ANP-based Bike Network Evaluation

This repository provides a framework for evaluating bike networks by fusing Knowledge Graphs (KGs) and Multi Criteria Decision Analysis (MCDA).

Key objectives:

• Identify key evaluation metrics and criteria based on knowledge base of bike network evaluation studies.
• Structure bike network evaluation as a decision modelling task and address interactions between commonly used metrics and criteria.
• Derive a bikeability index for Zurich's road network.

Features

1. Leveraging Bike Network Evaluations' KG Parse over 270 bike network metrics and over 40 qualitative criteria from KG and use it to determine preferences.
2. Analytic Network Process
   Use Analytic Network Process (ANP), an MCDA technique, to derive priority weights for criteria, metrics and rank edges based on matrix operations.
3. Bikeability Index
   Develop a Bikeability Index based on scholarly consensus and perform sensitivity analysis on how KG structure affects index outcomes.

Installation

1. Setup Environment
   Ensure you have Python 3.11 or higher and conda is installed. Clone the repository and set up the Python environment.

   git clone <repository-url>
   cd <repository-directory>
   conda env create -f environment.yml
   conda activate anp
   

   2. Directory Structure:

      ├── main.py                 # runs the entire workflow.
      ├── anp_utils.py            # Functions to construct required matrices for the NAP approach.
      ├── context_data_load.py    # Functions to load the relevant contextual information.
      ├── data_preprocessing.py   # Functions to preprocess relevant contextual data.
      ├── constants.py            # mappings for some of the data that are described in categorical terms.
      └── data_submission/        # Contains input or generated output data
         ├── input/               # Intermediate data necessary for full workflow reproducibility
           ├── metrics.nq         # N-quads to be stored in a graph database.
           ├── ...                # ... rest of the contextual data.
           ├── zurich_kreise/     # Dataset with Zurich's districts.
         ├── output/              # All generated data from this workflow is stored here
      

• data_submission/input/: Contains the essential input files (e.g., contextual data, road network, metrics n-quad file to load into KG.
• data_submission/output/: Stores generated data such as bikeability map and sensitivity analysis results relevant for the ANP-based bikeability evaluation paper.

Data and Reproducibility:

• Input data specific to this workflow can be downloaded and should be placed in the data_submission/input/ folder to match the directory structure above. In addition, create output folder inside data_submission/ if it is not there.
• The input data was generated using methodologies described in: metrics [1] and road network [2,3] and as part of the blinded for peer review project. Contextual data was downloaded from Zurich City Data Registry together with meta-information.
• The full workflow has been tested on Windows operating system, Lenovo ThinkPad X1 (12th Gen Intel(R) Core(TM) i7-1260P 2.10 GHz) and took up to 10 min.

Knowledge Graph Setup:

Install Blazegraph:

• Ensure Java 8+ is installed.

• Download blazegraph.jar and place it in the root directory.

• Start Blazegraph by navigating to the root directory and running the following in CLI:

  java -server -Xmx4g -jar blazegraph.jar
  

• Check that the port in the running local Blazegraph instance URL, e.g., http://127.0.0.1:9999/blazegraph/ matches the port of the kg key in the config.ini file.

• You can now open the local KG URL in a browser.

  • Under NAMESPACES tab create a new namespace called anp, with Mode set to quads. Set the anp namespace to in Use. Note that the namespace's name needs to match with the name in the kg key.
  • Under the UPDATE tab choose the metrics.nq file from the input/ directory as Type RDF data and Format set to N-Quads and clicking Update.

ANP workflow:

Run the following with the ANP workflow that loads contextual data, enriches road network and constructs a limit matrix with final priority vectors and edge ranking.

python main.py

The priority vectors, bikeability maps and sensitivity analysis will be stored the output/ directory.

Tips and Troubleshooting:

• Ensure Blazegraph namespaces match those specified in the configuration. For the input files, do not change names.
• For namespace-specific issues, consult the Blazegraph wiki or ensure proper SPARQL endpoint connectivity.

Acknowledgements:

blinded for peer review

References:

[1]: blinded for review

[2]: blinded for review

[3]: blinded for review