Knowledge Graph: Transit Network Analysis Using Translink Data, Python and Gephi

This research demonstrates how open transit data, combined with knowledge graph modeling and powerful network visualization tools like Gephi, can unlock insights about public transportation systems. I collected real-time data from Vancouver’s TransLink API, cleaned and transformed it using Python, and built a network graph that maps relationships between transit stops and routes. The graph analysis revealed patterns, vulnerabilities, and opportunities that can support data-informed transit planning.

Introduction

Public transit systems are complex networks that connect various parts of a city, facilitating daily commutes for millions of people. Understanding the intricacies of these networks is crucial for optimizing traffic flow, improving passenger experiences, and planning future urban development. This paper aims to unravel these complexities using a data-driven approach to network analysis and visualization.

Data Collection

The data was sourced from the local transit authority’s open API, which provides real-time information on bus routes, stops, schedules, and vehicle movements. An extensive data retrieval process was set up to query the API for various endpoints, collecting information about:

• Bus Stops: Names, region, identifiers
• Active Buses: Vehicle IDs, destinations
• Bus Routes: Trip start and end locations

Python scripts using requests and pandas handled data fetching and normalization. After parsing thousands of records, I created clean node-edge datasets:

• Nodes = Transit Stops
• Edges = Routes between stops, with weights derived from frequency

Each node was enriched with a region, and each edge included a Route ID and weight (number of buses).

Data Preprocessing

The raw data was subjected to rigorous cleaning and transformation to ensure accuracy and consistency. Python scripts were developed to automate these tasks, including:

• Parsing and normalizing JSON responses from the API
• Merging data from multiple sources to align stops with routes
• Calculating the weights of edges based on route frequency between stops

Network Graph Construction

Using Pythons NetworkX, I modeled the transit network as a graph library, where nodes represent transit stops and edges represent bus routes connecting these stops. Each edge was assigned a weight proportional to the number of buses traveling that route, capturing the intensity of service.

Visualization with Gephi

Gephi was utilized to visualize the network graph. The nodes and edges datasets were imported into Gephi, where nodes were arranged using force-directed algorithms to reveal patterns in the network’s structure.

• ForceAtlas2 layout for spatial clustering
• Node size: by degree (number of connected routes)
• Color: by region/modularity class
• Labels: Top stops by centrality

Descriptive Analytics

Initial observations of the network topology showed:

1. High-Degree Centrality Stops

Hub nodes: Nodes like “Downtown Waterfront Station”, “Coquitlam Central Station”, and “Bridgeport Station” appear frequently and connect to many others, these are referred to as hub nodes. Their high node size and central position in the graph indicate high degree centrality.

2. Regional Clusters and Modularity

The network naturally forms regional clusters visible in color-coded communities from Gephi. As seen above,

• Surrey/Langley forms one cluster.
• UBC/Vancouver/Downtown forms another.
• Maple Ridge/Coquitlam is separate. This reveals geographical modularity in the transit system.

Surrey/Langley

UBC/Vancouver/Downtown

3. Peripheral Nodes

Stops like “Ruskin”, “Bluewater”, or “Snug Cove” are peripheral with very few links. These likely represent edge-of-network zones with limited bus routes.

Snug Cove/Bluewater Peripheral Nodes

4. Most Frequent Source — Target Pairs

From the edges data, stops like"Coquitlam Central Station" → "Haney Place Exchange" and"Guildford Exchange" → "Newton Exchange"are repeated in different routes, indicating high-traffic corridors.

5. Network Density

The graph shows a sparse but strategically connected network — most stops are not universally connected but optimized through hub-and-spoke patterns, balancing coverage and efficiency.

Inferential Analytics

1. Regional Transit Equity Analysis

From the cluster sizes and number of routes, one can easily infer:

• Urban centers such as Downtown, Surrey, Richmond are well-served.
• Outer regions like Bowen Island, Deep Cove have fewer connections.

This points to transit inequality that can inform future expansions.

2. Route Redundancy Risks

Highly connected routes serve popular node pairs. An unexpected disruption on such a route could disproportionately affect access, suggesting the need for backup or alternate routing mechanisms.

3. Student vs. Commuter Route Overlay

Stops like “Brookswood Secondary School” or “Elgin Park Secondary” suggest school-specific routes, while stops like “Brentwood Station” or “Guildford Exchange” serve commuters. As a future step, one can model time-based demand profiles for optimization (e.g., peak student hours vs. general commute).

Final Thoughts

I carried out this project to bridge the gap between open source data and actionable transit intelligence. By turning raw bus routes into a knowledge graph, I highlighted how data science, network theory, and open tools can shape smarter cities.

Conclusion

The application of network analysis and visualization techniques yielded valuable insights into the operational characteristics of the transit network. This methodology can guide future enhancements in public transit services and contribute to more sustainable and efficient urban mobility solutions.