Empirical Dephasing Test on GDELT News Data

Empirical Media Dephasing Test: GDELT News Data

Session

2026-04-16-targeted-029

Hypothesis Tested

C2-H2: Co-Mention Dephasing Rate (PASS, 8.15)

Verdict: INCONCLUSIVE

Trend in expected direction but not statistically robust. 3/5 negative coherence trends, but only 0/5 reach significance threshold.

Method

Time-ordered chunk analysis on GDELT article titles:

1. Collect articles for each story, sort chronologically
2. Split into equal chunks of 25 articles each (minimum 6 chunks)
3. Build TF-IDF + SVD (d=12) density matrices per chunk
4. Three independent tests:

- Test A: Spearman rank correlation of coherence metrics vs time

- Test B: First-half vs second-half comparison (t-test)

- Test C: Exponential fit to per-element decay rates

Results

Figures

• fig1: Coherence and off-diagonal/diagonal ratio time series per story
• fig2: Entropy and purity evolution per story
• fig3: Summary: Spearman correlations and decay excess by story

Interpretation

A negative Spearman correlation between coherence and time indicates that

off-diagonal elements (co-mention correlations) are decaying relative to

diagonal elements (individual topic prominence). This is the empirical

signature of dephasing.

A positive decay rate excess (gamma_offdiag > gamma_diag/2) is the quantitative

measure: off-diagonal elements decay faster than the amplitude-damping-only

baseline predicts.

Limitations

1. Article TITLES only (not full text)
2. TF-IDF + SVD with d=12 may miss nuance
3. GDELT API rate limits reduced some story coverage
4. Story identification via keyword queries (not event clustering)
5. Results should be replicated with full article text and finer time resolution