Edit on GitHub
  • John Snow Labs
  • Nov 21, 2022

Extract Effective, Renewal, Termination dates (Small)

  • renewal
  • effective
  • termination
  • date
  • en
  • licensed

Description

This models extracts if a date is a Effective Date, a Renewal Date or a Termination Date, and also extracts the keywords surrounding that may be pointing about what kind of date it is. Please take into account that the keyword was not used to learn the date, all entities were training separately. But you can use the keywords to double check the date type is correct.

Predicted Entities

EFFDATE, EFFDATE_KEYWORD, RENDATE, RENDATE_KEYWORD, TERMINDATE, TERMINDATE_KEYWORD

Copy S3 URI

How to use

documentAssembler = nlp.DocumentAssembler()\
        .setInputCol("text")\
        .setOutputCol("document")
        
sentenceDetector = nlp.SentenceDetectorDLModel.pretrained("sentence_detector_dl","xx")\
        .setInputCols(["document"])\
        .setOutputCol("sentence")

tokenizer = nlp.Tokenizer()\
        .setInputCols(["sentence"])\
        .setOutputCol("token")

embeddings = nlp.RoBertaEmbeddings.pretrained("roberta_embeddings_legal_roberta_base","en") \
    .setInputCols(["sentence", "token"]) \
    .setOutputCol("embeddings")

ner_model = legal.NerModel.pretrained('legner_dates_sm', 'en', 'legal/models')\
        .setInputCols(["sentence", "token", "embeddings"])\
        .setOutputCol("ner")

ner_converter = nlp.NerConverter()\
        .setInputCols(["sentence","token","ner"])\
        .setOutputCol("ner_chunk")

nlpPipeline = nlp.Pipeline(stages=[
        documentAssembler,
        sentenceDetector,
        tokenizer,
        embeddings,
        ner_model,
        ner_converter])

empty_data = spark.createDataFrame([[""]]).toDF("text")

model = nlpPipeline.fit(empty_data)

text = ["""RENEWAL DATE. The date on which this Agreement shall renew, July 1st, pursuant to the terms and conditions contained herein."""]

res = model.transform(spark.createDataFrame([text]).toDF("text"))

Results

+----------+-----------------+
|     token|        ner_label|
+----------+-----------------+
|   RENEWAL|B-RENDATE_KEYWORD|
|      DATE|I-RENDATE_KEYWORD|
|         .|                O|
|       The|                O|
|      date|                O|
|        on|                O|
|     which|                O|
|      this|                O|
| Agreement|                O|
|     shall|                O|
|     renew|                O|
|         ,|                O|
|      July|        B-RENDATE|
|       1st|        I-RENDATE|
|         ,|                O|
|  pursuant|                O|
|        to|                O|
|       the|                O|
|     terms|                O|
|       and|                O|
|conditions|                O|
| contained|                O|
|    herein|                O|
|         .|                O|
+----------+-----------------+

Model Information

Model Name: legner_dates_sm
Compatibility: Legal NLP 1.0.0+
License: Licensed
Edition: Official
Input Labels: [sentence, token, embeddings]
Output Labels: [ner]
Language: en
Size: 16.1 MB

References

In-house dataset.

Benchmarking

label	 tp	 fp	 fn	 prec	 rec	 f1
I-RENDATE	 6	 2	 5	 0.75	 0.54545456	 0.631579
B-EFFDATE_KEYWORD	 5	 0	 0	 1.0	 1.0	 1.0
B-EFFDATE	 5	 0	 0	 1.0	 1.0	 1.0
B-TERMINDATE	 3	 1	 0	 0.75	 1.0	 0.85714287
I-TERMINDATE	 9	 4	 0	 0.6923077	 1.0	 0.8181818
I-EFFDATE	 15	 0	 0	 1.0	 1.0	 1.0
I-RENDATE_KEYWORD	 4	 0	 0	 1.0	 1.0	 1.0
I-EFFDATE_KEYWORD	 5	 0	 0	 1.0	 1.0	 1.0
I-TERMINDATE_KEYWORD	 5	 0	 0	 1.0	 1.0	 1.0
B-RENDATE	 2	 1	 2	 0.6666667	 0.5	 0.57142854
B-TERMINDATE_KEYWORD	 5	 0	 0	 1.0	 1.0	 1.0
B-RENDATE_KEYWORD	 3	 0	 1	 1.0	 0.75	 0.85714287
Macro-average 67 8 8 0.90491456 0.8996212 0.9022601
Micro-average 67 8 8 0.8933333 0.8933333 0.89333326