Dataset
llm-jp-corpus-v4-ja_patent
The ja_patent mirror of llm-jp-corpus-v4, assembled by the LLM-jp Corpus Building Working Group at NII, reproduces the Japanese patent sub-corpus from the larger LLM-jp Corpus v4 release. It is delivered as 621 jsonl.gz files totaling 58.2 GB in compressed form, with each line holding a JSON object that includes a text field and a meta field containing the document identifier, URL, and other provenance information.
Data preview
Dataset card
llm-jp-corpus-v4-ja_patent is a Japanese-language mirror of the ja_patent sub-corpus from the LLM-jp Corpus v4, distributed by Podtech for large language model training. The dataset compiles Japanese patent documents with provenance metadata suitable for text-generation research.
What this dataset contains
The dataset is a single text-generation corpus subset containing Japanese patent documents drawn from the LLM-jp Corpus v4 ja_patent release. It is published as a mirror, with the directory structure following the upstream repository and provenance fields preserving references back to the original data providers. According to the size category tag, the corpus falls in the 1M–10M range, though the precise file count and total byte size are not stated.
Structure and fields
Records are provided in JSON format and split into a train set of 68,111 rows. Each row carries two top-level fields: a text column of dtype string containing the document content, and a meta column that itself wraps a local_path string field along with dedup-related metadata. The masked sample rows illustrate patent text beginning with the Japanese (57)【要約】 abstract marker, confirming the source as Japanese patent specifications, with file paths organized under dataset/<year>/<class>/ directories.
How analysts use it
For investment analysts, the dataset functions primarily as a research-grade text corpus rather than a live market signal, and it is most relevant to teams building Japanese-language models for tasks such as patent classification, prior-art search, or technology landscaping. It could plausibly support thematic screens of innovation activity in Japan by mining patent abstracts, but the readme does not document any investment-grade aggregations, ticker mappings, or market coverage, so it should be treated as raw training material rather than a packaged alternative-data product.
Documentation & references 1
Related datasets
Browse moreParsed eligibility criteria from 13,229 ClinicalTrials.gov studies representing the candidate pool gathered by two first-stage retrievers during TREC Clinical Trials 2021–2023, formatted as typed entity-relation graphs for reranking models.
The clinical-trials-patient-graphs dataset is a small tabular collection of patient-level clinical records distributed in Parquet format by the 2001jdev publisher. It organizes entities and relations extracted from trial data into structured rows spanning 2021 through 2023.
This dataset is a parsed snapshot of ClinicalTrials.gov records, distributed in Parquet format on the Hugging Face Hub under an unspecified license. It contains roughly 52,000 trial entries drawn from public registry filings.
clinical-trials-trec-qrels is a tabular relevance-judgment file distributed by the 2001jdev user on the Hugging Face Hub. It maps clinical-trial topic identifiers to NCT registry IDs with graded relevance scores.
The clinical-trials-trec-topics dataset is a small tabular collection of clinical-trial topic records distributed in Parquet format on the Hugging Face Hub. It appears to be derived from TREC clinical-trial retrieval topics, with rows partitioned by year.
SEC EDGAR 10-K & 10-Q financial statements for 2,600+ NASDAQ companies.
Discussion
Questions or observations about this dataset? Join the conversation on X.
No comments yet. Start the conversation on X.