Trustworthy Group Recommendation

Key Contacts: Biplav Srivastava, Siva Likitha Valluru, Michael Huhns, Sriraam Natarajan

We study the problem of group recommendation, an information exploration paradigm that retrieves interesting items for users based on their profiles and past interactions/activities/history. Existing literature encourages using greedy methods, genetic and heuristic algorithms, topic diversification, and cost constraint bi-objective optimizations. Our objective is to build novel methods and useful tools for group recommendation with fairness, and drive different use cases (e.g., meal recommendation).

The underlying research directions and applications are summarized in the poster to the right (can be enlarged).

Technical Lead: Biplav Srivastava
Collaborators over the years: Siva Likitha Valluru, Sai Teja Paladi, Michael Widener, Rohit Sharma, Owen Bond, Ronak Shah, Austin Hetherington
External Collaborators: Aniket Gupta, Siwen Yan, Sriraam Natarajan, Tarmo Koppel, Sugata Gangopadhyay
Advisors: Michael Matthews, Paul Ziehl, Michael Huhns, Danielle McElwain

We introduce ULTRA (University Lead Team Builder from RFPs and Analysis), a novel AI-based system for assisting team formation when researchers respond to RFPs from funding agencies. This is an instance of the general problem of building teams when demand opportunities come periodically and potential members may vary over time. The novelties of our approach are that we: (a) extract technical skills needed about researchers and calls from multiple open data sources and normalize them using NLP techniques, (b) build teaming solutions based on constraints, (c) computationally and qualitatively evaluate our system in two diverse settings (US, India) to establish generality of our approach, and (d) create and publish a dataset that others can use.

(This research study has been certified as exempt from the IRB per 45 CFR 46.104(d)(3) and 45 CFR 46.111(a)(7) by University of South Carolina IRB#Pro00127449.)

Representative Publications

• [2024] AI-Assisted Research Collaboration with Open Data for Fair and Effective Response to Call for Proposals
  AAAI AI Magazine
  [Paper]
  
  
• [2024] Promoting Research Collaboration with Open Data Driven Team Recommendation in Response to Call for Proposals [Awarded Deployed Application Award from AAAI-IAAI 2024]
  The Thirty-Sixth Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-24)
  [Tool Websites at UofSC, IIT-R] [Demo Videos for UofSC, IIT-R] [Paper] [BibTex]
  
  


• [2024] ULTRA: Exploring Team Recommendations in Two Geographies Using Open Data in Response to Call for Proposals.
  ACM India Joint International Conference on Data Science and Management of Data (CODS-COMAD-2024)
  [Paper] [BibTex]
  
  
• [2022] ULTRA: A Data-driven Approach for Recommending Team Formation in Response to Proposal Calls.
  IEEE ICDM Workshop on AI for Nudging and Personalization (WAIN)
  [Paper] [BibTex]
  

Additional Tools

Collaborators over the years: Aniket Gupta, Biplav Srivastava, Karan Aggarwal, Sai Teja Paladi

Here, we describe some of the important tools that we have developed as part of the ULTRA effort. They started out as useful features that we then made into stand-alone capabilities recognizing their potentia for wider usage:

• KITE (right) is an unsupervised system for exploring textual data which can generate insights from a general as well as a domain-dependent perspective consisting of holistic views, entity-centric view, events view, domain-specific interpretation using industry taxonomies and a detailed full-text view transparently connecting the document to insight elements.

• We also developed a text-to-classification mapper, a tool that takes the input as a text and matching threshold as a number and returns the ACM or JEL classification codes and description based on the input text.

Representative Publications

• [2022] KITE - An Unsupervised, Effective and Inclusive Approach for Textual Content Exploration.
  [Tool Website] [Paper] [GitHub] [BibTex]
  
  


• [2022] A Text-to-Classification Mapper (Using ACM/JEL Subject Ontology Codes).
  [Tool Website]
  
  

Meal Recommendation

Key Contacts: Vansh Nagpal, Kausik Lakkaraju, Siva Likitha Valluru, Biplav Srivastava

Personal health can benefit significantly from meals planned over extended periods, whether for a single day, a three-day travel period, or an entire week. Our research addresses this through BEACON, a data-driven meal recommender system designed to help both healthy individuals and people with health conditions (such as diabetes) make optimized meal choices.

