Description
This is a small capital call NER, trained to extract contact and financial information from Capital Call Notices. These are the entities retrieved by the model:
Financial information:
FUND: Name of the Fund called
ORG: Organization asking the Fund for the Capital
AMOUNT: Amount called by ORG to FUND
DUE_DATE: Due date of the call
ACCOUNT_NAME: Organization's Bank Account Name
ACCOUNT_NUMBER: Organization's Bank Account Number
ABA: Routing Number (ABA)
BANK_ADDRESS: Contact address of the Bank
Contact information:
PHONE: Contact Phone
PERSON: Contact Person
BANK_CONTACT: Person to contact in Bank
EMAIL: Contact Email
Other additional information, not directly involved in the call:
OTHER_PERSON: Other people detected (People signing the call, people to whom is addressed the Notice, etc)
OTHER_PERCENTAGE: Percentages mentiones
OTHER_DATE: Other dates mentioned, not Due Date
OTHER_AMOUNT: Other amounts mentioned
OTHER_ADDRESS: Other addresses mentiones
OTHER_ORG: Other ORG mentiones
Predicted Entities
FUND, ORG, AMOUNT, DUE_DATE, ACCOUNT_NAME, ACCOUNT_NUMBER, BANK_ADDRESS, PHONE, PERSON, BANK_CONTACT, EMAIL, OTHER_PERSON, OTHER_PERCENTAGE, OTHER_DATE, OTHER_AMOUNT, OTHER_ADDRESS, OTHER_ORG, ABA
How to use
from pyspark.sql import functions as F
documentAssembler = nlp.DocumentAssembler() \
.setInputCol("text") \
.setOutputCol("document")
sentence = nlp.SentenceDetector() \
.setInputCols(["document"]) \
.setOutputCol("sentence")
tokenizer = nlp.Tokenizer() \
.setInputCols("sentence") \
.setOutputCol("token")
embeddings = nlp.BertEmbeddings.pretrained("bert_embeddings_sec_bert_base","en") \
.setInputCols(["sentence", "token"]) \
.setOutputCol("embeddings")\
.setMaxSentenceLength(512)
ner = finance.NerModel.pretrained('finner_capital_calls', 'en', 'finance/models')\
.setInputCols(["sentence", "token", "embeddings"]) \
.setOutputCol("ner")
converter = finance.NerConverterInternal()\
.setInputCols(["sentence", "token", "ner"]) \
.setOutputCol("ner_chunk")\
pipeline = nlp.Pipeline(stages=[documentAssembler,
sentence,
tokenizer,
embeddings,
ner,
converter
])
df = spark.createDataFrame([[""]]).toDF("text")
model = pipeline.fit(df)
lp = nlp.LightPipeline(model)
text = """Dear Charlotte R. Davis,
We hope this message finds you well. This is to inform you that a capital call for Upfront Ventures has been initiated. The amount requested is 7000 EUR and is due on 01.01.2024.
Kindly wire transfer the funds to the following account:
Account Green Planet Solutions LLC
Account Number 1234567-1XX
Routing Number 51903761
Bank First Republic Bank
If you require any further information, please do not hesitate to reach out to us at 3055 550818 or coxeric@example.com.
Thank you for your prompt attention to this matter.
Best regards,
James Wilson"""
result = model.transform(spark.createDataFrame([[text]]).toDF("text"))
from pyspark.sql import functions as F
result.select(F.explode(F.arrays_zip(result.ner_chunk.result, result.ner_chunk.metadata)).alias("cols")) \
.select(F.expr("cols['0']").alias("chunk"),
F.expr("cols['1']['entity']").alias("ner_label"),
F.expr("cols['1']['confidence']").alias("confidence")).show(truncate=False)
Results
+--------------------------+--------------+----------+
|chunk |ner_label |confidence|
+--------------------------+--------------+----------+
|Charlotte R. Davis |OTHER_PERSON |0.971875 |
|Upfront Ventures |FUND |1.0 |
|7000 EUR |AMOUNT |1.0 |
|01.01.2024 |DUE_DATE |1.0 |
|Green Planet Solutions LLC|ACCOUNT_NAME |0.999875 |
|1234567-1XX |ACCOUNT_NUMBER|1.0 |
|51903761 |ABA |1.0 |
|First Republic Bank |BANK_NAME |0.9999333 |
|3055 550818 |PHONE |1.0 |
|coxeric@example.com |EMAIL |1.0 |
|James Wilson |OTHER_PERSON |1.0 |
+--------------------------+--------------+----------+
Model Information
| Model Name: | finner_capital_calls |
| Compatibility: | Finance NLP 1.0.0+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [sentence, token, embeddings] |
| Output Labels: | [ner] |
| Language: | en |
| Size: | 16.3 MB |
References
In-house capital call notices
Benchmarking
Total test loss: 0.3542 Avg test loss: 0.0295
label tp fp fn prec rec f1
B-PERSON 47 0 0 1.0 1.0 1.0
I-OTHER_PERSON 214 0 0 1.0 1.0 1.0
I-AMOUNT 127 0 0 1.0 1.0 1.0
I-OTHER_PERCENTAGE 37 0 0 1.0 1.0 1.0
B-OTHER_DATE 25 0 0 1.0 1.0 1.0
I-BANK_ADDRESS 121 0 0 1.0 1.0 1.0
B-AMOUNT 170 0 0 1.0 1.0 1.0
B-OTHER_AMOUNT 409 0 0 1.0 1.0 1.0
I-ORG 311 18 0 0.9452888 1.0 0.971875
B-PHONE 79 0 0 1.0 1.0 1.0
I-DUE_DATE 153 0 0 1.0 1.0 1.0
B-FUND 124 0 0 1.0 1.0 1.0
B-ABA 97 0 0 1.0 1.0 1.0
I-ACCOUNT_NAME 223 0 0 1.0 1.0 1.0
I-OTHER_DATE 25 0 0 1.0 1.0 1.0
I-PHONE 119 0 0 1.0 1.0 1.0
B-BANK_ADDRESS 39 0 0 1.0 1.0 1.0
B-OTHER_ORG 139 0 6 1.0 0.95862067 0.97887325
I-OTHER_AMOUNT 307 0 0 1.0 1.0 1.0
I-FUND 131 0 0 1.0 1.0 1.0
I-BANK_NAME 139 0 0 1.0 1.0 1.0
B-EMAIL 73 0 0 1.0 1.0 1.0
I-BANK_CONTACT 52 0 0 1.0 1.0 1.0
B-BANK_CONTACT 30 0 0 1.0 1.0 1.0
B-OTHER_PERSON 116 0 0 1.0 1.0 1.0
B-ACCOUNT_NAME 97 0 0 1.0 1.0 1.0
B-DUE_DATE 127 0 0 1.0 1.0 1.0
B-OTHER_ADDRESS 11 0 0 1.0 1.0 1.0
B-ORG 147 6 0 0.9607843 1.0 0.98
B-BANK_NAME 113 0 0 1.0 1.0 1.0
B-OTHER_PERCENTAGE 74 0 0 1.0 1.0 1.0
I-OTHER_ADDRESS 38 0 0 1.0 1.0 1.0
B-ACCOUNT_NUMBER 97 0 0 1.0 1.0 1.0
I-PERSON 109 0 0 1.0 1.0 1.0
I-OTHER_ORG 283 0 18 1.0 0.9401993 0.969178
I-ACCOUNT_NUMBER 32 0 0 1.0 1.0 1.0
Macro-average 4435 24 24 0.997391 0.9971894 0.9972902
Micro-average 4435 24 24 0.99461764 0.99461764 0.99461764