Causal-SC (Causal Sentence Classification)

Figure 1.

Goal

The Causal-SC dataset is designed for Sentence Classification (SC) focusing on causality in the financial domain. Its primary goals are:

• To detect causal relationships in financial news and texts (e.g., “Stock rose due to strong earnings”).

• To distinguish between causal statements and non-causal (noise) text.

• To support the development of models that can understand cause-and-effect in financial markets.

Description

Causal-SC is a dataset from ChanceFocus, containing sentences annotated for the presence of causal logic.

Dataset Composition

• Source: Financial news and SEC filings.

• Classes:

  • Causal: Sentences containing a cause-effect relationship.

  • Noise: Sentences without causal content.

• Size: 1K - 10K examples.

• Modality: Text

• Language: English

Annotation Process

The dataset focuses on identifying explicit causal links, which are crucial for explaining market movements and financial events.

Example

Below is a representative example of the Causal Classification task:

Table 6 Causal-SC Examples

ID	Text	Label
0	The company’s profits surged because of higher demand in Asia.	Causal
1	The board meeting is scheduled for next Monday.	Noise

Task Description

Sentence Classification

• Input: A financial sentence.

• Output: Binary classification label (Causal or Noise).

• Challenge: Distinguishing subtle causal inferences from mere correlation or temporal sequence.

Evaluation Metrics

1. Accuracy: The percentage of correctly classified sentences.

2. F1-Score: Particularly important if the classes are imbalanced.

Why Use Causal-SC?

• Explainability: Helps in building systems that can explain why financial events happen.

• Event Logic: Crucial for event-driven trading strategies that rely on cause-effect analysis.

• Specialized Domain: Targets financial causality, which differs from general domain causality.

References

For dataset access and more details, visit the HuggingFace dataset page.