Job Market Paper

• Collaboration and Knowledge Sharing in Teams: A Graph-based Deep Learning Approach 

Being revised for resubmission to Marketing Science

Collaboration and knowledge sharing are foundational to modern organizations, especially in environments where teamwork drives performance. However, understanding how individuals contribute to and learn from multiple teams remains empirically and methodologically challenging. This paper introduces a Two-Stage Hypergraph Neural Network (TSHyGNN) framework to model agent behavior in multi-agent, multi-team settings. Leveraging hypergraph structures, the model captures both within-team collaboration and cross-team knowledge transfer. To address endogeneity in team formation, we adopt a two-stage estimation approach: the first-stage hypergraph model predicts team formation and produces a correction term, which is incorporated into the second-stage performance model to account for selection bias. Using salesforce data from a large financial services firm, we show that TSHyGNN outperforms traditional econometric and network-based models in predicting team outcomes. Simulation results suggest that replacing a novice with an experienced salesperson can generate approximately $10K in direct team collaboration gains and $16K from knowledge spillovers across teams, with sales increasing by 1-3%. The deep learning architecture is designed to be interpretable by aligning its components with collaboration and knowledge-sharing processes. This approach offers a scalable and flexible tool for team design, with broad applicability to settings characterized by overlapping team structures and dynamic collaboration networks.

Keywords: Teamwork; Knowledge Sharing; Team Formation and Design; Salesperson Evaluation; Hypergraph Neural Networks; Graph-Based Deep Learning

Presentations:

• AMA Sheth Doctoral Consortium, Manchester, UK, June 2024

• ISMS Marketing Science Conference, Sydney, Australia, June 2024

Working Papers

• Qianyin Xia, Yi Zhao, Cheng He: “AI-Driven Shopping Mall Design: Uncovering Foot Traffic Patterns with Graph Neural Networks.” 

Effective shopping mall layout design is critical for optimizing foot traffic flow and maximizing overall profitability. Retail tenants often encounter challenges in selecting optimal store locations when relying solely on historical traffic data, which can lead to suboptimal decisions. Likewise, shopping mall owners must balance revenue maximization with ensuring seamless consumer movement throughout the mall. Addressing these challenges requires a holistic understanding of the entire mall layout rather than evaluating individual store performance in isolation. This study proposes a novel approach that utilizes Graph Neural Networks (GNNs) to analyze and optimize shopping mall layouts by leveraging large-scale daily foot traffic data from one of the largest malls in China. The proposed GNN framework enables predictive analysis at both the node and edge levels, allowing for the forecasting of tenant store sales and foot traffic flows. The study evaluates the impact of different tenant placement strategies, comparing clustered (same-category) and mixed (cross-category) configurations. It also investigates the strategic positioning of essential non-sales service units such as information desks, play areas, and restrooms, which serve as key traffic attractors. The findings of this research provide actionable insights for shopping mall property owners and offer data-driven recommendations for designing layouts that enhance both profitability and the consumer experience. 

Keywords: Layout Design; Customer Foot Traffic; Graph Neural Networks

Presentations:

• ISMS Marketing Science Conference, Washington D.C., June 2025

• Qianyin Xia, Yi Zhao, and Sarang Sunder: “Modeling Agent Performance with Dynamic GNNs: The Role of Network Evolution, Office Culture, and Peer Influence” 

Achieving superior sales performance is increasingly dependent on collaborative team dynamics rather than isolated individual efforts, particularly in complex, service-based industries. As organizations grow and evolve, managers face challenges in understanding how shifting office structures, peer relationships, and evolving networks influence individual outcomes. Addressing these challenges requires moving beyond static or individual-centric models to frameworks that can capture the dynamic and relational nature of workplace performance. This study develops a dynamic Graph Neural Network (GNN) architecture to model the multi-layered relationships that shape salesperson performance within a large financial services firm. The model incorporates coworking ties, office affiliations, and temporal dependencies using a network decay process, enabling the prediction of individual sales outcomes based on evolving network structures. Empirical results show that coworker connections are more predictive than office-level affiliations, and that the proposed GNN framework consistently outperforms traditional benchmarks. Simulation studies further reveal that increasing office-level connectedness raises predicted sales by 3% to 4%, while the absence of highly central agents reduces average peer sales by 1.9%, with heterogeneous effects across individuals. This research offers a scalable and interpretable framework for talent evaluation, team formation, and organizational culture design in networked environments.

Keywords: Dynamic Graph Neural Networks; Network Evolution; Peer Influence; Sales Performance Prediction; Organizational Team Dynamics

Presentations:

• ISMS Marketing Science Conference, Miami, FL, June 2023

• GSU-Emory Brown Bag Seminar, Atlanta, GA, Aug 2023

• AI Machine Learning and BA Conference, Philadelphia, Dec 2023

• AMA Winter Conference, St Peter, Feb 2024

• Qianyin Xia and Yi Zhao: “Designing the Future Organization: Deep Generative Graph Learning for Team Prediction and Talent Fit Evaluation” 

Organizational design and talent strategy are essential drivers of success in business-to-business and service-oriented industries, where team-based structures and the effective deployment of human capital shape long-term performance. However, firms often struggle to predict how teams will evolve in new markets or how individual hires will integrate into existing networks. Addressing these challenges requires moving beyond traditional, feature-based models toward approaches that capture the relational dynamics within organizations. This research introduces a novel application of Generative Graph Neural Networks (Gen-GNNs) to model collaboration patterns and simulate strategic hiring outcomes within a large financial services firm. Leveraging rich organizational data encompassing individual, team, and contract-level attributes, we explore how key factors interact within dynamic team networks. Our generative framework forecasts future team configurations across new office locations and hiring scenarios, enabling managers to anticipate whether cohesive or fragmented teams will emerge, and to evaluate whether a prospective hire will enhance collaboration or introduce redundancy. Methodologically, this work advances the application of generative artificial intelligence (Gen-AI) in organizational design by integrating graph generation with performance-based evaluation. Empirical validation demonstrates that the model accurately predicts both observed team structures and performance outcomes, providing a scalable, data-driven tool to inform talent strategy and guide market expansion.

Keywords:  Generative Graph Neural Networks; Team Composition ; Strategic Hiring; Organizational Design

• Amit Agarwal, Qianyin Xia, Yi Zhao: “Modeling Scholarly Collaboration Across Marketing Journals: A Graph Neural Network Analysis of Cross-Channel Structures on the AMA Platform” 

Presentations

·   

Modeling Salesperson Networks Using Dynamic Graph Neural Networks

ISMS Marketing Science Conference, Miami, FL, June 2023

GSU-Emory Brown Bag Seminar, Atlanta, GA, Aug 2023

AI Machine Learning and BA Conference, Philadelphia, Dec 2023

AMA Winter Conference, St Peter, Feb 2024

Collaboration and Knowledge Sharing in Teams: A Graph-based Deep Learning Approach

    AMA Sheth Doctoral Consortium, Manchester, UK, June 2024

  ISMS Marketing Science Conference, Sydney, Australia, June 2024

AI-Driven Shopping Mall Design: Uncovering Foot Traffic Patterns with Graph Neural Networks

ISMS Marketing Science Conference, Washington D.C., USA, June 2025