Hugging Face's logo

spacy
/
zh_core_web_lg

Token Classification
spaCy
Chinese
Eval Results
Model card Files Community
zh_core_web_lg / README.md
osanseviero's picture
osanseviero
Update spaCy pipeline
7dfbadd
preview code
|
raw
history blame
3.69 kB
metadata
tags:
  - spacy
  - token-classification
language:
  - zh
license: MIT
model-index:
  - name: zh_core_web_lg
    results:
      - tasks:
          name: NER
          type: token-classification
          metrics:
            - name: Precision
              type: precision
              value: 0.7358998362
            - name: Recall
              type: recall
              value: 0.6910989011
            - name: F Score
              type: f_score
              value: 0.7127961011
      - tasks:
          name: POS
          type: token-classification
          metrics:
            - name: Accuracy
              type: accuracy
              value: 0.9037457747
      - tasks:
          name: SENTER
          type: token-classification
          metrics:
            - name: Precision
              type: precision
              value: 0.7896445968
            - name: Recall
              type: recall
              value: 0.7286499084
            - name: F Score
              type: f_score
              value: 0.7579220779
      - tasks:
          name: UNLABELED_DEPENDENCIES
          type: token-classification
          metrics:
            - name: Accuracy
              type: accuracy
              value: 0.7069146954
      - tasks:
          name: LABELED_DEPENDENCIES
          type: token-classification
          metrics:
            - name: Accuracy
              type: accuracy
              value: 0.7069146954

Details: https://spacy.io/models/zh#zh_core_web_lg

Chinese pipeline optimized for CPU. Components: tok2vec, tagger, parser, senter, ner, attribute_ruler.

Feature Description
Name zh_core_web_lg
Version 3.1.0
spaCy >=3.1.0,<3.2.0
Default Pipeline tok2vec, tagger, parser, attribute_ruler, ner
Components tok2vec, tagger, parser, senter, attribute_ruler, ner
Vectors 500000 keys, 500000 unique vectors (300 dimensions)
Sources OntoNotes 5 (Ralph Weischedel, Martha Palmer, Mitchell Marcus, Eduard Hovy, Sameer Pradhan, Lance Ramshaw, Nianwen Xue, Ann Taylor, Jeff Kaufman, Michelle Franchini, Mohammed El-Bachouti, Robert Belvin, Ann Houston)
CoreNLP Universal Dependencies Converter (Stanford NLP Group)
Explosion fastText Vectors (cbow, OSCAR Common Crawl + Wikipedia) (Explosion)
License MIT
Author Explosion

Label Scheme

View label scheme (101 labels for 4 components)
Component Labels
tagger AD, AS, BA, CC, CD, CS, DEC, DEG, DER, DEV, DT, ETC, FW, IJ, INF, JJ, LB, LC, M, MSP, NN, NR, NT, OD, ON, P, PN, PU, SB, SP, URL, VA, VC, VE, VV, X
parser ROOT, acl, advcl:loc, advmod, advmod:dvp, advmod:loc, advmod:rcomp, amod, amod:ordmod, appos, aux:asp, aux:ba, aux:modal, aux:prtmod, auxpass, case, cc, ccomp, compound:nn, compound:vc, conj, cop, dep, det, discourse, dobj, etc, mark, mark:clf, name, neg, nmod, nmod:assmod, nmod:poss, nmod:prep, nmod:range, nmod:tmod, nmod:topic, nsubj, nsubj:xsubj, nsubjpass, nummod, parataxis:prnmod, punct, xcomp
senter I, S
ner CARDINAL, DATE, EVENT, FAC, GPE, LANGUAGE, LAW, LOC, MONEY, NORP, ORDINAL, ORG, PERCENT, PERSON, PRODUCT, QUANTITY, TIME, WORK_OF_ART

Accuracy

Type Score
TOKEN_ACC 97.88
TAG_ACC 90.37
DEP_UAS 70.69
DEP_LAS 65.55
ENTS_P 73.59
ENTS_R 69.11
ENTS_F 71.28
SENTS_P 78.96
SENTS_R 72.86
SENTS_F 75.79