Towards automated check-worthy sentence detection using Gated Recurrent Unit

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• The rapid spread of information necessitates automated fact-checking systems.
• Identifying factual claims is the critical first step in automated fact-checking.
• Existing methods require improvement in accurately detecting check-worthy sentences.

Purpose of the Study:

• To propose and evaluate a novel framework for detecting check-worthy sentences.
• To assess the framework's ability to identify both the presence and importance of factual content.
• To compare the framework's performance against existing methods on standard and custom datasets.

Main Methods:

• Developed a gated recurrent unit (GRU) pipeline utilizing GloVe embeddings, termed the G2CW framework.
• The G2CW framework is designed for check-worthy sentence detection, distinguishing between factual and non-factual content.
• Evaluated the framework on the ClaimBuster dataset and a self-curated IndianClaim dataset.

Main Results:

• The G2CW framework achieved a high F1-score of 0.92, outperforming previous approaches.
• The framework demonstrated strong performance on the IndianClaim dataset, even when trained on the ClaimBuster dataset.
• This indicates the robustness and generalizability of the proposed G2CW method.

Conclusions:

• The G2CW framework represents a significant advancement in automated check-worthy sentence detection.
• The model's effectiveness on diverse datasets highlights its potential for real-world fact-checking applications.
• Further research can build upon this framework to enhance the accuracy and efficiency of automated fact-checking systems.