In this study, we present a model that optimizes product assortment strategies for a firm, informed by detailed customer transaction histories. The model harnesses data on the range of products offered to customers and their corresponding purchasing decisions. This information is crucial in constructing a partial order of preferences for each customer, highlighting their choices among the presented product options. 

We take a robust approach to model the consumers' purchasing behavior. It operates on the premise that, given a set of preferences, consumers are inclined to select the product that generates the least profit for the firm. This perspective is innovative in the realm of assortment optimization, effectively merging individual customer preferences with historical sales data to inform more strategic product offering decisions.

Live-streaming and video advertising commerce have emerged as an innovative marketing tool that enables social influencers to engage customers through sequential product demonstrations. However, the profitability of live-streaming and the means by which platforms can improve their returns through this medium are not fully understood. To address this gap, we investigate the impact of information-enhanced advertisement, specifically live-streaming, on the online advertisement market with information asymmetry. Utilizing a mechanism design framework, we model live-streaming e-commerce as a position auction that incorporates endogenous information provision. Our findings highlight the crucial role of information provision in enhancing the matching efficiency between products and heterogeneous consumers, as well as in mitigating the platform's information rent. Surprisingly, we identify that the live-streaming platform may engage in strategic placement of inferior seller at the top  position and charge elevated fees, thereby creating a position and payment paradox. Furthermore, our study demonstrates that both the platform and sellers can reap benefits from live-streaming, primarily through the deterrence of competition among sellers. These findings provide valuable insights into the optimal mechanisms and strategies for improving the efficiency of live-streaming and video advertising commerce, with significant implications for managerial practices in online shopping marketing.

Given the intricate nature of consumer behavior and the operational constraints faced by businesses, some industry professionals often rely on the independent demand model (IDM) to simplify demand estimation and assortment optimization. However, this approach fails to account for cannibalization effects among

products. To address this limitation, we propose an IDM-based optimization variant that uniformly adjusts the unit revenue contributions of all considered products. For last-choice regular models that meet certain bounding conditions, our heuristic—requiring only two linear programs to solve—achieves the best possible performance guarantees when the assortment constraints are totally unimodular (TUM) and downward feasible. More generally, we propose heuristic extensions for cases where constraints are not TUM or the bounding conditions are difficult to determine. Beyond its practical simplicity, the proposed methodology also facilitates the development of efficient heuristics for several previously studied problems, either by relaxing assumptions or improving approximation ratios. Comprehensive computational tests indicate that the proposed heuristics offer reliably outstanding performance. Furthermore, an examination of a real-world dataset demonstrates that our heuristics, developed from models with enhanced predictive accuracy, yield better optimization results.

In this paper, we consider predicting travel time for aircraft operated in multi-airport systems by modeling and simulating a multiclass queuing network, which can systematically capture the complicated coupling relationship among multiple airports and terminal airspace and the complex nature of flight trajectories following different traffic flow patterns. In this multiclass queuing network model, each class of queuing network, named a class of customers, is modeled with the data of a traffic flow pattern, which is identified for a cluster of flight trajectories. Airports and airspace sectors are correspondingly modeled as networked servers with nonhomogeneous and time-varying arrival rate, service rate and server capacity to serve those classes of customers following their specific routing probabilities. Then, all of the parameters for setting up the multiclass queuing network model can be properly estimated using historical 4D flight trajectory data. To illustrate the superiority of this model, both average travel time for each class of customers, i.e., aircraft following a particular flow pattern, and the arrival time for an individual flight are predicted via simulations of a multiclass queuing network, and furthermore, compared with the real travel time. A typical example of a multi-airport system, the Guangdong-Hong Kong-Macau Greater Bay Area in China, is utilized to showcase the prediction performance of the proposed multiclass queuing network simulation model. The simulation experiments of the case study demonstrate that the proposed model well fits this multi-airports system. For most of the time periods, the percentage error (PE) of simulated average travel time and real average travel time is less than 5%. The travel time prediction for a random individual flight can achieve around 1% of the percentage error in terms of point estimation.