Movie Scene Event Extraction with Graph Attention Network Based on Argument Correlation Information

Area of Science:

• Natural Language Processing
• Media Analysis
• Artificial Intelligence

Background:

• Movie scene event extraction is crucial for media analysis but underexplored.
• Existing methods struggle with argument roles sharing similar characteristics in movie scripts.

Purpose of the Study:

• To propose a new model for movie scene event extraction that leverages correlations between argument roles.
• To improve trigger and argument identification and classification in movie scripts.

Main Methods:

• Introduced the superior role concept (SRC) to model correlations between argument roles.
• Developed an attentive high-level argument role module to capture SRC information.
• Utilized an SRC-based graph attention network (GAT) to integrate role correlation into semantic embeddings.

Main Results:

• The proposed model achieved superior performance compared to existing methods on a new dataset (MovieSceneEvent) and a benchmark dataset.
• Demonstrated that incorporating argument role correlation information significantly enhances movie scene event extraction.

Conclusions:

• The novel approach effectively models argument role correlations, leading to improved movie scene event extraction.
• The SRC concept and GAT integration offer a promising direction for advancing media analysis tasks.