SciCode Wiki: Wikipedia Markdown Corpus for Scientific Concepts

A Wikipedia Markdown knowledge corpus constructed around the SciCode benchmark. The dataset consists of two parts: an original seed concept set derived from SciCode seed cards and an expanded concept set generated by a language model. The dataset was created by dabingzz and last updated on June 10, 2026.

Use Cases

• Benchmarking scientific concept retrieval systems based on the SciCode framework.
• Training language models on structured scientific knowledge from Wikipedia.
• Expanding knowledge graphs with machine-generated adjacent scientific concepts.

Strengths

• Data is structured into distinct original seed and machine-expanded concept sets.
• The expanded set adds 25 additional concepts per SciCode problem via a language model (deepseek-v4-flash).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and overall dataset scale are unknown, which may limit suitability assessment.

Provenance

Source

Wikipedia pages retrieved for concepts related to the SciCode benchmark.

Collection Method

Concepts were sourced from SciCode seed cards and expanded using the deepseek-v4-flash language model, followed by Wikipedia page crawling.

Freshness

Last updated 2026-06-10 07:41:55; freshness should be verified.