This research introduces a breakthrough method that can automatically discover important association patterns among different population subgroups in complex data, helping researchers more accurately analyze characteristics across different populations and providing a scientific basis for personalized decision-making.

This study analyzes COVID-19's impact on e-commerce using three years of sales data from 339 Chinese cities on Alibaba's platform, revealing two key findings: first, e-commerce sales followed a pattern of decline and recovery during the pandemic, demonstrating digital resilience; second, logistics capacity emerged as the critical operational driver affecting sales fluctuations. These insights provide valuable guidance for platforms and policymakers in digital strategy planning and logistics infrastructure investment.

This work advances causal inference by introducing a doubly robust estimator for ATE (average treatment effect) that ensures consistency, dimension-free scalability, and valid statistical inference, validated through real-world applications.

This research advances algorithmic fairness through multi-dimensional measurement, interpretable enhancements, and a novel no-worse-off metric, while balancing fairness-economic trade-offs.

The PIPL negatively impacted data-intensive firms, especially in B2C sectors, but those with stronger analytics and AI talent better mitigated declines in revenue, productivity, profitability, and expansion efforts.

China exhibits lower GPT exposure than the U.S., with impacts concentrated in knowledge jobs, finance, and premium cities. Sectoral/regional inequality may intensify, requiring targeted reskilling and balanced AI diffusion strategies.

AI-personalized services combined with consumer participation nudges boost purchases by 22% through effort-driven preference alignment. AI-consumer co-creation maximizes outcomes, outperforming solo AI or nudging strategies.

The Consumer Preference Transformer (CPT) integrates deep learning with sequential search theory to predict dynamic preferences while explaining decision-making processes, outperforming existing models in accuracy and interpretability.

Customizing LLMs via domain-specific theory and supervised fine-tuning (SFT) bridges gaps in expertise, trust, and satisfaction between AI and human doctors in medical consultation.

Last-mile home delivery significantly boosts sales and customer spending on online retail platforms, despite its high costs. Using advanced machine learning models, it also highlights strategies to optimize delivery capacity while ensuring fairness and maximizing profits.

Human-aligned framework evaluates LLMs: GPT-4 exceeds humans, reveals intelligence tradeoffs, RLHF hinders creativity, predicts labor needs, guides AI adoption.

Customers browse more products but make fewer clicks and purchases, while clicks and purchases from sponsored ads increase. Additionally, customers quickly adapt their expectations and behaviors in response to de-personalized search results. These findings provide valuable insights for platforms and regulators to balance privacy concerns with user experience and market performance.

When e-commerce platforms provide sellers with more market analytics tools and information, sellers launch more innovative products. Specifically, sellers adopt new attributes previously absent in their product lines and update products more quickly. These new products tend to be priced higher and generate more revenue. The research also reveals that sellers mimic product features from other sellers they follow through analytics tools, leading to customer base alignment between competing sellers. This indicates platforms can effectively drive innovation by providing market information, without additional incentives.

Combining human ideas with AI refinement optimizes creative output in product ideation. We integrate human creativity with AI refinement (human draft + AI polish) offers firms a strategic hybrid workflow to maximize innovation efficiency.

A large-scale Alibaba experiment simulating data regulation reveals sharp declines in customer engagement and sales, disproportionately harming niche merchants and consumers reliant on tailored matches.

As serialized digital content becomes more popular, small partitioning (SP) increases total consumption, while large partitioning (LP) boosts progression rates and deeper engagement. These findings offer valuable guidance for optimizing content strategies, especially for high-quality and popular content.

Can showing customers others’ creative designs inspire them to create and buy their own? A large-scale e-commerce experiment reveals that high-quality designs motivate users to start projects, while simpler designs help them finish and purchase. The study highlights how social learning can reduce effort barriers in customizing products, offering valuable insights for businesses.

By conducting a large-scale experiment with over 555,800 customers on an e-commerce platform, researchers found that lower recommendation relevance leads to increased search activity, indicating a substitution effect. Different product categories show either complementary or substitution relationships, highlighting the roles of demand fulfillment and formation in channel interactions, offering valuable insights for e-commerce platform design.

Online retailers' delivery speed promises affect customer behavior and business performance. Research finds their pros and cons and proposes an optimization model and management strategies.

