KDD ADS submission on multi-channel uplift policy learning
A KDD 2027 ADS submission introduces ReAlloc for support-aware causal reallocation across multiple marketing channels.
KDDUplift ModelingDecision Intelligence
GeoAI · Machine Learning · Decision
Changjian Liu
Spatial intelligence for learning and decision-making.
I build learning systems for spatial intelligence, causal response estimation, and industrial decision-making. My work connects spatio-temporal big data, GeoAI, advertising algorithms, uplift modeling, and causal inference, with a focus on stable response laws under observational and non-stationary settings.
Core field Spatial Intelligence
Decision layer Causal response learning
Applied systems Pricing, uplift, allocation
News
Latest updates
NeurIPS submission on deployable causal action geometry
A NeurIPS 2026 submission introduces deployable causal action geometry for continuous decisions under temporal non-stationarity.
NeurIPSCausal InferenceDecision Intelligence
OpFlow preprint released on arXiv
OpFlow is now available as an arXiv preprint, introducing opportunity-conditioned choice potentials for robust OD flow prediction.
OpFlowOD FlowSpatial Intelligence
Blog
Research notes
Research Notes
Stable Response Estimation Under Temporal Drift
A research note on why observational response models become brittle when logs, policies, and environments move together.
causal inferenceuplift modelingtemporal drift
Design Notes
Designing a Spatial Intelligence Homepage
How the homepage visual language can signal spatial intelligence without turning the site into a heavy demo page.
spatial intelligencevisual designthree.js
About
Researcher profile
I am currently pursuing my master's degree in Spatio-Temporal Big Data at Peking University, after completing my bachelor's degree in Computer Science and Technology at China University of Geosciences, Beijing.
My academic background sits at the intersection of spatial intelligence and modern machine learning. On the application side, I work on industrial decision systems, including pricing, advertising optimization, and causal decision-making.
Research
Research directions
Methods that connect spatial structure, response estimation, and real decisions.
GeoAI
Spatial Intelligence
Spatio-temporal prediction, trajectory modeling, graph neural networks, spatial flow, and explainable spatial regression.
GeoAImobility flowtrajectory modeling
uplift
Causal Response Learning
Stable treatment-response estimation from non-stationary observational logs, orthogonal learning, and decision-oriented causal modeling.
upliftorthogonal learningnon-stationarity
pricing
Industrial Decision Systems
Pricing, budget allocation, multi-treatment optimization, and robust policy design for online platforms.
pricingallocationpolicy optimization
Publications
Publications (representative papers are highlighted)
Last update: Jul 2026
Multi-channel Uplift Policy Learning
ReAlloc formulates fixed-budget multi-channel marketing as simplex-constrained uplift policy learning, combining an orthogonal teacher, explanation-guided student, and support-aware local reallocation for stable production decisions.
multi-channel upliftcausal policy learningresource allocation
Learning Deployable Causal Action Geometry under Temporal Non-Stationarity
A causal response learning framework for continuous decisions under temporal drift, using anchored link-scale contrasts, orthogonal pilots, and profiled morphology selection to learn deployable action geometry.
causal action geometrytemporal non-stationarityorthogonal learning
OpFlow: Learning Opportunity-Conditioned Choice Potentials for Robust OD Flow Prediction
A mechanism-constrained framework for robust origin-destination flow prediction that learns row-centered choice potentials and separates transferable allocation laws from origin demand scale.
OD flow predictionchoice potentialsdistribution shift
Integrating trajectory data and demographic characteristics: a trajectory semantic model for predicting travel flow and conducting interaction analysis
A trajectory semantic modeling framework that combines movement traces with demographic structure for travel-flow prediction and interaction analysis.
graph neural networksspatio-temporal predictionmobility flow modeling
High precision prediction of time-varying photovoltaic power based on dynamic adjacency matrix and temporal spectral graph convolution network
A dynamic graph forecasting model for time-varying photovoltaic power prediction under volatile renewable-energy generation patterns.
graph neural networksphotovoltaic forecastingdynamic adjacency
VisitFrequency-Diffusion: Leveraging Recurrent Visits for Long-Term Individual Trajectory Forecasting
A generative trajectory forecasting study that uses recurrent visit patterns to improve long-horizon individual mobility prediction.
generative modelinghuman mobilitylong-horizon trajectory forecasting
Experience
Decision systems in practice
2025.11 - Present
Alibaba / Alimama
Worked on managed marketing pricing and ad-equity uplift modeling for advertising decision systems.
- Participated in cost optimization and downstream allocation design.
- Connected pricing response models with decision-facing evaluation.
2025.06 - 2025.10
Didi Chuxing
Modeled short-term LTV under coupon-package interventions and optimized personalized allocation under budget constraints.
- Built personalized treatment allocation logic for coupon packages.
- Online A/B test reported +6% purchase rate and +3% ROI.
Visitor Analytics
Global research reach
Region-level visitor aggregation for the academic homepage. The production path uses a Cloudflare Worker and D1, without storing IP addresses.
...Total visits
...Countries
...Regions
LoadingData source