Description
This is a Multiclass classification model which identifies law area labels(civil_law, penal_law, public_law, social_law) in French-based Court Cases.
Predicted Entities
civil_law, penal_law, public_law, social_law
How to use
document_assembler = nlp.DocumentAssembler() \
.setInputCol("text") \
.setOutputCol("document")
embeddings = nlp.BertSentenceEmbeddings.pretrained("sent_bert_multi_cased", "xx")\
.setInputCols(["document"]) \
.setOutputCol("sentence_embeddings")
docClassifier = legal.ClassifierDLModel.pretrained("legclf_law_area_prediction_french", "fr", "legal/models")\
.setInputCols(["sentence_embeddings"])\
.setOutputCol("category")
nlpPipeline = nlp.Pipeline(stages=[
document_assembler,
embeddings,
docClassifier
])
df = spark.createDataFrame([["par ces motifs, le Juge unique prononce : 1. Le recours est irrecevable. 2. Il n'est pas perçu de frais judiciaires. 3. Le présent arrêt est communiqué aux parties, au Tribunal administratif fédéral et à l'Office fédéral des assurances sociales. Lucerne, le 2 juin 2016 Au nom de la IIe Cour de droit social du Tribunal fédéral suisse Le Juge unique : Meyer Le Greffier : Cretton"]]).toDF("text")
model = nlpPipeline.fit(df)
result = model.transform(df)
result.select("text", "category.result").show(truncate=100)
Results
+----------------------------------------------------------------------------------------------------+------------+
| text| result|
+----------------------------------------------------------------------------------------------------+------------+
|par ces motifs, le Juge unique prononce : 1. Le recours est irrecevable. 2. Il n'est pas perçu de...|[social_law]|
+----------------------------------------------------------------------------------------------------+------------+
Model Information
| Model Name: | legclf_law_area_prediction_french |
| Compatibility: | Legal NLP 1.0.0+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [sentence_embeddings] |
| Output Labels: | [class] |
| Language: | fr |
| Size: | 22.3 MB |
References
Train dataset available here
Benchmarking
label precision recall f1-score support
civil_law 0.93 0.91 0.92 613
penal_law 0.94 0.96 0.95 579
public_law 0.92 0.91 0.92 605
social_law 0.97 0.98 0.97 478
accuracy - - 0.94 2275
macro-avg 0.94 0.94 0.94 2275
weighted-avg 0.94 0.94 0.94 2275