Description
This is the Multi-Label Text Classification model that can be used to identify 30+ classes to facilitate analysis, discovery, and comparison of legal texts in English related to Law Stack Exchange.
Predicted Entities
business, california, canada, civil-law, constitutional-law, consumer-protection, contract-law, copyright, corporate-law, criminal-law, employment, england-and-wales, european-union, fraud, gdpr, germany, intellectual-property, international, internet, landlord, legal-terms, liability, licensing, police, privacy, property, real-estate, rental-property, software, tax-law, terms-of-service, trademark, traffic, united-kingdom, united-states, us-constitution
How to use
document_assembler = nlp.DocumentAssembler() \
.setInputCol("text") \
.setOutputCol("document") \
.setCleanupMode("shrink")
embeddings = nlp.InstructorEmbeddings.pretrained("instructor_large", "en") \
.setInstruction("Represent for multilabel classification:") \
.setInputCols(["document"]) \
.setOutputCol("sentence_embeddings")
classifierdl = nlp.MultiClassifierDLModel.pretrained('legmulticlf_law_stack_exchange', 'en', 'legal/models') \
.setInputCols(["sentence_embeddings"])\
.setOutputCol("class")
clf_pipeline = nlp.Pipeline(stages=[document_assembler,
embeddings,
classifierdl])
df = spark.createDataFrame([["I've seen this label on coke cans at one point, and I was wondering: is this legally enforceable?If it's not, is it possible for a retailer in any way to disallow the resale of an item purchased?Something like, I don't know, maybe a license you have to agree to in order to be allowed to purchase said item?This comes in the larger context of these new tech releases (GPUs, consoles) and how the producers/retailers could legally prevent scalpers."]]).toDF("text")
model = clf_pipeline.fit(df)
result = model.transform(df)
result.select("text", "class.result").show(truncate=False)
Results
+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+------------------------+
|text |result |
+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+------------------------+
|I've seen this label on coke cans at one point, and I was wondering: is this legally enforceable?If it's not, is it possible for a retailer in any way to disallow the resale of an item purchased?Something like, I don't know, maybe a license you have to agree to in order to be allowed to purchase said item?This comes in the larger context of these new tech releases (GPUs, consoles) and how the producers/retailers could legally prevent scalpers.|[contract-law, business]|
+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+------------------------+
Model Information
| Model Name: | legmulticlf_law_stack_exchange |
| Compatibility: | Legal NLP 1.0.0+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [sentence_embeddings] |
| Output Labels: | [class] |
| Language: | en |
| Size: | 14.0 MB |
References
Train dataset available here
Benchmarking
label precision recall f1-score support
business 1.00 0.82 0.90 51
california 1.00 1.00 1.00 74
canada 1.00 0.97 0.99 74
civil-law 0.98 0.94 0.96 52
constitutional-law 1.00 1.00 1.00 53
consumer-protection 1.00 0.86 0.92 28
contract-law 0.98 0.89 0.94 199
copyright 0.94 0.97 0.95 246
corporate-law 1.00 0.88 0.94 26
criminal-law 0.92 0.95 0.94 198
employment 1.00 1.00 1.00 83
england-and-wales 0.96 0.99 0.97 93
european-union 0.91 0.93 0.92 72
fraud 1.00 0.97 0.99 39
gdpr 1.00 1.00 1.00 87
germany 1.00 1.00 1.00 44
intellectual-property 0.82 0.93 0.87 89
international 0.91 1.00 0.95 71
internet 0.96 0.96 0.96 83
landlord 0.97 0.94 0.96 36
legal-terms 1.00 1.00 1.00 56
liability 1.00 0.93 0.96 42
licensing 0.99 0.92 0.95 73
police 0.98 1.00 0.99 51
privacy 1.00 0.83 0.90 69
property 0.89 0.97 0.93 32
real-estate 1.00 0.97 0.98 32
rental-property 1.00 1.00 1.00 41
software 0.92 0.83 0.87 69
tax-law 1.00 1.00 1.00 38
terms-of-service 0.96 0.96 0.96 25
trademark 1.00 1.00 1.00 45
traffic 1.00 1.00 1.00 30
united-kingdom 0.97 0.94 0.96 161
united-states 0.77 0.87 0.81 644
us-constitution 1.00 0.95 0.98 43
micro avg 0.92 0.93 0.93 3149
macro avg 0.97 0.95 0.96 3149
weighted avg 0.92 0.93 0.93 3149
samples avg 0.92 0.94 0.92 3149