SECBERT<T>

Models & Types · AiDotNet.Finance.NLP

SEC-BERT neural network model for domain-specific financial language processing.

Inheritance: NeuralNetworkBase<T> → FinancialModelBase<T> → FinancialNLPModelBase<T> → SECBERT<T>

For Beginners

SEC-BERT is a language model trained exclusively on SEC filings (10-K annual reports, 10-Q quarterly reports, 8-K current reports). It understands the unique language and structure of regulatory documents, making it ideal for extracting information from corporate disclosures, identifying risk factors, and classifying financial statements.

How It Works

SEC-BERT is a BERT-based model specifically pretrained on SEC filings (10-K, 10-Q, etc.).

Reference: Loukas et al., “SEC-BERT: A Pre-trained Financial Language Model”, 2022.

Example

// Define architecture for SEC filing classification (512 tokens, 10 filing categories)
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.OneDimensional,
    taskType: NeuralNetworkTaskType.MultiClassClassification,
    inputHeight: 512, inputWidth: 1, inputDepth: 1, outputSize: 10);
            
// Training mode: BERT pre-trained on SEC 10-K, 10-Q, and 8-K filings
var model = new SECBERT<double>(architecture);
            
// ONNX inference mode: load pre-trained SEC-BERT model
var onnxModel = new SECBERT<double>(architecture, "secbert.onnx");

Constructors

Constructor Summary
SECBERT(NeuralNetworkArchitecture<T> architecture, SECBERTOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) Creates a SEC-BERT network in native mode for training.
SECBERT(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, SECBERTOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) Creates a SEC-BERT network in native mode for training.

Properties

Property Summary
TrainingObjectiveKind TrainingObjectiveKind { get; }

Methods

Method Summary
T EvaluateTrainingObjective(Tensor<T> input, Tensor<T> target)
Tensor<T> ResolveTrainingTarget(Tensor<T> input, Tensor<T> proposedTarget)