Cenying Yang

Cenying Yang

Auto-generated profile, based on publicly available information

Cenying Yang

City University of Hong Kong

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 May 2025 in Production and Operations Management
Verified
Journal Article
Closed access
CYCenying Yang
Cenying Yang
AAAshish Agarwal
Ashish Agarwal
PKPrabhudev Konana
Prabhudev Konana
Various sensors embedded in smart products can now provide alerts and warnings to users. However, these alerts can also influence user general behavior. We evaluate the effect of advanced driving assistance systems (ADAS) warnings on general driving behavior using automotive telematics data from a large automotive company. We categorize ADAS systems in terms of the response urgency and theorize that less urgent alerts lead to deliberate learning (System 2), improving driving behavior, while highly urgent warnings trigger automatic responses (System 1), potentially leading to risk compensation and worsening driving behavior. Our results show that the presence of a blind spot detection warning, which is less urgent with no immediate action, reduces the daily number of hard braking (speeding) events by 6.76% (9.34%). However, lane departure and forward collision warnings, which are highly urgent and require immediate responses, increase the daily number of hard braking (speeding) events by 5.65% (5.34%). Additionally, we find that, over time, the positive (negative) impact of the blind spot detection (lane departure and forward collision) feature improves (worsens). Finally, we quantify that the presence of the blind spot detection feature decreases the collision rate by 2.17% (3.14%) through a reduction in hard braking (speeding) incidents. However, the presence of lane departure and forward collision features reduces the safety benefits of preventing collisions by 1.71% (1.66%) due to an increase in hard braking (speeding) events. Our results call for the need to integrate user behavior into the design of smart features such as ADAS and related services.

21 November 2025 in Applied Soft Computing
Verified
Journal Article
Closed access
QCQian Chen
Qian Chen
CYCenying Yang
Cenying Yang
JZJiajia Zhang
Jiajia Zhang
HLHuale Li
Huale Li
LJLin Jiang
Lin Jiang

17 August 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
YZYanjie Zhou
Yanjie Zhou
HHHan Huang
Han Huang
YGYuan Gao
Yuan Gao
CYCenying Yang
Cenying Yang
QCQian Chen
Qian Chen
Spend-threshold promotions create an applied intelligent-system task: infer useful add-on products while ensuring that every proposed bundle closes a residual spending gap, stays compact, remains sufficiently relevant, and avoids unnecessary expenditure. Existing recommenders typically rank individual products or score pre-existing bundles; they do not jointly construct several complete alternatives under coupled operational rules. We design a transparent, constraint-aware retail recommender whose inference engine is Adaptive Centroid--Dinkelbach Diverse Top-K (\method). A transaction module estimates personalized relevance, basket complementarity, and popularity, while a rule layer represents promotion eligibility, bundle size, uniqueness, and minimum quality. The first bundle establishes the utility-per-spend quality anchor. Later bundles balance selection-pattern difference and new-item coverage against bounded quality deterioration and overspending. If a requested Hamming distance makes a later rank infeasible, only that diversity request is relaxed; all operational and quality constraints remain binding. A chronological counterfactual evaluation on public UCI Online Retail II transactions uses 500 correlated cart contexts from 100 independent invoices, seven common-input comparators, invoice-clustered inference, exhaustive small-instance verification, parameter sensitivity, and size-robustness tests. \method{} ranks first in diversity and second in both overspending and held-out union recall over five bundles. It retains 96.27\% of the first-bundle quality benchmark and returns all five bundles in 99.6\% of scenarios. Relative to Danna--Woodruff pool diversification, diversity increases by 10.67\% and overspending decreases by 6.82\%, at a 0.986 percentage-point reduction in mean quality retention. The reported sample means form an empirically non-dominated operating point among the tested methods, rather than evidence of universal metric dominance.

27 August 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
YLYixuan Liu
Yixuan Liu
WZWenkang Zhang
Wenkang Zhang
CYCenying Yang
Cenying Yang
Digitalization makes clinical information mobile, but the people who generate and act on the resulting information remain exposed to their physical environments. We study how ambient heat reshapes a decentralized care process on an at-home fetal-monitoring platform in China, using 1.1 million user-days that record, for the same user, monitoring, an objective fetal score, consultation, and the physician's assessment. Day-today temperature variation moves the two sides of the platform in opposite directions. Under moderate heat, users keep monitoring and their sessions record more abnormal scores near term, yet an abnormal score is about 10% less likely to be followed by a consultation within two days, with no clear evidence of catch-up within the week. Under extreme heat, physicians are more likely to flag a consultation as abnormal, conditional on the recorded score and clinical variables. The increase is concentrated on reassuring recordings, consistent with a heat-shifted prior. Together, these patterns reveal a risk-response divergence: patient escalation falls as recorded risk rises under moderate heat, whereas physician escalation rises for reassuring recordings under extreme heat. This divergence is greatest in the final month of pregnancy, when monitoring matters most. Digital resilience therefore depends not only on keeping remote care available, but also on preserving the link between recorded risk and downstream action.