Research shows adding unrelated brands' free samples to e-commerce orders significantly increases the sampled brand's sales, with effects lasting up to 14 months. This method both acquires new customers and boosts sales across the brand's entire product line. Sending samples to consumers who recently browsed related products or purchased non-essential items works best. This innovation combining offline logistics with online data creates a win-win-win for platforms, brands, and consumers.

Research shows experienced blood center nurses significantly increase donation volumes, especially from less confident donors. This influence continues into donors' future visits. Therefore, charitable organizations can substantially increase overall donations without additional costs by strategically matching staff with donor characteristics (particularly low-confidence donors).

Integrating reinforcement learning (RL) with historical randomized experiments via a Bayesian recurrent Q-network enables holistic optimization of intervention sequences along customer journeys while balancing exploration-exploitation tradeoffs.

Mobile app adoption and IoT tracking synergistically transform offline retail by enhancing customer discovery and enabling hyper-localized store strategies, driving measurable gains in offline consumption.

E-commerce pickup stations serve beyond package delivery as powerful offline marketing channels. Alibaba's experiment shows that online notifications driving consumers to collect free samples at pickup points generate significantly more sales than waiting for natural walk-ins. This method screens for genuinely interested customers—those willing to make an extra trip become more likely buyers. Last-mile delivery points can transform into precision marketing locations, with online-to-offline customer flow producing superior results compared to relying solely on offline traffic.

Online-driven offline interactions boost online sales

Data Network Effect (DNE) enhances system value through data similarity and algorithmic interactions. Platforms must balance local/global algorithms to optimize personalization and systemic robustness.

Many e-book reviews come from users who haven't finished—or even started—reading. This creates a U-shaped pattern: beginners and completers give positive ratings, while those who quit midway rate lowest. Incomplete reading produces shallow reviews that hurt future sales. Consumers should be skeptical of reviews, while platforms should display reading progress to improve credibility.

This study reveals a paradoxical effect in referral program design—combining social proof (sender’s purchase status) and monetary incentives (referral rewards) nullifies their individual benefits, suggesting motivational crowding-out or signaling conflicts.

When inviting others to try products, how you phrase the invitation matters more than the reward size. Messages emphasizing helping friends benefit rather than personal gain lead to more and better referrals. This altruistic approach reduces psychological barriers, enabling more genuine recommendations to those who truly need the product. For businesses, this improves customer acquisition without additional cost.

When e-commerce sites offer registration at the beginning of your browsing experience (rather than waiting until checkout), users are not only more likely to register but also make more purchases over time. This simple design change doesn't affect immediate shopping decisions but significantly increases long-term spending. For businesses, it offers a cost-free method to boost sales.

The study reveals a paradoxical conflict in referral incentives: non-shareable promo codes enhance sender purchases, while shareable codes amplify referrals—yet combining both eliminates these advantages entirely, as they activate competing psychological mechanisms.

This study demonstrates that strategically displaying things in common (TIC) between users fosters friendship formation, particularly among weak ties from diverse backgrounds, by triggering belief updates.

Unlike traditional one-size-fits-all solutions, this research recognizes that identical treatments can produce dramatically different results across subgroups—sometimes even opposite effects. The powerful framework developed in this study precisely identifies which specific individuals will respond best to different interventions, a breakthrough that empowers organizations to efficiently deploy limited resources for maximum benefit.

This research identifies a critical tipping point in online review systems: review quantity determines whether customer feedback helps or harms conversion rates. Reviews primarily influence the initial product consideration phase rather than final purchase decisions, revealing that social proof operates most powerfully early in the consumer journey.

While monetary incentives generated more initial downloads, information-based approaches produced significantly better long-term business outcomes. Effective digital adoption strategies should focus on attracting the right users through information rather than simply maximizing download numbers through financial incentives.

When incentivizing prosocial behaviors, rewards designed for social networks significantly outperform traditional individual incentives. Experiments demonstrate that when rewards are contingent on bringing others to participate, they not only substantially improve cost-effectiveness but also activate individuals with rich offline social capital who typically show lower online engagement.

Researchers found that sending mobile alerts to existing donors works effectively, but the family replacement program—which leverages family bonds and social relationships during medical emergencies—shows distinct advantages. When patients' relatives donate on their behalf, it activates powerful social networks and obligations that formal systems cannot replicate.