Epistemic Vector Analysis for the First Signal from Preprint Watch

This repository provides a fully reproducible pipeline for analyzing epistemic disruption signals in scientific preprints using a triplet classification model based on Thomas Kuhn's theory of scientific revolutions. The tool was developed as part of the Preprint Watch, a KNOWDYN sponsored project, and is designed to detect, classify, and visualize paradigm-shifting contributions within preprint datasets. Visit the project website to scan preprints for paradigm shifts in real time: https://preprintwatch.com

Triplet Model: (N, M, P)

Each preprint is classified according to an epistemic triplet:

• N: Novelty stage
• M: Method stage
• P: Paradigm stage

Each stage is mapped onto ordinal Kuhnian cycle values:

• 1: Normal Science
• 2: Model Drift
• 3: Crisis
• 4: Model Revolution
• 5: Paradigm Shift

Vector Representation

Each preprint is represented as a vector in a hybrid semantic-metric space:

$$ \vec{v}_i = \left[ (D_N, N),\ (D_M, M),\ (D_P, P) \right] $$

Where:

• $$D_k$$ is the assigned discipline
• $$k$$ in $${N, M, P}$$
• $$N, M, P \in \{1, 2, 3, 4, 5\}$$

The 48 Epistemic Scenarios

The triplet combinations are validated against 48 epistemic scenarios introduced in the 2020 arXiv paper "A Novel Kuhnian Ontology for Epistemic Classification of STM Scholarly Articles ". URL: https://arxiv.org/abs/2002.03531

These scenarios are generated from the full cross-product:

$$ M \in \{M1, M2, M3\} \\\ N \in \{N1, N2, N3\} \\\ P \in \{P1, P2, P3, P4, P5, P6\} $$

However, only 48 of these are considered epistemically valid. Each triplet encodes a coherent relationship among:

• The epistemic function of a method (e.g., new vs. refined)
• The novelty claim of empirical results
• The Kuhnian stage of the underlying conceptual model or paradigm

For example:

• Scenario 1: (M1, N1, P1) → "Existing method confirms known result under normal science"
• Scenario 48: (M3, N3, P5) → "Novel method produces new observation triggering a paradigm shift"

Signal Intensity

The epistemic signal intensity is computed from the Paradigm stage:

$$ \text{SignalIntensity}_i = P_i $$

Entropy Calculation

To capture disciplinary nonlinearity (multdisciplinary and transdisciplinary research), the cross-entropy of the triplet disciplines is computed:

$$ H_i = -\sum_{d \in D} p(d) \log_2 p(d) $$

Where $$p(d)$$ is the empirical probability of discipline $$d$$ across the triplet $$D = \{D_N, D_M, D_P\}$$

Nonlinear Signal Metric

A composite epistemic signal strength is defined as:

$$ \text{NonlinearSignal}_i = P_i + H_i $$

This metric accounts for both disruptive potential and cross-disciplinary entropy.

Epistemic Spectrogram

The final visualization is a spectrogram-style heatmap where:

• X-axis: Discipline × Role (e.g., Biology:M, AI:P)
• Y-axis: Preprints (sorted by signal intensity + entropy)
• Color: Kuhnian stage from Normal (P1) to Shift (P5), using a Baby Blue → Cool Orange palette

How to Use

1. Load and run the script
2. Upload your Preprint Watch signal file (CSV format, freely available upon request from ipcontrol@knowdyn.co.uk)
3. Let the script analyze and visualize the data
4. Extract final plots and metrics for reporting or publication

License

This project is maintained by KNOWDYN for the Preprint Watch project. For scientific use only.