Project topic:
Co-traveler detection for coordinated trip generation using spatiotemporal co-occurrence mining on human trajectories, with a time-expanded interaction graph representation. The goal is to detect pairs/groups who repeatedly move together (or strongly co-occur in space-time) so their coordinated trips can be modeled downstream.
Search process (initial pass):
We conducted an initial literature pass primarily through Google Scholar, using keyword combinations based on the project framing:
Co traveler detection
Travelling companions trajectories
Spatiotemporal co-occurrence mining
Recurring co-traveling pattern
Time-expanded graph trajectories
This initial pass yielded 13 papers that were closest to the topic (including papers hosted on ACM Digital Library, arXiv, and journals). We removed 2 older papers to prioritize recent work and align with the assignment expectation of recent conference/journal sources.
From the remaining 11 papers, 4 papers co-authored by Shuai An were identified, forming a coherent research lineage on co-location co-travel pattern discovery and statistical work. These 4 are used as main references for problem framing, definitions, and evaluation direction.
Scoping:
We initially struggled to scope the contribution into a clean semester project and considered two framings:
Spatial data mining framing: define co-travel events and mine recurring co-travel groups/patterns
Spatial graph modeling / learning framing: build a time-expanded interaction graph and apply representation learning (embeddings + clustering or link prediction)
After discussion with Shuai An, the project direction was finalized as "a spatial graph modeling or learning project, meaning construct a time expanded interaction graph and do embedding based clustering or classification" for co-traveler detection.
Key backbone sources (the two papers we will follow closely)
These two papers serve as the backbone for problem setup and evaluation structure:
Discovering Traveling Companions using Autoencoders (2020): Selected as the primary baseline and reference for a representation-learning approach to companion discovery.
Graph-based representation for identifying individual travel activities with spatiotemporal trajectories and POI data (2022): Selected as the main reference for constructing a mobility graph representation and learning embeddings from it.
Paper shortlist (11) and what evidence each provides
A) Co-occurrence / co-location
Towards geographically robust statistically significant regional colocation pattern detection (2022)
Evidence: definition evidence (co-location patterns), method evidence (significance-based mining), evaluation evidence (demonstrates why naive co-occurrence leads to spurious results).
Reducing False Discoveries in Statistically-Significant Regional-Colocation Mining: A Summary of Results (2024)
Evidence: method evidence (false discovery / multiple comparisons control), evaluation evidence (reduced false discoveries and computational cost).
Discovering Super-Colocation Patterns: A Summary of Results (2025)
Evidence: definition evidence (higher-order co-location patterns), method evidence (mining super-colocations), relevant for extending from pairwise companions to group co-travel.
Towards Recurring Co-traveling Pattern Detection: A Summary of Results (2025)
Evidence: definition evidence (recurrence-based co-travel), evaluation direction for recurring/group co-travel discovery which matches the coordinated-trip motivation.
B) Trajectory mining
Travel Trajectory Frequent Pattern Mining Based on Differential Privacy Protection (2021)
Evidence: method evidence (trajectory frequent pattern mining), data evidence (trajectory record handling), evaluation evidence (pattern mining with privacy constraints).
C) Graph representation and time-expanded graph
Graph-based representation for identifying individual travel activities with spatiotemporal trajectories and POI data (2022)
Evidence: method evidence (constructs weighted directed mobility graphs), representation evidence (embedding approach), evaluation evidence (classification task pipeline).
Offline map matching using time-expanded graph for low-frequency data (2021)
Evidence: modeling evidence (time-expanded graph construction), robustness evidence (handles low-frequency trajectories), useful for justifying time-expanded modeling choices.
D) Time-aware mobility pattern mining
Exploring human mobility: a time-informed approach to pattern mining and sequence similarity (2025)
Evidence: method evidence (time-aware pattern mining and similarity), supports feature design and ablations for time-sensitive co-travel definitions.
CoBAD: Modeling Collective Behaviors for Human Mobility Anomaly Detection (2025)
Evidence: modeling evidence (collective event sequences and co-occurrence event graph), relevant as a modern example of modeling cross-person spatiotemporal dependencies with graph structure.
E) Direct co-traveler detection baselines
Discovering Traveling Companions using Autoencoders (2020)
Evidence: method evidence (deep representation learning for companions), evaluation evidence (baseline comparisons), used as the primary baseline backbone.
Discovery of Travelling Companions from Trajectories with Different Sampling Rates (2020)
Evidence: robustness evidence (sampling mismatch), definition evidence (companion criteria under unequal sampling), informs a defensible co-travel definition for course project constraints.
Summary of search results (required: DBLP, Google Scholar, Amazon)
DBLP search summary
We used DBLP primarily to confirm authors and find closely-related CS venue entries (ACM workshops / SIGSPATIAL / arXiv CoRR indexing). DBLP indexing supports tracking the Shuai An / Shekhar line of work and related co-occurrence / co-travel mining directions, including ACM-published “summary of results” entries and related arXiv-indexed work.
Google Scholar
Google Scholar was the main engine for discovering cross-venue work connected to:
Deep learning companion discovery (autoencoder baseline)
Mobility graph construction using trajectories and POI (graph representation backbone)
Time-expanded graph modeling for low-frequency trajectories (supporting justification)
Time-aware pattern mining and similarity measures (feature and evaluation inspiration)
Amazon.com search
We searched Amazon for trajectory, mobility, and spatial data mining references to cite for background definitions, standard methods, and terminology. Main books identified:
Computing with Spatial Trajectories
Handbook of Mobility Data Mining
Spatial Data Mining: Theory and Application
Availability check (U of M library and peers):
All papers are available in PDF format for downloading.