Mixed Text Dataset: 20,000 Samples from Wikipedia, Gutenberg, and CNN/DailyMail

20,000 text samples compiled from three distinct sources: Wikipedia, Project Gutenberg, and CNN/DailyMail. The dataset was created by author 'brograrnmer' and last updated on May 5, 2026. Preprocessing involved regex cleaning to replace certain patterns with whitespace.

Use Cases

• Train language models on diverse writing styles based on the mixture of encyclopedic, literary, and news text.
• Benchmark text summarization models using the CNN/DailyMail news articles mentioned in the description.
• Study stylistic differences in human-written text across the three documented source domains.
• Pre-train or fine-tune models for tasks requiring general domain knowledge from the Wikipedia subset.

Strengths

• Contains 20,000 total text samples, providing a substantial corpus for training.
• Explicitly documents the sample count from each of its three sources: 10,000 from Wikipedia, 3,000 from Gutenberg, and 7,000 from CNN/DailyMail.
• Specifies the preprocessing method used, which was regex_cleaning.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain tasks.
• The Wikipedia dump is from March 2022, so the information may not be current.

Provenance

Source

Compiled from Wikipedia (20220301 dump), Project Gutenberg (via GutenDex API), and CNN/DailyMail (Version 3.0.0).

Collection Method

Aggregation from multiple open sources, followed by regex-based cleaning.

Time Range

The Wikipedia component is from a March 2022 snapshot. Temporal coverage for other sources is unspecified.

Freshness

Last updated 2026-05-05 08:12:08; freshness should be verified.