Research map / Graph learning / Graph self-supervised learning
Improving IP Geolocation With Target-Centric IP Graph (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes TCGeo, which makes the target host the central node of an IP graph and aggregates multiple last-hop routers, to alleviate sparsity in GNN-based IP geolocation.
Problem
GNN-based IP geolocation with router-centric graphs suffers from sparse graphs (14.24% of targets with fewer than 10 neighbors, 9.73% isolated), which limits graph learning accuracy.
Approach
Designate the target host as the central node and aggregate multiple last-hop routers (instead of only the router with smallest last-hop latency) to build a target-centric IP graph, used for targets with fewer than 10 neighbors while router-centric graphs are kept for the rest.
Key results
- TCGeo reduces MAE to 1.271 km on New York, 2.658 km on Los Angeles and 5.189 km on Shanghai, versus 1.316, 2.793 and 5.457 km for TrustGeo.
- Target-centric graph reduces isolated nodes from 9.73% to 4.46% and sparsity ratio (targets with fewer than 10 neighbors) from 14.24% to 8.64%.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use target-centric IP graphs for targets with fewer than 10 neighbors, rather than relying on the router with the smallest last-hop latency | TCGeo vs TrustGeo · New York | MAE (km): 1.316 -> 1.271 | Helped |
| Use target-centric IP graphs for targets with fewer than 10 neighbors, rather than relying on the router with the smallest last-hop latency | TCGeo vs TrustGeo · New York | Median error (km): 0.888 -> 0.866 | Helped |
| Use target-centric IP graphs for targets with fewer than 10 neighbors, rather than relying on the router with the smallest last-hop latency | TCGeo vs TrustGeo · Los Angeles | MAE (km): 2.793 -> 2.658 | Helped |
| Use target-centric IP graphs for targets with fewer than 10 neighbors, rather than relying on the router with the smallest last-hop latency | TCGeo vs TrustGeo · Los Angeles | Median error (km): 1.786 -> 1.745 | Helped |
| Use target-centric IP graphs for targets with fewer than 10 neighbors, rather than relying on the router with the smallest last-hop latency | TCGeo vs TrustGeo · Shanghai | MAE (km): 5.457 -> 5.189 | Helped |
| Use target-centric IP graphs for targets with fewer than 10 neighbors, rather than relying on the router with the smallest last-hop latency | TCGeo vs TrustGeo · Shanghai | Median error (km): 3.619 -> 3.461 | Helped |
| Aggregate multiple last-hop routers around the target host | Target-centric IP graph construction · Three real-world datasets | Isolated-node proportion: 9.73% to 4.46% | Helped |
| Aggregate multiple last-hop routers around the target host | Target-centric IP graph construction · Three real-world datasets | Sparsity ratio: 14.24% to 8.64% | Helped |
| Increase the number of neighboring nodes | GNN-Geo model · IP graphs with varying neighbor counts | Distance error | Helped |
| Increase the number of neighbors beyond 10 | TCGeo · IP geolocation | Geolocation performance | No clear effect |
Limitations
- Target-centric graph applied only to targets with fewer than 10 neighbors; benefit beyond that threshold not explored.
- Evaluated only on three datasets in one student-abstract-length study; limited ablation.
Experimental setup
Datasets: New York IP geolocation dataset Los Angeles IP geolocation dataset Shanghai IP geolocation dataset
Benchmarks: New York IP geolocation dataset Los Angeles IP geolocation dataset Shanghai IP geolocation dataset
Metrics: MAE (km) median error (km)
Baselines: XLBoost-Geo TabNet GNN-Geo GraphGeo TrustGeo
Builds on: TrustGeo GraphGeo
Closest papers by meaning
- Probabilistic Routing for Graph-Based Approximate Nearest Neighbor Search (ICML 2024 · similarity 0.80)
- CSPG: Crossing Sparse Proximity Graphs for Approximate Nearest Neighbor Search (NeurIPS 2024 · similarity 0.79)
- Spatio-Temporal Pivotal Graph Neural Networks for Traffic Flow Forecasting (AAAI 2024 · similarity 0.79)
- GeoRouter: Dynamic Paradigm Routing for Worldwide Image Geolocalization (NeurIPS 2026 · similarity 0.79)
- Navigable Graphs for High-Dimensional Nearest Neighbor Search: Constructions and Limits (NeurIPS 2024 · similarity 0.78)
- Semantic Routing via Autoregressive Modeling (NeurIPS 2024 · similarity 0.78)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.