Analytics

Training results and comparison of your models on test data.

LanguageModelMetricDownload results JSON ↓

Accuracy

—

Macro F1

—

ROC-AUC

—

No results for this language and model.

Model comparison

Confusion matrix

ROC curves

ROC curves and ROC-AUC have not been computed in the saved experiments.

Detailed statistics

Intent distribution

Classes the models were trained on: user turns from train, dev and test.

Frequent entities

Annotated slots in user turns.

Your message sentiment

Only messages from your session in the selected language. Older messages without sentiment analysis are excluded.

Dialogue length, turns

Multi-turn dialogue sets only; single phrases (MASSIVE, ruSLUn) are not counted.

RU / EN · results by language

Datasets and domains differ; language scores are not a direct comparison. “—” means no experiment is available.

Error examples

NER / SENTIMENT

Entity recognition and sentiment

Amounts, currencies and dates use rules; people, places and organizations use the NER model.

NER · EN

People, places and organizations

Pretrained model · Runs locally

Babelscape/wikineural-multilingual-ner ↗

CC BY-NC-SA 4.0

Control check · n=4

Precision100.0%

Recall70.0%

F182.4%

Micro Precision / Recall / F1: exact entity type and span matching.

A small set of synthetic project examples. Not used for training; not an independent quality benchmark.

Sentiment · EN

Sentiment

Pretrained model · Runs locally

cardiffnlp/twitter-roberta-base-sentiment-latest ↗

CC BY 4.0

Control check · n=9

Precision100.0%

Recall100.0%

F1100.0%

Accuracy100.0%

Macro Precision / Recall / F1; message-level Accuracy.

A small set of synthetic project examples. Not used for training; not an independent quality benchmark.

Examples and results ↓