Hugging Face's logo

nlpaueb
/
sec-bert-base

Fill-Mask
Transformers
PyTorch
TensorFlow
English
bert
pretraining
finance
financial
Inference Endpoints
Model card Files Community
2
sec-bert-base / README.md
nlpaueb's picture
nlpaueb
Update README.md
cb24dd4
preview code
|
raw
history blame
8.97 kB
metadata
language: en
pipeline_tag: fill-mask
license: cc-by-sa-4.0
thumbnail: https://i.ibb.co/0yz81K9/sec-bert-logo.png
tags:
  - finance
  - financial
widget:
  - text: Total net sales [MASK] 2% or $5.4 billion during 2019 compared to 2018.
  - text: Total net sales decreased 2% or $5.4 [MASK] during 2019 compared to 2018.
  - text: >-
      During 2019, the Company [MASK] $67.1 billion of its common stock and paid
      dividend equivalents of $14.1 billion.
  - text: >-
      During 2019, the Company repurchased $67.1 billion of its common [MASK]
      and paid dividend equivalents of $14.1 billion.
  - text: >-
      During 2019, the Company repurchased $67.1 billion of its common stock and
      paid [MASK] equivalents of $14.1 billion.
  - text: >-
      During 2019, the Company repurchased $67.1 billion of its common stock and
      paid dividend [MASK] of $14.1 billion.

SEC-BERT

sec-bert-logo

SEC-BERT is a family of BERT models for the financial domain, intended to assist financial NLP research and FinTech applications.
SEC-BERT consists of the following models:

  • SEC-BERT-BASE (this model)
  • SEC-BERT-NUM: We replace every number token with a [NUM] pseudo-token handling all numeric expressions in a uniform manner, disallowing their fragmentation)
  • SEC-BERT-SHAPE: We replace numbers with pseudo-tokens that represent the number’s shape, so numeric expressions (of known shapes) are no longer fragmented.
    (e.g. , '53.2' becomes '[XX.X]' and '40,200.5' becomes '[XX,XXX.X]').

Pre-training corpus

The model was pre-trained on 260,773 10-K filings from 1993-2019, publicly available at U.S. Securities and Exchange Commission (SEC)

Pre-training details

  • We created a new vocabulary of 30k subwords by training a BertWordPieceTokenizer from scratch on the pre-training corpus.
  • We trained BERT using the official code provided in Google BERT's GitHub repository.
  • We then used Hugging Face's Transformers conversion script to convert the TF checkpoint in the desired format in order to be able to load the model in two lines of code for both PyTorch and TF2 users.
  • We release a model similar to the English BERT-BASE model (12-layer, 768-hidden, 12-heads, 110M parameters).
  • We chose to follow the same training set-up: 1 million training steps with batches of 256 sequences of length 512 with an initial learning rate 1e-4.
  • We were able to use a single Google Cloud TPU v3-8 provided for free from TensorFlow Research Cloud (TFRC), while also utilizing GCP research credits. Huge thanks to both Google programs for supporting us!

Load Pretrained Model 

from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("nlpaueb/sec-bert-base")
model = AutoModel.from_pretrained("nlpaueb/sec-bert-base")

Use LEBAL-BERT variants as Language Models

Sample Masked Token
Total net sales [MASK] 2% or $5.4 billion during 2019 compared to 2018. decreased
Model Predictions (Probability)
BERT-BASE-UNCASED increased (0.221), were (0.131), are (0.103), rose (0.075), of (0.058)
SEC-BERT-BASE increased (0.678), decreased (0.282), declined (0.017), grew (0.016), rose (0.004)
SEC-BERT-NUM increased (0.665), decreased (0.281), grew (0.028), declined (0.015), rose (0.008)
SEC-BERT-SHAPE increased (0.793), decreased (0.145), grew (0.042), declined (0.011), rose (0.003)
Model Masked Token Predictions (Probability)
Total net sales [MASK] 2% or $5.4 billion during 2019 compared to 2018.
BERT-BASE-UNCASED decreased increased (0.221), were (0.131), are (0.103), rose (0.075), of (0.058)
SEC-BERT-BASE decreased increased (0.678), decreased (0.282), declined (0.017), grew (0.016), rose (0.004)
SEC-BERT-NUM decreased increased (0.665), decreased (0.281), grew (0.028), declined (0.015), rose (0.008)
SEC-BERT-SHAPE decreased increased (0.793), decreased (0.145), grew (0.042), declined (0.011), rose (0.003)
Total net sales decreased 2% or $5.4 [MASK] during 2019 compared to 2018.
BERT-BASE-UNCASED billion
SEC-BERT-BASE billion
SEC-BERT-NUM billion
SEC-BERT-SHAPE billion
During 2019, the Company [MASK] $67.1 billion of its common stock and paid dividend equivalents of $14.1 billion.
BERT-BASE-UNCASED repurchased ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-BASE repurchased ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-NUM repurchased ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-SHAPE repurchased ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
During 2019, the Company repurchased $67.1 billion of its common [MASK] and paid dividend equivalents of $14.1 billion.
BERT-BASE-UNCASED stock ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-BASE stock ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-NUM stock ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-SHAPE stock ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
During 2019, the Company repurchased $67.1 billion of its common stock and paid [MASK] equivalents of $14.1 billion.
BERT-BASE-UNCASED dividend ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-BASE dividend ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-NUM dividend ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-SHAPE dividend ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
During 2019, the Company repurchased $67.1 billion of its common stock and paid dividend [MASK] of $14.1 billion.
BERT-BASE-UNCASED equivalents ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-BASE equivalents ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-NUM equivalents ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')
SEC-BERT-SHAPE equivalents ('letter', '0.38'), ('dealer', '0.04'), ('employment', '0.03'), ('award', '0.03'), ('contribution', '0.02')

About Us

AUEB's Natural Language Processing Group develops algorithms, models, and systems that allow computers to process and generate natural language texts.

The group's current research interests include:

  • question answering systems for databases, ontologies, document collections, and the Web, especially biomedical question answering,
  • natural language generation from databases and ontologies, especially Semantic Web ontologies, text classification, including filtering spam and abusive content,
  • information extraction and opinion mining, including legal text analytics and sentiment analysis,
  • natural language processing tools for Greek, for example parsers and named-entity recognizers, machine learning in natural language processing, especially deep learning.

The group is part of the Information Processing Laboratory of the Department of Informatics of the Athens University of Economics and Business.

Manos Fergadiotis on behalf of AUEB's Natural Language Processing Group