Standard Test Criteria

1. Click Add & update button
2. The Tasks window appears. Select the task type:
   • Text similarity: Measures similarity metrics between model outputs and reference texts.
3. Click
4. Select one or more metrics
5. Click Update to apply changes

The following metrics of Text similarity are available:

Test criteria / Metric	How it tests	Best for
BLEU	Measures how precisely a model’s output matches reference text using n-gram overlap.	Evaluating translation and short text similarity.
Fuzzy Match	Measures how closely the model’s output resembles the reference text, allowing for minor differences in wording or order.	Checking approximate correctness.
ROUGE-1	Measures unigram (single word) overlap between model output and reference text.	Short summarization and text generation tasks.
ROUGE-2	Measures bigram (two-word sequence) overlap between model output and reference text.	Evaluating contextual accuracy.
ROUGE-L	Measures the longest common subsequence (LCS) between model output and reference text to capture fluency and word order similarity.	Longer text evaluation where structure matters.
ROUGE-LSUM	Measures LCS-based similarity across multiple sentences, suitable for evaluating longer summaries.	Summarization tasks.