Performing Department
Agricultural & Resource Eco
Non Technical Summary
The Coordinated Innovation Network (CIN) in big data and modern artificial intelligence (AI) will promote multidisciplinary research into food production, processing, and consumption systems in collaboration with several institutions, including academic, governmental, and international partners. The CIN will use big data to address food systems questions, exploit a unique mobility dataset of 200 million cell phone users, and complete three immediate research applications. Most agricultural AI innovations are limited to production data analysis, indicating a clear need for new research relying on big data beyond the farmgate and AI systems to answer research questions of national scope focusing on producer marketing, food retailing, and consumer choices. We will design a unique geospatial cyberinfrastructure that combines large-scale heterogeneous mobility data with multiple other sources. Spatial and temporal data filtering and indexing will facilitate frontier food systems research. We will construct location-based heterogeneous graphs to develop new insights regarding the producer, retailer, and consumer interactions and behaviors adopting modern AI systems. Trajectory mining will enable us to leverage data correlations and construct user visitation models. These novel data science applications will support multidisciplinary frontier research of CIN members. The CIN will be established and sustained through transdisciplinary training of undergraduate and graduate students, virtual networking, and an annual research symposium. Stakeholder engagement will be maintained in all project aspects. A user-friendly database that integrates visualization and AI tools will be made available through a dedicated web interface and GitHub.
Animal Health Component
0%
Research Effort Categories
Basic
30%
Applied
35%
Developmental
35%
Goals / Objectives
The goals of this project fall under four main categories. The first involves network building. We aim to establish a Coordinated Innovation Network (CIN) that promotes big data and Artificial Intelligence (AI) systems to conduct food systems research. This CIN will build a network of collaborators interested in using big data to research questions related to food production, processing, and consumption systems, exploiting a unique dataset of 200 million cell phone users with the help of AI systems, and completing three immediate research applications. We will bring together experts from multiple disciplines, including agricultural economics, geography, and computer science, to identify a synergistic solution to siloing in the agricultural and food sciences. The network of collaborators will include internal researchers from UConn's College of Agriculture, Health and Natural Resources (CAHNR), the School of Engineering, the College of Liberal Arts and Sciences, and the School of Business, as well as external collaborators from academic institutions and governmental agencies in the U.S. and abroad.The second goal involves research. We will strategically build on the unique data science collaborations at the University of Connecticut (UConn) to create transformative research that improves academic and policymaker understanding of the U.S. agriculture and food system. Specifically, we will use mobility big data to conduct food systems research, combining multiple datasets to answer questions related to producer marketing decisions, retailer networks, and consumer behaviors. This innovative research will be conducted through the CIN by designing, integrating, analyzing, and interpreting AI and data mining applications to the food system. We will leverage AI to address critical questions in food and agriculture by automating processes that create scalable insights from big geospatial and mobility data. This includes addressing topics spanning from agriculture and food retailing to consumer choices and health.The third goal involves data science. We will design multiple novel AI systems to acquire knowledge, iteratively extract information and learn from the mesmerizing patterns in our high-dimensional input data. We will develop new deep learning approaches, formulating a flexible and powerful way to learn from the training data as a nested hierarchy of concepts, where more complicated and high-level concepts will build on simpler ones. This includes designing data analysis techniques that lead to methodological advances that will enable a more efficient and scalable platform for researchers and stakeholders to discover new producer, processor, and consumer patterns from large mobility datasets, such as identifying frequent visitations from large-scale datasets for food market decision support.The fourth goal involves stakeholders and students. We will combine transdisciplinary network building and large-scale cyberinfrastructure development for novel insights in both data and agricultural sciences, improve food system collaborations, and provide policy-relevant research findings. We will also incorporate a training program that engages graduate and undergraduate students in multiple disciplines, ensuring graduates with workforce-ready skills. Finally, we will ensure strong stakeholder involvement in all project elements through collaborations with UConn's Department of Extension.
Project Methods
A systems-based approach is necessary when studying food and agriculture due to the myriad of complex factors interacting throughout the food system. We will develop a Coordinated Innovation Network (CIN), through which we will pursue innovative techniques and create new knowledge in both data and agricultural sciences by applying big data and AI techniques to questions concerning food and agriculture. We will emphasize post-farmgate activities that have been understudied despite the increasing use of big data in production agriculture, with a primary focus on analyzing agricultural data by connecting multi-scale, multi-domain, and multi-format datasets. While we are building on pre-existing collaborations between ARE and CSE, which demonstrates the collaborative abilities of the Co-PIs, we will have additional internal and external collaborators and grow the network through annual research symposia, monthly virtual seminars, and a new multi-state Hatch project. We will collaborate with stakeholders to establish research priorities, and both undergraduate and graduate students will receive training on interdisciplinary research processes.As part of this project, we will complete three research applications in food retailing, the restaurant industry, and agricultural direct marketing (detailed below), with additional research projects forthcoming as collaboration increases. Using mobility big data, we can empirically investigate how retailers structure their supply chains by identifying storage and transportation patterns in the food supply chain network, thereby deriving deeper insights into how firms are managing the supply process of food products. Using an equity and social justice framework, we will employ AI systems to compare the neighborhood demographic and socioeconomic characteristics of warehouse vs retail locations and assess whether neighborhood-level racial/ethnic segregation and poverty are associated with individual consumer food shopping behavior and choice of grocery stores. Another highly innovative aspect of this research is the ability to define market size, assessing competition effects while also linking to questions concerning equity and access to SNAP and other food assistance resources.We will also take an interdisciplinary approach to analyze the restaurant industry. AI systems and mobility big data will be used to identify the effect of distance on choices, assess whether neighborhood-level racial/ethnic segregation and poverty are associated with consumer restaurant choice, and study the impact of COVID-19 on delivery apps (through driver identification) and changes in consumer patterns during and after the crisis. We will use machine learning to construct location-based heterogeneous graphs based on different types of business entities (supermarkets, grocery stores, fast-food restaurants, etc.), and by learning a generative representation of each node (business entities, mobile users) in these heterogeneous graphs, we can discover the hidden relationships among them. We will then use the dynamics of these heterogeneous graphs to predict the future status of each node (business) or edge (relationships).We will also use AI systems and mobility big data to answer questions concerning agricultural direct marketing not previously possible due to data constraints. Using trajectory mining, we can leverage the correlations between users, POIs, and activities by constructing the semantic relationships between user profiles, POI categories, and activity types. Using the correlations, given any two of the three components, we can build a market recommendation model predicting consumer interest in potential POI visitations. We will assess consumer behavior regarding distinct marketing arrangements on a national level, including attributes that impact consumer market choice and potential spillover effects. From a supply chain perspective, we will identify where producers are coming from and how many establishments they supply, allowing for the creation of local food network maps.As part of our dissemination and network-building strategy, we will publish our data through a dedicated webpage, ensuring stakeholder use and increased visibility for network growth. This web-based interface will be designed to interactively display and visualize the mobility data analysis results and engage relevant stakeholders (such as researchers, government officials, policymakers, and NGOs) in the decision-making process relevant to producer, processors, and consumers markets to understand and interpret essential considerations like supplier networks, retailer inventory and the neighborhood impacts of store location choice. We will use the web-based interface to interact with the location data and the deep learning modeling results to capture, manage, analyze, and display geographically referenced information.