XLM-R QA Register Selector
This XLM-RoBERTa Base classifier selects web documents for question-answer mining. It does not extract the answer spans.
The trained task
Training uses register-labelled material in English, Finnish, Swedish and French. The model distinguishes NOT_QA from QA_NEW. Its card uses inputs of up to 512 tokens. This selector is a separate trained stage from the token classifiers that mark questions and answers.XLM-R QA Register Selector: model card and trained artifactsXLM-R QA Register Selector: output labels and configuration
Practical assessment
Practical guidance. Use the selector to narrow a collection before span extraction. Inspect both rejected and accepted pages: an error at this stage can remove useful question-answer pairs before the extractor sees them.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | xlmr-qa-register |
| Base model | XLM-RoBERTa Base.XLM-R QA Register Selector: model card and trained artifacts |
| Labels | NOT_QA and QA_NEW.XLM-R QA Register Selector: output labels and configuration |
| License | CC BY-SA 4.0.XLM-R QA Register Selector: model card and trained artifacts |
Related models
- XLM-R English QA Extractor
TurkuNLP · QA web mining
The English QA Extractor is a separate token classifier for locating question and answer spans.
Launch date not confirmedDownloadable - XLM-R Finnish QA Extractor
TurkuNLP · QA web mining
The Finnish QA Extractor labels question and answer spans within source documents.
Launch date not confirmedDownloadable