keep JAFSCD's content available to all, regardless of their resources. We welcome anyone to become an individual shareholder; donations are tax-deductible to the extent allowed by law.
|
|
|
|
