sv_core_news_sm
| Property | Value |
|---|---|
| Version | 3.7.0 |
| License | CC BY-SA 4.0 |
| Author | Explosion |
| Source Data | UD Swedish Talbanken v2.8, Stockholm-Umeå Corpus (SUC) v3.0 |
What is sv_core_news_sm?
sv_core_news_sm is a lightweight Swedish language model developed by Explosion AI for the spaCy framework. It's specifically optimized for CPU performance, making it accessible for users without specialized hardware requirements. The model provides a comprehensive suite of natural language processing capabilities for Swedish text analysis.
Implementation Details
The model implements a pipeline architecture with multiple components: tok2vec (for tokenization and vector representations), tagger, morphologizer, parser, lemmatizer, attribute ruler, and named entity recognition (NER). It achieves impressive accuracy scores across various metrics, including 99.99% token accuracy and 94.90% lemmatization accuracy.
- Token Accuracy: 99.99% with 99.95% F-score
- POS Tagging Accuracy: 95.11%
- Morphological Analysis Accuracy: 94.07%
- Dependency Parsing: 81.86% UAS, 76.73% LAS
- Named Entity Recognition: 74.90% F-score
Core Capabilities
- Part-of-speech tagging with extensive tag set
- Morphological analysis supporting 381 different labels
- Dependency parsing with 35 dependency relations
- Named entity recognition for 8 entity types (EVN, LOC, MSR, OBJ, ORG, PRS, TME, WRK)
- Sentence segmentation with 93.68% F-score
Frequently Asked Questions
Q: What makes this model unique?
The model's strength lies in its optimization for CPU usage while maintaining high accuracy across multiple NLP tasks. It's particularly notable for its comprehensive coverage of Swedish language features and high token accuracy.
Q: What are the recommended use cases?
This model is ideal for production environments requiring Swedish language processing, including text classification, information extraction, and linguistic analysis. It's particularly suitable for applications where CPU-based processing is preferred over GPU acceleration.
