Get Started with TempoEval

Welcome to the TempoEval documentation.

This section guides you through setting up the framework and running your first evaluation.

• Installation

  Detailed guide on installing via pip, setting up Java for HeidelTime, and configuring environment variables.

• Quick Start

  Run your "Hello World" evaluation in 5 minutes. Learn the core loop: Extract -> Compute.

Why use TempoEval?

Most RAG metrics (like Ragas or TruLens) focus on semantic context. TempoEval focuses specifically on time.

The Temporal Distinction

Query: "Who was president in 1999?"

• President Clinton (1993-2001) ✅ Temporally correct
• President Bush (2001-2009) ❌ Temporally incorrect

Both are semantically similar (both presidents), but temporally distinct. TempoEval catches this.

How Focus Time Works

The Process:

1. Extract Query Focus Time (QFT): Query "What caused the 2008 financial crisis?" → {2008}
2. Extract Document Focus Times (DFT):
3. Doc 1: "Lehman Brothers collapsed in 2008..." → {2008} ✅
4. Doc 2: "The 1929 Wall Street Crash..." → {1929} ❌
5. Doc 3: "COVID-19 pandemic in 2020..." → {2020} ❌
6. Compare: Match QFT with each DFT using set intersection
7. Score: Temporal Precision = ⅓ = 33% (only Doc 1 matches)