Description
This is a md version of German Legal Judgement Text Classifier written in German legal writing style “Urteilsstil” (judgement style), which will retrieve if a text is either conclusion, definition, other or subsumption .
Predicted Entities
conclusion, definition, subsumption, other
How to use
document_assembler = nlp.DocumentAssembler() \
.setInputCol("text") \
.setOutputCol("document")
tokenizer = nlp.Tokenizer() \
.setInputCols(["document"]) \
.setOutputCol("token")
classifierdl = legal.BertForSequenceClassification.pretrained("legclf_judgement_medium","de", "legal/models")\
.setInputCols(["document", "token"])\
.setOutputCol("label")
bert_clf_pipeline = nlp.Pipeline(stages=[document_assembler,
tokenizer,
classifierdl])
text = ["Insoweit ergibt sich tatsächlich im Ergebnis ein Verzicht der Arbeitnehmer in Höhe der RoSi-Zulage ."]
empty_df = spark.createDataFrame([[""]]).toDF("text")
model = bert_clf_pipeline.fit(empty_df)
res = model.transform(spark.createDataFrame([text]).toDF("text"))
Results
+----------------------------------------------------------------------------------------------------+-------------+
|text |result |
+----------------------------------------------------------------------------------------------------+-------------+
|Insoweit ergibt sich tatsächlich im Ergebnis ein Verzicht der Arbeitnehmer in Höhe der RoSi-Zulage .|[subsumption]|
+----------------------------------------------------------------------------------------------------+-------------+
Model Information
| Model Name: | legclf_judgement_medium |
| Compatibility: | Legal NLP 1.0.0+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [document, token] |
| Output Labels: | [class] |
| Language: | de |
| Size: | 409.8 MB |
| Case sensitive: | true |
| Max sentence length: | 128 |
References
An in-house augmented version of this dataset
Benchmarking
label precision recall f1-score support
conclusion 0.74 0.79 0.76 189
definition 0.91 0.88 0.90 160
other 0.85 0.82 0.83 163
subsumption 0.71 0.70 0.70 159
accuracy - - 0.80 671
macro-avg 0.80 0.80 0.80 671
weighted-avg 0.80 0.80 0.80 671