atheer2104/swedish-patent-cpc-subclass-new

Patent CPC Multi-Label Classification Dataset

This dataset contains historical Swedish patent documents (1885-1972) classified according to the Cooperative Patent Classification (CPC) system. Each document can have multiple labels, making this a multi-label classification task with significant implications for patent retrieval and prior art search.

Dataset Description

The dataset includes patent claims text extracted from digitally recreated versions of historical Swedish patents, generated using Optical Character Recognition (OCR) from original paper documents. The text quality varies due to OCR limitations, but all CPC labels were manually assigned by patent engineers at PRV (Swedish Patent and Registration Office), ensuring high reliability for machine learning applications.

Background and Motivation

Historical Swedish patents from 1885-1972 were originally classified using the German Patent Classification (DPK) system. When Sweden adopted the International Patent Classification (IPC) system in 1972, approximately 20% of documents were manually reclassified. Later, when the Cooperative Patent Classification (CPC) system was introduced, most IPC labels were automatically converted to CPC, with manual classification when automatic conversion wasn't possible.

This dataset addresses the challenge of making historical patents more accessible by providing CPC classifications that enable modern patent search and retrieval systems to effectively index and retrieve older Swedish patents.

Subclass Classification

• Examples: 35,831 documents (after filtering)
• Train examples: 31776
• Test examples: 4041
• Unique labels: 497 CPC subclass labels
• Single-label documents: 23,416 (65.3%)
• Multi-label documents: 12,415 (34.7%)
• Average labels per document: 1.49
• Max labels per document: 12
• Average documents per label: 107.82
• Task: Multi-label classification of patent claims text to CPC subclass level

Data Characteristics

The dataset exhibits a long-tailed label distribution typical of patent classification tasks:

• Most documents (65.3%) have only one CPC subclass label
• Label frequency follows a power-law distribution with few highly frequent labels and many rare labels
• The hierarchical nature of CPC classification creates natural class imbalances

Label Distribution Breakdown:

• 1 label: 23,424 documents
• 2 labels: 8,689 documents
• 3 labels: 2,650 documents
• 4+ labels: 1,268 documents (decreasing exponentially)

Data Fields

• text: Patent claims text (Swedish) extracted from OCR-processed historical documents
• labels: List of CPC subclass classification codes (multi-label format)
• dokID: Unique document identifier
• dokNr: Patent document number
• dokKod: Document code classification
• full_cpc_codes: Complete list of all CPC codes assigned to the document

Data Processing

Text Extraction

Patent claims were automatically extracted using spaCy with the "sv_core_news_sm" pipeline for Swedish text processing. Claims were identified by:

• Start markers: "patentanspråk" or "patentkrav"
• End markers: "härtill ritning" or "kongl boktryckeriet"

Label Preprocessing

• CPC labels related to index codes and section Y were removed (providing only supplementary information)
• Labels with fewer than 10 associated documents were filtered out to ensure adequate training data
• Only CPC subclass level labels are included in this version

Quality Considerations

• Original documents: 269,052 total patent documents
• Used documents: 35,831 (after filtering for quality and label requirements)
• Text quality varies due to OCR processing of historical documents
• All CPC labels manually verified by patent engineering professionals

Usage

from datasets import load_dataset

# Load dataset
dataset = load_dataset("atheer2104/swedish-patent-cpc-subclass-new")

# Access train and test splits
train_data = dataset['train']
test_data = dataset['test']

# Example of accessing multi-label data
example = train_data[0]
text = example['text']
labels = example['label']  # This is a list of labels

print(f"Text: {text}")
print(f"Labels: {labels}")

Multi-Label Classification Challenges

This dataset presents several characteristics that make it challenging for machine learning models:

1. Class Imbalance: Highly skewed label distribution with some labels having hundreds of examples while others have only the minimum of 10
2. Multi-label Complexity: 34.7% of documents have multiple labels requiring models to predict label combinations
3. Historical Text: OCR artifacts and historical Swedish language variations may impact text quality
4. Domain Specificity: Technical patent language requiring domain knowledge for optimal performance

Evaluation Metrics

For multi-label classification tasks, consider using these metrics:

• Precision@k (P@k): Precision of top-k predictions
• Recall@k (R@k): Recall of top-k predictions
• F1@k: Harmonic mean of P@k and R@k
• nDCG@k: Normalized Discounted Cumulative Gain at rank k
• Micro-F1: Overall F1 score across all labels
• Macro-F1: Average F1 score across all labels

Recommended evaluation at k = 1, 3, 5 to capture different aspects of model performance.

Dataset Splits

The dataset should be split using stratified sampling to maintain similar label distributions across train/validation/test sets. Recommended splits:

• Train: 90%
• Test: 10%
• Use nested validation (90/10 split of training data) for hyperparameter tuning

Related Research

This dataset was created as part of research investigating whether incorporating legacy classification information (DPK) improves CPC classification performance. The research demonstrated that combining textual content with legacy classification data significantly improves multi-label patent classification accuracy.

Limitations

• Text quality varies due to OCR processing of historical documents
• Limited to Swedish patents from 1885-1972
• Only documents with both original DPK and manually-assigned CPC labels were used
• Minimum threshold of 10 documents per label may exclude rare but important patent categories
• OCR errors may impact model training and evaluation

Citation

@mastersthesis{Salim2025PatentClassification,
   author = {Salim, Atheer},
   institution = {KTH, School of Electrical Engineering and Computer Science (EECS)},
   pages = {70},
   school = {KTH, School of Electrical Engineering and Computer Science (EECS)},
   title = {Machine Learning for Classifying Historical Swedish Patents: A Comparison of Textual and Combined Data Approaches},
   series = {TRITA-EECS-EX},
   number = {2025:571},
   keywords = {Multi-label Text Classification, Machine Learning, Patent Classification, Deep Learning, Natural Language Processing},
   abstract = {Patents are essential for protecting intellectual property and advancing innovation, but the accessibility of historical patents is often limited by outdated classification systems. This thesis investigates whether incorporating legacy classification data alongside patent claims improves the performance of machine learning models in classifying historical Swedish patents into the modern CPC system.},
   year = {2025},
   url = "https://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-368254"
}