Sogou News: 2.9 Million Chinese News Articles in Pinyin for Text Classification

A mixture of 2,909,551 Chinese news articles from the SogouCA and SogouCS corpora, categorized into 5 classes. The dataset was created by Xiang Zhang, Junbo Zhao, and Yann LeCun, with Chinese characters converted to Pinyin. Classification labels are derived from the news article's URL domain.

Use Cases

• Train a multi-class text classifier based on news article content and title.
• Benchmark natural language processing models on a large-scale Chinese news corpus.
• Study the relationship between URL domains and article topic categories.
• Analyze the characteristics of news text after conversion from characters to Pinyin.

Strengths

• Large scale with 2,909,551 total news articles.
• Contains 90,000 training and 12,000 testing samples per class for balanced evaluation.
• Provides a clear mapping of 5 classification labels (sports, finance, entertainment, automobile, technology).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Data may reflect source bias inherent to the Sogou news corpora and the time period of collection.

Provenance

Source

SogouCA and SogouCS news corpora, via Hugging Face.

Collection Method

Web crawled, with labels determined by URL domain.

Time Range

null

Freshness

null

Geography

null