METEOR¶
Metric for Evaluation of Translation with Explicit ORdering.
Overview¶
| Field | Value |
|---|---|
| Type | Generation |
| Range | [0, 1] |
| Higher is better | Yes |
Description¶
METEOR aligns generated text with reference using exact matches, stems, synonyms, and paraphrases. More flexible than BLEU for capturing semantic similarity.
Configuration¶
_target_: autorag_research.evaluation.metrics.generation.MeteorConfig
alpha: 0.9
beta: 3.0
gamma: 0.5
Options¶
| Option | Type | Default | Description |
|---|---|---|---|
| alpha | float | 0.9 | Recall weight |
| beta | float | 3.0 | Fragmentation penalty |
| gamma | float | 0.5 | Fragmentation weight |
When to Use¶
Good for:
- Tasks allowing paraphrasing
- When synonyms should be rewarded
- More forgiving evaluation than BLEU
Limitations:
- Language-specific resources needed
- Slower than BLEU