20 Newsgroups Text Corpus with Concept Drift

OpenML's '20_newsgroups.drift' dataset is a variant of the classic 20 Newsgroups text collection, likely modified to study concept drift in document classification. The dataset contains posts from 20 different Usenet discussion groups, a common benchmark for NLP tasks. Its specific drift-related adaptations are not detailed in the available metadata.

Use Cases

• Benchmarking concept drift detection algorithms on text data using document timestamps or group labels.
• Training classifiers to predict newsgroup category from post text and headers.
• Analyzing vocabulary or topic shifts over time within specific discussion groups.
• Evaluating domain adaptation techniques by treating different time periods or newsgroups as separate domains.

Strengths

• Based on the well-established 20 Newsgroups corpus, a standard NLP benchmark.
• Explicitly designed for studying concept drift, a specific and important ML challenge.

Limitations

• Specific modifications for drift, row count, and temporal range are unknown from metadata.
• Potential class imbalance or sampling bias inherent to the original Usenet collection.

Provenance

Source

OpenML, derived from the original 20 Newsgroups dataset.

Collection Method

null

Time Range

null

Freshness

null

Geography

null