Edit on GitHub
  • John Snow Labs
  • Oct 18, 2022

Legal Relation Extraction (Confidentiality, sm, Bidirectional)

  • legal
  • en
  • re
  • licensed
  • confidentiality

Description

IMPORTANT: Don’t run this model on the whole legal agreement. Instead:

  • Split by paragraphs. You can use notebook 1 in Finance or Legal as inspiration;
  • Use the legclf_cuad_confidentiality_clause Text Classifier to select only these paragraphs;

This is a Legal Relation Extraction Model to identify the Subject (who), Action (web), Object(the indemnification) and Indirect Object (to whom) from confidentiality clauses.

This model is a sm model without meaningful directions in the relations (the model was not trained to understand if the direction of the relation is from left to right or right to left).

There are bigger models in Models Hub trained also with directed relationships.

Predicted Entities

is_confidentiality_indobject, is_confidentiality_object, is_confidentiality_subject

Copy S3 URI

How to use

documentAssembler = nlp.DocumentAssembler()\
  .setInputCol("text")\
  .setOutputCol("document")
  
sentencizer = nlp.SentenceDetectorDLModel\
        .pretrained("sentence_detector_dl", "en") \
        .setInputCols(["document"])\
        .setOutputCol("sentence")
        
tokenizer = nlp.Tokenizer()\
        .setInputCols("sentence")\
        .setOutputCol("token")
        
pos_tagger = nlp.PerceptronModel()\
    .pretrained() \
    .setInputCols(["sentence", "token"])\
    .setOutputCol("pos_tags")
    
dependency_parser = nlp.DependencyParserModel() \
    .pretrained("dependency_conllu", "en") \
    .setInputCols(["sentence", "pos_tags", "token"]) \
    .setOutputCol("dependencies")
    
embeddings = nlp.RoBertaEmbeddings.pretrained("roberta_embeddings_legal_roberta_base","en") \
    .setInputCols(["sentence", "token"]) \
    .setOutputCol("embeddings")
    
ner_model = legal.NerModel.pretrained('legner_confidentiality', 'en', 'legal/models') \
        .setInputCols(["sentence", "token", "embeddings"]) \
        .setOutputCol("ner")
        
ner_converter = nlp.NerConverter() \
        .setInputCols(["sentence","token","ner"]) \
        .setOutputCol("ner_chunk")
        
re_filter = legal.RENerChunksFilter()\
    .setInputCols(["ner_chunk", "dependencies"])\
    .setOutputCol("re_ner_chunks")\
    .setMaxSyntacticDistance(10)\
    .setRelationPairs(['CONFIDENTIALITY_ACTION-CONFIDENTIALITY_SUBJECT','CONFIDENTIALITY_ACTION-CONFIDENTIALITY','CONFIDENTIALITY_SUBJECT-CONFIDENTIALITY_INDIRECT_OBJECT'])
    
reDL = legal.RelationExtractionDLModel.pretrained("legre_confidentiality", "en", "legal/models") \
    .setPredictionThreshold(0.5) \
    .setInputCols(["re_ner_chunks", "sentence"]) \
    .setOutputCol("relations")
    
pipeline = nlp.Pipeline(stages=[documentAssembler,sentencizer, tokenizer,pos_tagger,dependency_parser, embeddings, ner_model, ner_converter,re_filter, reDL])
text = "Each party will promptly return to the other upon request any Confidential Information of the other party then in its possession or under its control."
data = spark.createDataFrame([[text]]).toDF("text")
model = pipeline.fit(data)
res = model.transform(data)

Results


+----------------------------+-----------------------+-------------+-----------+--------------------+-------------------------------+-------------+-----------+------------------------+----------+
|relation                    |entity1                |entity1_begin|entity1_end|chunk1              |entity2                        |entity2_begin|entity2_end|chunk2                  |confidence|
+----------------------------+-----------------------+-------------+-----------+--------------------+-------------------------------+-------------+-----------+------------------------+----------+
|is_confidentiality_subject  |CONFIDENTIALITY_SUBJECT|0            |9          |Each party          |CONFIDENTIALITY_ACTION         |11           |30         |will promptly return    |0.9745122 |
|is_confidentiality_indobject|CONFIDENTIALITY_SUBJECT|0            |9          |Each party          |CONFIDENTIALITY_INDIRECT_OBJECT|39           |43         |other                   |0.89561754|
|is_confidentiality_object   |CONFIDENTIALITY_ACTION |11           |30         |will promptly return|CONFIDENTIALITY                |62           |85         |Confidential Information|0.9981041 |
+----------------------------+-----------------------+-------------+-----------+--------------------+-------------------------------+-------------+-----------+------------------------+----------+

Model Information

Model Name: legre_confidentiality
Compatibility: Legal NLP 1.0.0+
License: Licensed
Edition: Official
Language: en
Size: 409.9 MB

References

In-house annotated examples from CUAD legal dataset

Benchmarking

                       label   Recall Precision       F1   Support
is_confidentiality_indobject    1.000     1.000    1.000        28
   is_confidentiality_object    1.000     0.981    0.990        51
  is_confidentiality_subject    0.970     1.000    0.985        33
                         Avg    0.990     0.994    0.992        -
                Weighted-Avg    0.991     0.991    0.991        -