Description
This models should be run on each sentence of the NDA clauses, and will retrieve a series of 1..N labels for each of them. The possible clause types detected my this model in NDA / MNDA aggrements are:
- Parties to the Agreement - Names of the Parties Clause
- Identification of What Information Is Confidential - Definition of Confidential Information Clause
- Use of Confidential Information: Permitted Use Clause and Obligations of the Recipient
- Time Frame of the Agreement - Termination Clause
- Return of Confidential Information Clause
- Remedies for Breaches of Agreement - Remedies Clause
- Non-Solicitation Clause
- Dispute Resolution Clause
- Exceptions Clause
- Non-competition clause
Predicted Entities
APPLIC_LAW, ASSIGNMENT, DEF_OF_CONF_INFO, DISPUTE_RESOL, EXCEPTIONS, NAMES_OF_PARTIES, NON_COMP, NON_SOLIC, PREAMBLE, REMEDIES, REQ_DISCL, RETURN_OF_CONF_INFO, TERMINATION, USE_OF_CONF_INFO
How to use
document_assembler = (
nlp.DocumentAssembler().setInputCol("text").setOutputCol("document")
)
sentence_splitter = (
nlp.SentenceDetector()
.setInputCols(["document"])
.setOutputCol("sentence")
.setCustomBounds(["\n"])
)
embeddings = (
nlp.UniversalSentenceEncoder.pretrained()
.setInputCols("sentence")
.setOutputCol("sentence_embeddings")
)
classsifierdl_pred = nlp.MultiClassifierDLModel.pretrained('legmulticlf_mnda_sections', 'en', 'legal/models')\
.setInputCols(["sentence_embeddings"])\
.setOutputCol("class")
clf_pipeline = nlp.Pipeline(stages=[document_assembler, sentence_splitter, embeddings, classsifierdl_pred])
df = spark.createDataFrame([["Governing Law.\nThis Agreement shall be govern..."]]).toDF("text")
res = clf_pipeline.fit(df).transform(df)
res.select('text', 'class.result').show()
res.select('class.result')
Results
[APPLIC_LAW] Governing Law.\nThis Agreement shall be govern...
Model Information
| Model Name: | legmulticlf_mnda_sections |
| Compatibility: | Legal NLP 1.0.0+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [sentence_embeddings] |
| Output Labels: | [class] |
| Language: | en |
| Size: | 12.9 MB |
References
In-house MNDA
Benchmarking
label precision recall f1-score support
APPLIC_LAW 0.93 0.96 0.95 28
ASSIGNMENT 0.95 0.91 0.93 22
DEF_OF_CONF_INFO 0.92 0.80 0.86 30
DISPUTE_RESOL 0.76 0.89 0.82 28
EXCEPTIONS 0.77 0.91 0.83 11
NAMES_OF_PARTIES 0.94 0.88 0.91 33
NON_COMP 1.00 0.91 0.95 23
NON_SOLIC 0.88 0.94 0.91 16
PREAMBLE 0.79 0.85 0.81 26
REMEDIES 0.91 0.91 0.91 32
REQ_DISCL 0.92 0.92 0.92 13
RETURN_OF_CONF_INFO 1.00 0.96 0.98 24
TERMINATION 1.00 0.77 0.87 13
USE_OF_CONF_INFO 0.85 0.88 0.86 32
micro-avg 0.89 0.89 0.89 331
macro-avg 0.90 0.89 0.89 331
weighted-avg 0.90 0.89 0.89 331
samples-avg 0.87 0.89 0.88 331