Dataset Viewer
Auto-converted to Parquet Duplicate
description
string
claim1
string
The expression "cosmetic product" is understood to mean a product as defined in Regulation (EC) No <CIT> relating to cosmetic products. The invention is particularly suitable for the application of a cosmetic product, such as a makeup product, for example to keratin fibres, and notably for the application of mascara. ...
Applicator member (<NUM>, <NUM>) for applying a product to keratin fibres, notably to eyelashes, comprising: - a core (<NUM>, <NUM>) made of polymer material and extending along a longitudinal axis (X2-X2'), - first application elements (<NUM>, <NUM>) that project from the core (<NUM>, <NUM>) and are integral with said...
Hereinafter, Radio frame structures used in 3rd Generation Partnership Project (3GPP) Release <NUM> (referred to as Long Term Evolution (LTE)) and subsequent releases will be described, and then carrier aggregation (CA) introduced in 3GPP Release <NUM> (which is referred to as LTE-Advanced) will be described. Further, ...
A radio terminal (<NUM>) configured to perform Licensed Assisted Access, LAA, which is carrier aggregation between a primary cell operated in a licensed spectrum and a secondary cell operated in an unlicensed spectrum or shared spectrum, the radio terminal (<NUM>) comprising: a transceiver (<NUM>) configured to receive...
For example, there is an inspection device that uses an ultrasonic wave or the like. It is desirable to increase the detection sensitivity. <CIT> discloses a system for monitoring tissue sample processing comprising a transmitter outputting an energy and a receiver detecting the transmitted energy. <CIT> and <CIT> dis...
An inspection device (<NUM>, <NUM>), comprising: a transmitter (<NUM>) configured to transmit a first ultrasonic wave (10w), the first ultrasonic wave (10w) including a plurality of burst waves (10b) having a first period Tp; a receiver (<NUM>) on which the first ultrasonic wave (10w) is incident, the receiver (<NUM>) ...
"Many health conditions result in significant impairment to mobility, which may be short of complete(...TRUNCATED)
"An actuator system (<NUM>) for a wearable orthotic device, the actuator system (<NUM>) comprising:\(...TRUNCATED)
"A typical solid oxide fuel cell stack includes multiple fuel cells separated by metallic interconne(...TRUNCATED)
"A method of forming an interconnect for a fuel cell stack, comprising:\ncompressing an interconnect(...TRUNCATED)
"Many solutions have become known for the packaging and storage of liquids, solids, lump and bulk ma(...TRUNCATED)
"A removable lid to close the opening of a container, in particular to reclose metal cans used for b(...TRUNCATED)
"The present disclosure relates to the technical field of air treatment, in particular to an air pur(...TRUNCATED)
"A centrifugal fan impeller, comprising a hub (<NUM>) and a plurality of fan blades (<NUM>) distribu(...TRUNCATED)
"Variable geometry turbines (VGT) can be employed in various applications, such as in commercial veh(...TRUNCATED)
"A variable geometry turbine (VGT), comprising:\n- an intake channel (<NUM>), arranged for receiving(...TRUNCATED)
"For hearing aid users, certain physical activities induce low-frequency vibrations that excite the (...TRUNCATED)
"A hearing assistance apparatus (<NUM>) for detecting user activities from within the hearing assist(...TRUNCATED)
"In recent years, the population of people raising a pet has increased, in addition attachment and i(...TRUNCATED)
"A liquid dispenser, comprising:\na tank (<NUM>) having an upper opening and a chamber defined by a (...TRUNCATED)
End of preview. Expand in Data Studio

Coarse Cleaned EP Patents

This dataset contains the description and first claim from granted European Patent (EP) documents that have been cleaned for NLP tasks such as pretraining language models.

Contains one years worth of English granted EP patents from 04 September 2024 to 04 Sep 2025.

Each line is a JSON object representing a single patent with the following fields:

  • description: The cleaned English description text.
  • claim1: The cleaned first claim in English.

Cleaning and Processing Steps

  1. Language Filtering

    • Only patents with an English description (<description lang="en">) are included.
    • Only the first claim (<claims lang="en"><claim num="0001">) is extracted.
  2. Text Extraction

    • Text is extracted recursively from child XML elements.
    • Paragraphs (<p>) are preserved and separated by double newlines (\n\n).
  3. Special Token Replacement
    Certain XML tags are replaced with tokens, with the original content removed:

    • <table> → <TAB>
    • <img> → <IMG>
    • <figref> → <FIG>
    • <patcit> → <CIT>
    • <nplcit> → <NPL>
    • <chemistry> → <CHM>
    • <maths> → <MAT>
    • <heading> → <HEAD>
    • <ol>, <ul>, <dl> → removed entirely
  4. Number Normalization

    • Numbers with optional SI units (e.g., "20 mm", "3.5 kg") are replaced with <NUM>.
    • Numbers inside words like "CO2" are preserved.
  5. Whitespace Normalization

    • Redundant spaces are removed.
    • Paragraph separation is maintained using double newlines.
  6. Comments Removal

    • XML comments such as <!-- EPO <DP n="1"> --> are removed.
  7. Trimmed outliers

    • Longest 10% of claims and descriptions (seperately) removed.
    • Shortest 2% of claims and descriptions (seperately) removed.
  8. Exact deduping

    • Removed exact duplicates based on claim1 and description.
  9. Paragraph outliers

    • Removed paragraphs less than 40 characters long.
    • Removed paragraphs greater than 10240 characters long.
    • Removed duplicate paragraphs within same document.
  10. Output Format

  • Saved as JSONL, one patent per line.
  • Suitable for downstream NLP processing, including language model pretraining.

Github

Cleaning script and notebook example available at https://github.com/mhurhangee/openpatent

Downloads last month
74