Skip to content

How to Create a Temporal Evolution Experiment

Complete guide for setting up experiments to track semantic change across historical periods.

Overview

Temporal evolution experiments analyze how term meanings change over time by combining historical documents with anchor terms across defined time periods.

Prerequisites

Before creating an experiment:

  • [ ] OntExtract installed and running
  • [ ] User account created and logged in
  • [ ] Historical documents ready for upload (spanning the target time range)

Step 1: Create Anchor Terms

First, create anchor terms to define the concepts to be tracked:

  1. Navigate to Library → Terms
  2. Click Add New Anchor Term
  3. Enter:
  4. Term text - The word/phrase to track (e.g., "agent")
  5. Domain - Subject area
  6. Notes - Research context

See Create Anchor Terms for detailed instructions.

Step 2: Create the Experiment

  1. Click Experiments in the main navigation
  2. Click New Experiment

Create New Experiment

Fill in Experiment Details

Field Description Example
Name Descriptive experiment name "Agent Temporal Evolution 1910-2024"
Description Research goals and scope "Tracking the semantic evolution of 'agent' in AI literature"
Start Year Beginning of time range 1910
End Year End of time range 2024
Status Experiment state draft, active, completed

Temporal Periods

OntExtract automatically generates temporal periods based on the date range. Options include:

  • Auto-generate - System creates periods based on document dates
  • Manual - Define custom period boundaries

Click Create Experiment to save.

Manage Temporal Terms

After creating a temporal evolution experiment, access the Manage Temporal Terms feature to configure the timeline in detail.

Accessing the Manager

  1. Go to the experiment's detail page
  2. Click Manage Temporal Terms button

Timeline Configuration

The Temporal Term Manager provides two ways to set up periods:

Method Description
Auto-generate from documents Creates artifact markers for each document's publication year
Manual Entry Manually specify time period boundaries

Adding Semantic Events

The timeline can be annotated with semantic change events:

  1. Click Add Event in the Periods & Events section
  2. Select the event type (e.g., amelioration, pejoration, drift)
  3. Specify the time range (from/to periods)
  4. Add a description of the semantic shift
  5. Link related documents as evidence

Period Cards

The timeline displays period cards showing:

  • Year - The period's date marker
  • Source badge - ARTIFACT (auto-generated) or MANUAL
  • Documents - Papers associated with that period
  • Events - Semantic change events spanning periods

Period boundaries are color-coded:

  • Green (START) - Beginning of a defined period
  • Red (END) - End of a defined period

Saving Configuration

Click Save Configuration to persist the temporal setup before proceeding to analysis.

Step 3: Add Documents

Upload Documents

  1. Go to the experiment's Document Pipeline
  2. Click Add Documents
  3. Upload files with publication dates
  4. Documents are automatically assigned to periods based on publication date

Document Requirements

For meaningful temporal analysis:

  • Multiple documents per period - More data improves accuracy
  • Date coverage - Documents spanning the full time range
  • Consistent domain - Documents from related subject areas

See Upload Documents for detailed upload instructions.

Step 4: Process Documents

LLM Text Cleanup (Recommended)

For scanned or OCR'd documents:

  1. In Document Pipeline, click the broom icon for each document
  2. Review suggested corrections
  3. Accept or modify changes
  4. Save cleaned version

Run Processing Operations

From the Document Pipeline or individual document pages:

  1. Segmentation - Split into paragraphs/sentences
  2. Embeddings - Generate vector representations
  3. Entity Extraction - Identify named entities

Use Run Local Tools for batch processing without API costs.

Step 5: Run Analysis

LLM Orchestration (Advanced)

For AI-assisted analysis:

  1. Go to the experiment's Document Pipeline
  2. Click Start LLM Orchestration
  3. Review the generated strategy
  4. Approve or modify the approach
  5. Execute the analysis

Manual Analysis

Explore results through:

  • Timeline View - Visual evolution across periods
  • Document Comparison - Side-by-side period analysis
  • Term Context - Source passages for each period

Step 6: Review Results

Timeline Visualization

The timeline shows: - Term usage frequency per period - Semantic change events - Context snippets from source documents

Export Options

Export analysis results:

  • CSV - Tabular data for spreadsheets
  • JSON - Structured results for further processing

Troubleshooting

No Documents in Period

  • Check document publication dates
  • Verify period boundaries
  • Upload additional documents for sparse periods

Processing Errors

  • Ensure documents have text content
  • Run LLM cleanup on problematic documents
  • Check API keys for external services

Missing Analysis Results

  • Verify all processing steps completed
  • Check anchor term associations
  • Review experiment status