The system leverages several innovative components:

• Variable Meal Configurations: Users can customize meal structures (breakfast, lunch, dinner) with different components (main course, side dishes, beverages, desserts)
• Flexible Time Horizons: Recommendations span from single days up to multi-day periods, enabling better long-term planning
• Rich Recipe Representation (R3): Recipes are converted from text to a structured, multimodal format that captures both content and preparation processes
• Advanced Recommendation Algorithms: Using contextual bandits and reinforcement learning, the system learns user preferences and provides highly personalized recommendations
• Comprehensive Goodness Metrics: Evaluations consider duplicate avoidance, meal coverage, and alignment with user dietary constraints
• Fairness Evaluation: Ensures recommendations are unbiased and not stereotyped toward any group of food items or user demographics

The system has been deployed and tested with real users, demonstrating significant improvements over baseline methods in balancing convenience with nutritional needs. BEACON represents a novel approach to meal recommendation that moves beyond simple food suggestions to comprehensive meal planning with health optimization.

Representative Publications

• [2025] A Novel Approach to Balance Convenience and Nutrition in Meals With Long-Term Group Recommendations and Reasoning on Multimodal Recipes and its Implementation in BEACON
  Presented at WAIN workshop at IEEE ICDM 2025
  [Paper] [Demo] [GitHub]
  
• [2025] An empirical study on the use of Large Language Models (LLMs) to translate recipes into a semi-structured data format
  Conference Poster at the National Big Data and Health Science Conference
  [Poster]



• [2024] A Novel Approach to Balance Convenience and Nutrition in Meals With Long-Term Group Recommendations and Reasoning on Multimodal Recipes and its Implementation in BEACON
  Presented at National Big Data and Health Science Conference 2025
  as A Digital Twin for Increasing Adherence to Dietary Guidelines with Personalized, AI-Driven, Long Duration Meal Recommendations. [Paper] [BibTex]
  
• [2024] BEACON: Balancing Convenience and Nutrition in Meals With Long-Term Group Recommendations and Reasoning on Multimodal Recipes
  Arxiv Preprint
  [Paper] [BibTex]
  

Recipe Recommendation

Collaborators over the years: Vishal Pallagani, Vedant Khandelwal, Kausik Lakkaraju, Revathy Venkataramanan, Biplav Srivastava
External Collaborators: Priyadharsini Ramamurthy, Hem Chandra Joshi, Sathyanarayanan N. Aakur, Ram Manohar Singh

Cooking domain is a popular use-case to demonstrate decision-support (AI) capabilities in service of benefits like precision health with tools ranging from information retrieval interfaces to task-oriented chatbots. The recipes today are handled as textual documents which makes it difficult for machines to read, reason and handle ambiguity. This demands a need for better representation of the recipes, overcoming the ambiguity and sparseness that exists in the current textual documents. We constructed a machine-understandable rich recipe representation (R3), in the form of plans, from the recipes available in natural language. R3 is infused with additional knowledge such as information about allergens and images of ingredients, possible failures and tips for each atomic cooking step.

To show the benefits of R3, we also built TREAT, a tool for recipe retrieval which uses R3 to perform multi-modal reasoning on the recipe's content (plan objects - ingredients and cooking tools), food preparation process (plan actions and time), and media type (image, text). R3 leads to improved retrieval efficiency and new capabilities that were hither-to not possible in textual representation.

Representative Publications

• [2022] A Rich Recipe Representation as Plan to Support Expressive Multi-Modal Queries on Recipe Content and Preparation Process
  Workshop on Knowledge Engineering for Planning and Scheduling (KEPS), International Conference on Automated Planning and Scheduling (ICAPS)
  [Paper] [BibTex]