Reconstruct USF Network Demo

This demo simulates semantic fluency data from a known semantic network (the USF animal subset) and evaluates how well various network estimation methods can reconstruct the original network.

It demonstrates the power and limitations of different modeling approaches as the number of simulated participants increases.

—

What This Script Does:

  1. Imports the USF semantic network (Nelson et al., 1999)

  2. Generates simulated fluency data using censored random walks over the USF network

  3. Fits new networks to the simulated data using several methods: - Naive Random Walk - Conceptual Network - Pathfinder - Correlation-Based Network - (Optionally) U-INVITE

  4. Calculates similarity metrics between the estimated networks and the original USF network

  5. Exports evaluation results to a CSV for each simulation round

—

Key Parameters:

  • numsubs: Number of pseudo-participants to simulate

  • listlength: Number of items per fluency list

  • methods: List of network estimation techniques to apply

Performance Metrics:

  • Cost: Structural difference between estimated and true network

  • SDT (Signal Detection Theory) measures: Hits, misses, false alarms, correct rejections

—

SNAFU Functions Used:

  • snafu.read_graph

  • snafu.gen_lists

  • snafu.naiveRandomWalk, conceptualNetwork, pathfinder, correlationBasedNetwork

  • snafu.cost, costSDT

  • snafu.DataModel, snafu.Fitinfo

Output:

  • usf_reconstruction_results.csv: A line-by-line record of how each method performed as participant count increased

Note:

Hierarchical U-INVITE is not supported in this demo, but code structure hints at how it could be added in future experiments.